Research ArticleInformation Filtering via Biased Random Walk onCoupled Social Network

Da-Cheng Nie1 Zi-Ke Zhang23 Qiang Dong1 Chongjing Sun1 and Yan Fu1

1 Web Sciences Center School of Computer Science amp Engineering University of Electronic Science and Technology of ChinaChengdu 610054 China

2 Institute of Information Economy Hangzhou Normal University Hangzhou 311121 China3 Alibaba Research Center for Complexity Sciences Hangzhou Normal University Hangzhou 311121 China

Correspondence should be addressed to Da-Cheng Nie niedachenggmailcom

Received 27 March 2014 Revised 30 June 2014 Accepted 8 July 2014 Published 22 July 2014

Academic Editor Yolanda Blanco Fernandez

Copyright copy 2014 Da-Cheng Nie et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The recommender systems have advanced a great deal in the past two decades However most researchers focus their attentionson mining the similarities among users or objects in recommender systems and overlook the social influence which plays animportant role in usersrsquo purchase process In this paper we design a biased random walk algorithm on coupled social networkswhich gives recommendation results based on both social interests and usersrsquo preference Numerical analyses on two real data setsEpinions and Friendfeed demonstrate the improvement of recommendation performance by taking social interests into accountand experimental results show that our algorithm can alleviate the user cold-start problemmore effectively compared with themassdiffusion and user-based collaborative filtering methods

1 Introduction

In the past two decades the Web 20 and its applicationshave greatly accelerated the development of the InternetThey bring our lives much convenience as well as over-whelm us with too many resources in the informationocean One typical scenario is online shopping in our dailylife When we are confronted with millions of books onhttpwwwAmazoncom or billions of different kinds ofcommodities on httpwwwTaobaocom indeed it is verydifficult to choose the relevant ones from countless candi-dates This is the so-called Information Overload problem [1]Therefore an automatic way that can help us make the rightdecision under the Information Overload is a significant issuefor both academic and industrial communities

Search engines provide a way to help users find the usefulinformation which alleviates this dilemma partially a userinputs the keywords and then the search engine returns theresults accordingly However if different users input the samekeywords the search engine will return the same resultsBesides when users resort to a search engine theymust knowhow to clearly describe what they want by the keywords But

inmost situations users do not knowwhat they really want orit is hard for them to find appropriate keywords In this casethe recommender systems [2] have been designed to solvethis problem

Recently social networks (SN) [3 4] have become apowerful tool to characterize social relationship in onlinesocial services emerging with various Web 20 applications[5] in evolutionary games [6 7] community detection [8]medical science [9] and so forth By taking advantage ofsocial relationship in recommender systems many tradi-tional challenges can be partially solved such as the cold-startproblem [10] and data sparsity problem [11] However mostresearches are focused onmining the similarities amongusersor objects in recommender systems and the social influenceis seldom taken into account

Coupled networks (CN) also known as interdependentnetworks [12] are usually composed of two layers of net-works [12 13] such as electricityinternet networks [14] andairportrailway networks [15] Being similar with interdepen-dent networks a coupled social network (CSN) also containsthe coupling nodes (users) which form a leader-followerrelationship in the layer of social network and collecting

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 829137 10 pageshttpdxdoiorg1011552014829137

2 The Scientific World Journal

relationship in the layer of information network Figure 1gives an illustration of a simple CSN with five users andfive objects where circles denote users and squares representobjects the social network (upper layer) consists of five usersand the information network (lower layer) consists of fiveobjects and five users where the users are the same as thosein the social network It can be seen that 119874

5will not be

recommended to1198804in the user-object network because only

1198805collects object 119874

5and the value of similarity between 119880

4

and1198805is zero However 119880

4follows119880

5in the social network

which indicates that 1198804may have similar interests with 119880

5

to some extent thus we can accordingly recommend 1198745to

1198804via social network Therefore by making use of the social

relationship between users the user cold-start problem canbe partially solved When a new user comes to the systemwe can recommend himher some objects through the socialnetwork

Moreover most researchers focus their attention onmin-ing the similarities among users or objects in recommendersystems and many researchers use the social interest tofilter the recommendations but we use the social interestto supplement the recommendations instead of filteringthem To our knowledge the random walk algorithm oncoupled social network remains yet to be investigated inrecommender systems

The contributions of this paper can be summarized asfollows (1) We use the social interest to supplement therecommendations instead of filtering them and we obtainmore accurate recommendations (2) We first propose abiased random walk recommendation algorithm on coupledsocial network which considers the social interests as well asusersrsquo preference in the recommender systems This methodcan improve the performance of recommendations (3)Com-pared with the mass diffusion (MD) [16 17] and user-basedCF (UCF) [18 19] methods the proposed algorithm canalleviate the user cold-start problem more effectively

This paper is organized as follows We introduce therelated works in Section 2 In Section 3 we propose a biasedrandom walk algorithm on coupled social network InSection 4 we describe the data sets and metrics used in thispaper We evaluate the performance of the proposed methodin Section 5 Finally we summarize this paper in Section 6

2 Related Works

Collaborative filtering (CF) [18ndash25] is the most frequentlyused technology in recommender systems which uses thecollection history of users for mining the potential objects ofinterest to the target user However the CF algorithm onlytakes the similar users or objects into account and will lead tothe same recommendation results to diverse users namelyit is not conducive to the personalized recommendationMeanwhile the CF algorithm cannot deal with the cold-startproblem [10] that is when a new user or object is addedto the system it is difficult to obtain recommendation or tobe recommended because of lack of enough information Toalleviate this problem many methods have been proposedsuch as content-based [26] trust-aware [27 28] social-impact [29] and tag-aware [30] methods

U1

U2

U3

U4

U5

O1

O2

O3

O4

O5

U1U2

U3U4

U5

Social network

Information network

Figure 1 Illustration of a coupled social network with five users andfive objects where circles denoteusers and squares represent objects(color online)The social network (upper layer) consists of five usersthe information network (lower layer) consists of five objects andfive users while user nodes are the same in the social network

Random walk [31] is a mathematical formalization of apath that consists of a succession of random edges which issuccessfully used in recommender systems based on bipartitenetwork [16 32] namely mass diffusion (MD for short)method [16] Accordingly many methods based on massdiffusion were proposed [17 33] Furthermore random walkwas successfully used in many fields such as social network[34] and Top-k search [35] However there is a lack of studyof random walk on coupled social network in recommendersystems

Massa and Avesani [36] proposed a social propagationmethod that is based on usersrsquo distance from a fixed prop-agation horizon which increased the coverage of recom-mender systems Esslimani et al [37] proposed a feedbackeffect between similarity and social influence in onlinecommunities By utilizing the social relations we can obtainthe strength of social relationship between users and wecan use this social relationship to generate more accuraterecommendation results Meanwhile the literature [36 38]demonstrated that recommendation performance can beimproved by taking into consideration the effect of socialnetwork and the methods are both filtering the uselessinformation by social relationship

Lai et al [39] proposed a hybrid personal trust modelwhich adaptively combines the rating-based trust model andexplicit trust metric to resolve the drawback caused by insuf-ficient past rating records Community-based recommendersystems have attracted much research attention the authors[40] proposed a novel community-based framework thatemploys PLSA-based model incorporating social activenessand dynamic interest to discover communities Wei et al [41]proposed a multicollaborative filtering trust network algo-rithm an improved version of CF algorithmdesigned towork

The Scientific World Journal 3

onWeb 20 platform which can improve the prediction accu-racy compared with the original CF algorithm We believethat if the social relationship can be used to supplementthe user-object network like the aforementioned example ofFigure 1 we will get more accurate recommendations andalleviate the user cold-start problem Motivated by this weproposed a biased random walk (diffusion-based) methodon coupled social network to generate recommendationsTherefore new users can obtain recommendations as long asthey are connected to others in social networks

3 Method

In this section we introduce the approach of diffusion oncoupled social networks Generally a recommender systemconsists of two sets 119880 = 119880

1 1198802 119880

119898 and 119874 =

1198741 1198742 119874

119899 representing the 119898 users and 119899 objects

respectively Denote119860119898times119899

by the adjacent matrix of the user-object bipartite network of which each element 119886

119894120572= 1

if user 119880119894has collected object 119874

120572 and 119886

119894120572= 0 otherwise

Analogously denote 119861119898times119898

by the nonsymmetric adjacentmatrix of user-user directed social network of which eachelement 119887

119894119895= 1 if the user119880

119894has linked to user119880

119895 and 119887

119894119895= 0

otherwise

Random Walk on Social Network Let 1198751015840 be the 119898 times 119898transition probabilitymatrix of a directed social networkTheprobability that a randomwalker at user119880

119894goes to user119880

119895on

social network can be described as

1199011015840

119894119895=

119887119894119895

119896out119894

if 119896out119894= 0

0 otherwise(1)

where 119896out119894

is the out-degree in social network that is thenumber of leaders of user 119880

119894 Denote 1199041015840

119894(119905) by the probability

from other users to user 119880119894at time 119905 Therefore we have

1199041015840

119894(119905 + 1) =

119898

sum

119895=1

119887119894119895

119896out119894

1199041015840

119895(119905) if 119896out

119894= 0

0 otherwise(2)

The initial probability for target user 119880119894is given by 1199041015840

119894(0) = 1

and 1199041015840119895(0) = 0 for all of the other user119880

119895 Thus we can obtain

the probability that a randomwalker goes from the target userto all other users at time 119905

Random Walk on Bipartite Network Let 11987510158401015840 be the 119898 times119899 transition probability matrix of a bipartite network Theprobability that a random walker at user119880

119894goes to object119874

120572

on bipartite network can be described as

11990110158401015840

119894120572=

119886119894120572

119896119894

if 119896119894= 0

0 otherwise(3)

where 119896119894denotes the number of collected objects of user 119880

119894

and the probability that a randomwalker at object119874120572goes to

user 119880119895on bipartite network can be described as

11990110158401015840

120572119895=

119886119895120572

119896120572

if 119896120572= 0

0 otherwise(4)

where 119896120572denotes the number of users who have collected

object119874120572on bipartite network Denote 11990410158401015840

119894(119905) and 11990410158401015840

120572(119905) by the

probability of user 119880119894and object 119874

120572on bipartite network at

time 119905 respectively Therefore we have

11990410158401015840

119894(119905 + 1) =

119899

sum

120572=1

119886119894120572

119896119894

11990410158401015840

120572(119905) if 119896

119894= 0

0 otherwise

11990410158401015840

120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

11990410158401015840

119895(119905) if 119896

120572= 0

0 otherwise

(5)

Similar to random walk on social network the initial proba-bility for target user119880

119894is given by 11990410158401015840

119894(0) = 1 But the difference

is the fact that there are two different nodes on bipartitenetwork and the initial probability 11990410158401015840

119895(0) = 0 and 11990410158401015840

120572(0) = 0

for all the other user119880119895and object 120572 In the odd time step and

119905 ge 3 the probability of 11990410158401015840120572(119905)means the probability of target

user 119880119894selecting uncollected object 119874

120572 Therefore we can

obtain the recommendation list according to this probabilityfor target user

Biased Random Walk on Coupled Social Network Let 119875 bethe119872 times119872 transition probability matrix of a coupled socialnetwork where119872 = 119898 + 119899 In order to solve the user cold-start problem suppose that a random walker at user 119880

119894goes

to their neighbors (leaders) on directed social network withprobability 120582 isin (0 1) and to their neighbors on bipartitenetworkwith probability 1minus120582Whatrsquosmore a randomwalkerat object119874

120572goes to all users who collect object119874

120572with equal

probability Thus the target user finds the potential objectsnot only through other users with similar collecting intereston bipartite network but also through their friends (leaders)on directed social network Denote 119904

119894(119905) and 119904

120572(119905) by the

probability of walker user119880119894and object119874

120572on coupled social

network at time 119905 respectively Therefore we have

119904119894(119905 + 1) =

120582 sdot

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) + (1 minus 120582)

sdot

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) if 119896

119894= 0 119896

out119894= 0

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

0 otherwise

4 The Scientific World Journal

119904120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

119904119895(119905) if 119896

120572= 0

0 otherwise(6)

That is to say initially we assign the target user one unitof resource Then 120582 (0 le 120582 le 1) proportion of the resourceis evenly distributed to the userrsquos social neighbors throughthe directed links (social network) and 1 minus 120582 proportion isdistributed to collected objects through the undirected links(bipartite network) In (6) when 119896out

119894= 0 then 120582 = 0

it means that user 119880119894has no outlinks in social network

therefore heshe will distribute all of hisher resources tobipartite network Similarly when 119896

119894= 0 then 120582 = 1 user

119880119894will distribute all of hisher resources to social network

The initial score for target user 119880119894is given by 119904

119894(0) = 1

119904119895(0) = 0 and 119904

120572(0) = 0 for all the other user 119880

119895and object

120572 Thus we can obtain the recommendations by ranking thescore 119904

120572of all objects at time 119905 for target user At time 119905 = 2

the recommendations are obtained only from social networkthat is hisher social leaders At time 119905 = 3 and 120582 = 0 therecommendations are obtained only from bipartite networkand it is the pure MD algorithm

Thus the probability that a random walker arrives atthe object at time 119905 is recognized as the possibility thatthe target user purchases this object We call this algorithmbiased random walk (BRW) For the example in Figure 1 thetransition probability matrix 119875 for coupled social network isgiven in the following equation

119875 =

1198801

1198802

1198803

1198804

1198805

1198741

1198742

1198743

1198744

1198745

1198801119880211988031198804119880511987411198742119874311987441198745

((((((((

(

0 12 0 0 0 14 14 0 0 0

14 0 14 0 0 14 14 0 0 0

16 0 0 16 16 0 14 14 0 0

0 0 0 0 12 0 0 14 14 0

0 0 12 0 0 0 0 0 0 12

12 12 0 0 0 0 0 0 0 0

13 13 13 0 0 0 0 0 0 0

0 0 12 12 0 0 0 0 0 0

0 0 0 1 0 0 0 0 0 0

0 0 0 0 1 0 0 0 0 0

))))))))

)

(7)

Consider 120582 = 05 and 119905 = 2 then 119875(1198803 1198745) = 00833

which means users 1198803and 119874

5are reachable within 2 steps

with 00833 probability through the coupled social networkOn the other hand without social network the random walkdistance on the original bipartite network 11987510158401015840(119880

3 1198745) = 0 for

an arbitrary time 119905 because1198803and119874

5are not reachable from

each other in bipartite network

4 Data and Metrics

41 Data Sets To evaluate our algorithmrsquos performance tworeal data sets are analyzed in the experiments The datasets are from httpwwwepinionscom and httpwwwfriendfeedcom both of which provided user-objectscollecting information and user-user social relationshipThe Epinions data set was collected by Paolo Massa ina 5-week crawl (NovemberDecember 2003) from thehttpwwwepinionscom website [36] and the Friendfeeddata set was collected by Fabio Celli et al from httpwwwfriendfeedcom (September 6 2009 to September19 2009) [42] We extract a smaller data set by randomlysampling the whole records of user activities in both Epinionsand Friendfeed data sets 4066 users 7649 objects 154122collected links and 217071 social links in total were foundin the Epinions data set Friendfeed contains 4148 userswho collected 5700 objects 96942 collected links and386804 social links Table 1 shows the basic statistics for

two representative data sets Denote |119880| |119874| and 119873119877by

the number of users objects and ratings respectivelySparsity = 119873

119877(|119880| times |119874|) denotes the data sparsity of

user-objects network

42 Metrics To test our algorithmrsquos performance each infor-mation network is randomly divided into two parts thetraining set consists of 90 entries and the remaining entriesconstitute the testing set The training set is treated asknown information used for generating recommendationswhile the training set is regarded as unknown informationused for testing the performance of the recommendationresults To evaluate the proposed algorithm we employedfive different metrics that characterize not only the accuracyof recommendations but also the diversification which aredefined as follows

(1) Precision [43] Precision represents the probability that theselected objects appeared in the recommendation list whichis shown as

Precision119894=119873119894

119903119904

119871 (8)

where Precision119894represents user 119906

119894rsquos precision 119873119894

119903119904denotes

the number of recommended objects that appeared in the119880119894rsquos

testing set and 119871 represents the length of recommendation

The Scientific World Journal 5

Table 1 Properties of the tested data sets

Data sets Users Objects Collecting links Social links SparsityEpinions 4066 7649 154122 217017 5 times 10

minus3

Friendfeed 4148 5700 96942 386804 41 times 10minus3

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0

120582

08

07

06

05

04

03

02

t

(a) Epinions

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0120582

t

07

06

05

04

03

02

01

(b) Friendfeed

Figure 2 Ranking score values on Epinions and Friendfeed data sets (color online)

list By averaging over all usersrsquo precisions we can obtain thewhole recommender systemsrsquo precision as

Precision = 1119898

119898

sum

119894=1

Precision119894 (9)

where119898 represents the number of users Obviously a higherprecision means a higher recommendation accuracy

(2) Recall [43] Recall represents the probability that therecommended objects appeared in userrsquos collected list shownas

Recall119894=119873119894

119903119904

119873119894119901

(10)

where Recall119894represents user 119906

119894rsquos recall and119873119894

119901is the number

of objects collected by user 119906119894in the testing set Averaging

over all individualsrsquo recall we can obtain the recall of thewhole recommender system

(3) F-Measure [43] Generally speaking for each user recall issensitive to 119871 and a larger 119871 generally gives a higher recall buta lower precision The F-measure that assigns equal weightfor precision and recall is defined as

119865-measure119894=2 sdot precision

119894sdot recall

119894

precision119894+ recall

119894

(11)

By averaging over all usersrsquo119865-measure we can also obtainthe whole systemrsquos 119865-measure

(4) HD [17] HD is a metric to measure the diversity ofusersrsquo recommendation lists It uses the Hamming distance

to measure the difference of recommendation lists betweenusers 119906

119894and 119906

119895 which is defined as

HD119894119895(119871) = 1 minus

119876119894119895(119871)

119871 (12)

where 119876119894119895(119871) is the number of commonly recommended

objects shown in top-119871 locations of users 119906119894and 119906

119895rsquos recom-

mendation list Averaging over all pairs of usersrsquo HD119894119895(119871) we

can obtain theHDof the recommender algorithmObviouslyhigher HD means higher diversity of users

(5) Ranking Score (119903) [44] Generally the recommendersystem aims to generate a ranking list for the target userrsquosuncollected objects through the prediction score In therecommender systems one of the most used metrics toevaluate the algorithmrsquos performance is ranking score whichmeasures the usersrsquo satisfaction of the ranking list and isdefined as follows

119903119894120572=119871119894120572

119873119894

(13)

where 119871119894120572is the position of uncollected object 120572 in user 119880

119894rsquos

ranking list and 119873119894is the length of the user 119880

119894rsquos ranking list

By averaging all linksrsquo ranking score value we can obtain thewhole systemrsquos ranking score value 119903 A small 119903 means therecommender system puts the userrsquos favorite objects in a topplace in the recommender list hence the smaller 119903 is thebetter an algorithmrsquos performance will be

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

2 The Scientific World Journal

relationship in the layer of information network Figure 1gives an illustration of a simple CSN with five users andfive objects where circles denote users and squares representobjects the social network (upper layer) consists of five usersand the information network (lower layer) consists of fiveobjects and five users where the users are the same as thosein the social network It can be seen that 119874

5will not be

recommended to1198804in the user-object network because only

1198805collects object 119874

5and the value of similarity between 119880

4

and1198805is zero However 119880

4follows119880

5in the social network

which indicates that 1198804may have similar interests with 119880

5

to some extent thus we can accordingly recommend 1198745to

1198804via social network Therefore by making use of the social

relationship between users the user cold-start problem canbe partially solved When a new user comes to the systemwe can recommend himher some objects through the socialnetwork

Moreover most researchers focus their attention onmin-ing the similarities among users or objects in recommendersystems and many researchers use the social interest tofilter the recommendations but we use the social interestto supplement the recommendations instead of filteringthem To our knowledge the random walk algorithm oncoupled social network remains yet to be investigated inrecommender systems

The contributions of this paper can be summarized asfollows (1) We use the social interest to supplement therecommendations instead of filtering them and we obtainmore accurate recommendations (2) We first propose abiased random walk recommendation algorithm on coupledsocial network which considers the social interests as well asusersrsquo preference in the recommender systems This methodcan improve the performance of recommendations (3)Com-pared with the mass diffusion (MD) [16 17] and user-basedCF (UCF) [18 19] methods the proposed algorithm canalleviate the user cold-start problem more effectively

This paper is organized as follows We introduce therelated works in Section 2 In Section 3 we propose a biasedrandom walk algorithm on coupled social network InSection 4 we describe the data sets and metrics used in thispaper We evaluate the performance of the proposed methodin Section 5 Finally we summarize this paper in Section 6

2 Related Works

Collaborative filtering (CF) [18ndash25] is the most frequentlyused technology in recommender systems which uses thecollection history of users for mining the potential objects ofinterest to the target user However the CF algorithm onlytakes the similar users or objects into account and will lead tothe same recommendation results to diverse users namelyit is not conducive to the personalized recommendationMeanwhile the CF algorithm cannot deal with the cold-startproblem [10] that is when a new user or object is addedto the system it is difficult to obtain recommendation or tobe recommended because of lack of enough information Toalleviate this problem many methods have been proposedsuch as content-based [26] trust-aware [27 28] social-impact [29] and tag-aware [30] methods

U1

U2

U3

U4

U5

O1

O2

O3

O4

O5

U1U2

U3U4

U5

Social network

Information network

Figure 1 Illustration of a coupled social network with five users andfive objects where circles denoteusers and squares represent objects(color online)The social network (upper layer) consists of five usersthe information network (lower layer) consists of five objects andfive users while user nodes are the same in the social network

Random walk [31] is a mathematical formalization of apath that consists of a succession of random edges which issuccessfully used in recommender systems based on bipartitenetwork [16 32] namely mass diffusion (MD for short)method [16] Accordingly many methods based on massdiffusion were proposed [17 33] Furthermore random walkwas successfully used in many fields such as social network[34] and Top-k search [35] However there is a lack of studyof random walk on coupled social network in recommendersystems

Massa and Avesani [36] proposed a social propagationmethod that is based on usersrsquo distance from a fixed prop-agation horizon which increased the coverage of recom-mender systems Esslimani et al [37] proposed a feedbackeffect between similarity and social influence in onlinecommunities By utilizing the social relations we can obtainthe strength of social relationship between users and wecan use this social relationship to generate more accuraterecommendation results Meanwhile the literature [36 38]demonstrated that recommendation performance can beimproved by taking into consideration the effect of socialnetwork and the methods are both filtering the uselessinformation by social relationship

Lai et al [39] proposed a hybrid personal trust modelwhich adaptively combines the rating-based trust model andexplicit trust metric to resolve the drawback caused by insuf-ficient past rating records Community-based recommendersystems have attracted much research attention the authors[40] proposed a novel community-based framework thatemploys PLSA-based model incorporating social activenessand dynamic interest to discover communities Wei et al [41]proposed a multicollaborative filtering trust network algo-rithm an improved version of CF algorithmdesigned towork

The Scientific World Journal 3

onWeb 20 platform which can improve the prediction accu-racy compared with the original CF algorithm We believethat if the social relationship can be used to supplementthe user-object network like the aforementioned example ofFigure 1 we will get more accurate recommendations andalleviate the user cold-start problem Motivated by this weproposed a biased random walk (diffusion-based) methodon coupled social network to generate recommendationsTherefore new users can obtain recommendations as long asthey are connected to others in social networks

3 Method

In this section we introduce the approach of diffusion oncoupled social networks Generally a recommender systemconsists of two sets 119880 = 119880

1 1198802 119880

119898 and 119874 =

1198741 1198742 119874

119899 representing the 119898 users and 119899 objects

respectively Denote119860119898times119899

by the adjacent matrix of the user-object bipartite network of which each element 119886

119894120572= 1

if user 119880119894has collected object 119874

120572 and 119886

119894120572= 0 otherwise

Analogously denote 119861119898times119898

by the nonsymmetric adjacentmatrix of user-user directed social network of which eachelement 119887

119894119895= 1 if the user119880

119894has linked to user119880

119895 and 119887

119894119895= 0

otherwise

Random Walk on Social Network Let 1198751015840 be the 119898 times 119898transition probabilitymatrix of a directed social networkTheprobability that a randomwalker at user119880

119894goes to user119880

119895on

social network can be described as

1199011015840

119894119895=

119887119894119895

119896out119894

if 119896out119894= 0

0 otherwise(1)

where 119896out119894

is the out-degree in social network that is thenumber of leaders of user 119880

119894 Denote 1199041015840

119894(119905) by the probability

from other users to user 119880119894at time 119905 Therefore we have

1199041015840

119894(119905 + 1) =

119898

sum

119895=1

119887119894119895

119896out119894

1199041015840

119895(119905) if 119896out

119894= 0

0 otherwise(2)

The initial probability for target user 119880119894is given by 1199041015840

119894(0) = 1

and 1199041015840119895(0) = 0 for all of the other user119880

119895 Thus we can obtain

the probability that a randomwalker goes from the target userto all other users at time 119905

Random Walk on Bipartite Network Let 11987510158401015840 be the 119898 times119899 transition probability matrix of a bipartite network Theprobability that a random walker at user119880

119894goes to object119874

120572

on bipartite network can be described as

11990110158401015840

119894120572=

119886119894120572

119896119894

if 119896119894= 0

0 otherwise(3)

where 119896119894denotes the number of collected objects of user 119880

119894

and the probability that a randomwalker at object119874120572goes to

user 119880119895on bipartite network can be described as

11990110158401015840

120572119895=

119886119895120572

119896120572

if 119896120572= 0

0 otherwise(4)

where 119896120572denotes the number of users who have collected

object119874120572on bipartite network Denote 11990410158401015840

119894(119905) and 11990410158401015840

120572(119905) by the

probability of user 119880119894and object 119874

120572on bipartite network at

time 119905 respectively Therefore we have

11990410158401015840

119894(119905 + 1) =

119899

sum

120572=1

119886119894120572

119896119894

11990410158401015840

120572(119905) if 119896

119894= 0

0 otherwise

11990410158401015840

120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

11990410158401015840

119895(119905) if 119896

120572= 0

0 otherwise

(5)

Similar to random walk on social network the initial proba-bility for target user119880

119894is given by 11990410158401015840

119894(0) = 1 But the difference

is the fact that there are two different nodes on bipartitenetwork and the initial probability 11990410158401015840

119895(0) = 0 and 11990410158401015840

120572(0) = 0

for all the other user119880119895and object 120572 In the odd time step and

119905 ge 3 the probability of 11990410158401015840120572(119905)means the probability of target

user 119880119894selecting uncollected object 119874

120572 Therefore we can

obtain the recommendation list according to this probabilityfor target user

Biased Random Walk on Coupled Social Network Let 119875 bethe119872 times119872 transition probability matrix of a coupled socialnetwork where119872 = 119898 + 119899 In order to solve the user cold-start problem suppose that a random walker at user 119880

119894goes

to their neighbors (leaders) on directed social network withprobability 120582 isin (0 1) and to their neighbors on bipartitenetworkwith probability 1minus120582Whatrsquosmore a randomwalkerat object119874

120572goes to all users who collect object119874

120572with equal

probability Thus the target user finds the potential objectsnot only through other users with similar collecting intereston bipartite network but also through their friends (leaders)on directed social network Denote 119904

119894(119905) and 119904

120572(119905) by the

probability of walker user119880119894and object119874

120572on coupled social

network at time 119905 respectively Therefore we have

119904119894(119905 + 1) =

120582 sdot

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) + (1 minus 120582)

sdot

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) if 119896

119894= 0 119896

out119894= 0

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

0 otherwise

4 The Scientific World Journal

119904120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

119904119895(119905) if 119896

120572= 0

0 otherwise(6)

That is to say initially we assign the target user one unitof resource Then 120582 (0 le 120582 le 1) proportion of the resourceis evenly distributed to the userrsquos social neighbors throughthe directed links (social network) and 1 minus 120582 proportion isdistributed to collected objects through the undirected links(bipartite network) In (6) when 119896out

119894= 0 then 120582 = 0

it means that user 119880119894has no outlinks in social network

therefore heshe will distribute all of hisher resources tobipartite network Similarly when 119896

119894= 0 then 120582 = 1 user

119880119894will distribute all of hisher resources to social network

The initial score for target user 119880119894is given by 119904

119894(0) = 1

119904119895(0) = 0 and 119904

120572(0) = 0 for all the other user 119880

119895and object

120572 Thus we can obtain the recommendations by ranking thescore 119904

120572of all objects at time 119905 for target user At time 119905 = 2

the recommendations are obtained only from social networkthat is hisher social leaders At time 119905 = 3 and 120582 = 0 therecommendations are obtained only from bipartite networkand it is the pure MD algorithm

Thus the probability that a random walker arrives atthe object at time 119905 is recognized as the possibility thatthe target user purchases this object We call this algorithmbiased random walk (BRW) For the example in Figure 1 thetransition probability matrix 119875 for coupled social network isgiven in the following equation

119875 =

1198801

1198802

1198803

1198804

1198805

1198741

1198742

1198743

1198744

1198745

1198801119880211988031198804119880511987411198742119874311987441198745

((((((((

(

0 12 0 0 0 14 14 0 0 0

14 0 14 0 0 14 14 0 0 0

16 0 0 16 16 0 14 14 0 0

0 0 0 0 12 0 0 14 14 0

0 0 12 0 0 0 0 0 0 12

12 12 0 0 0 0 0 0 0 0

13 13 13 0 0 0 0 0 0 0

0 0 12 12 0 0 0 0 0 0

0 0 0 1 0 0 0 0 0 0

0 0 0 0 1 0 0 0 0 0

))))))))

)

(7)

Consider 120582 = 05 and 119905 = 2 then 119875(1198803 1198745) = 00833

which means users 1198803and 119874

5are reachable within 2 steps

with 00833 probability through the coupled social networkOn the other hand without social network the random walkdistance on the original bipartite network 11987510158401015840(119880

3 1198745) = 0 for

an arbitrary time 119905 because1198803and119874

5are not reachable from

each other in bipartite network

4 Data and Metrics

41 Data Sets To evaluate our algorithmrsquos performance tworeal data sets are analyzed in the experiments The datasets are from httpwwwepinionscom and httpwwwfriendfeedcom both of which provided user-objectscollecting information and user-user social relationshipThe Epinions data set was collected by Paolo Massa ina 5-week crawl (NovemberDecember 2003) from thehttpwwwepinionscom website [36] and the Friendfeeddata set was collected by Fabio Celli et al from httpwwwfriendfeedcom (September 6 2009 to September19 2009) [42] We extract a smaller data set by randomlysampling the whole records of user activities in both Epinionsand Friendfeed data sets 4066 users 7649 objects 154122collected links and 217071 social links in total were foundin the Epinions data set Friendfeed contains 4148 userswho collected 5700 objects 96942 collected links and386804 social links Table 1 shows the basic statistics for

two representative data sets Denote |119880| |119874| and 119873119877by

the number of users objects and ratings respectivelySparsity = 119873

119877(|119880| times |119874|) denotes the data sparsity of

user-objects network

42 Metrics To test our algorithmrsquos performance each infor-mation network is randomly divided into two parts thetraining set consists of 90 entries and the remaining entriesconstitute the testing set The training set is treated asknown information used for generating recommendationswhile the training set is regarded as unknown informationused for testing the performance of the recommendationresults To evaluate the proposed algorithm we employedfive different metrics that characterize not only the accuracyof recommendations but also the diversification which aredefined as follows

(1) Precision [43] Precision represents the probability that theselected objects appeared in the recommendation list whichis shown as

Precision119894=119873119894

119903119904

119871 (8)

where Precision119894represents user 119906

119894rsquos precision 119873119894

119903119904denotes

the number of recommended objects that appeared in the119880119894rsquos

testing set and 119871 represents the length of recommendation

The Scientific World Journal 5

Table 1 Properties of the tested data sets

Data sets Users Objects Collecting links Social links SparsityEpinions 4066 7649 154122 217017 5 times 10

minus3

Friendfeed 4148 5700 96942 386804 41 times 10minus3

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0

120582

08

07

06

05

04

03

02

t

(a) Epinions

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0120582

t

07

06

05

04

03

02

01

(b) Friendfeed

Figure 2 Ranking score values on Epinions and Friendfeed data sets (color online)

list By averaging over all usersrsquo precisions we can obtain thewhole recommender systemsrsquo precision as

Precision = 1119898

119898

sum

119894=1

Precision119894 (9)

where119898 represents the number of users Obviously a higherprecision means a higher recommendation accuracy

(2) Recall [43] Recall represents the probability that therecommended objects appeared in userrsquos collected list shownas

Recall119894=119873119894

119903119904

119873119894119901

(10)

where Recall119894represents user 119906

119894rsquos recall and119873119894

119901is the number

of objects collected by user 119906119894in the testing set Averaging

over all individualsrsquo recall we can obtain the recall of thewhole recommender system

(3) F-Measure [43] Generally speaking for each user recall issensitive to 119871 and a larger 119871 generally gives a higher recall buta lower precision The F-measure that assigns equal weightfor precision and recall is defined as

119865-measure119894=2 sdot precision

119894sdot recall

119894

precision119894+ recall

119894

(11)

By averaging over all usersrsquo119865-measure we can also obtainthe whole systemrsquos 119865-measure

(4) HD [17] HD is a metric to measure the diversity ofusersrsquo recommendation lists It uses the Hamming distance

to measure the difference of recommendation lists betweenusers 119906

119894and 119906

119895 which is defined as

HD119894119895(119871) = 1 minus

119876119894119895(119871)

119871 (12)

where 119876119894119895(119871) is the number of commonly recommended

objects shown in top-119871 locations of users 119906119894and 119906

119895rsquos recom-

mendation list Averaging over all pairs of usersrsquo HD119894119895(119871) we

can obtain theHDof the recommender algorithmObviouslyhigher HD means higher diversity of users

(5) Ranking Score (119903) [44] Generally the recommendersystem aims to generate a ranking list for the target userrsquosuncollected objects through the prediction score In therecommender systems one of the most used metrics toevaluate the algorithmrsquos performance is ranking score whichmeasures the usersrsquo satisfaction of the ranking list and isdefined as follows

119903119894120572=119871119894120572

119873119894

(13)

where 119871119894120572is the position of uncollected object 120572 in user 119880

119894rsquos

ranking list and 119873119894is the length of the user 119880

119894rsquos ranking list

By averaging all linksrsquo ranking score value we can obtain thewhole systemrsquos ranking score value 119903 A small 119903 means therecommender system puts the userrsquos favorite objects in a topplace in the recommender list hence the smaller 119903 is thebetter an algorithmrsquos performance will be

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World Journal 3

onWeb 20 platform which can improve the prediction accu-racy compared with the original CF algorithm We believethat if the social relationship can be used to supplementthe user-object network like the aforementioned example ofFigure 1 we will get more accurate recommendations andalleviate the user cold-start problem Motivated by this weproposed a biased random walk (diffusion-based) methodon coupled social network to generate recommendationsTherefore new users can obtain recommendations as long asthey are connected to others in social networks

3 Method

In this section we introduce the approach of diffusion oncoupled social networks Generally a recommender systemconsists of two sets 119880 = 119880

1 1198802 119880

119898 and 119874 =

1198741 1198742 119874

119899 representing the 119898 users and 119899 objects

respectively Denote119860119898times119899

by the adjacent matrix of the user-object bipartite network of which each element 119886

119894120572= 1

if user 119880119894has collected object 119874

120572 and 119886

119894120572= 0 otherwise

Analogously denote 119861119898times119898

by the nonsymmetric adjacentmatrix of user-user directed social network of which eachelement 119887

119894119895= 1 if the user119880

119894has linked to user119880

119895 and 119887

119894119895= 0

otherwise

Random Walk on Social Network Let 1198751015840 be the 119898 times 119898transition probabilitymatrix of a directed social networkTheprobability that a randomwalker at user119880

119894goes to user119880

119895on

social network can be described as

1199011015840

119894119895=

119887119894119895

119896out119894

if 119896out119894= 0

0 otherwise(1)

where 119896out119894

is the out-degree in social network that is thenumber of leaders of user 119880

119894 Denote 1199041015840

119894(119905) by the probability

from other users to user 119880119894at time 119905 Therefore we have

1199041015840

119894(119905 + 1) =

119898

sum

119895=1

119887119894119895

119896out119894

1199041015840

119895(119905) if 119896out

119894= 0

0 otherwise(2)

The initial probability for target user 119880119894is given by 1199041015840

119894(0) = 1

and 1199041015840119895(0) = 0 for all of the other user119880

119895 Thus we can obtain

the probability that a randomwalker goes from the target userto all other users at time 119905

Random Walk on Bipartite Network Let 11987510158401015840 be the 119898 times119899 transition probability matrix of a bipartite network Theprobability that a random walker at user119880

119894goes to object119874

120572

on bipartite network can be described as

11990110158401015840

119894120572=

119886119894120572

119896119894

if 119896119894= 0

0 otherwise(3)

where 119896119894denotes the number of collected objects of user 119880

119894

and the probability that a randomwalker at object119874120572goes to

user 119880119895on bipartite network can be described as

11990110158401015840

120572119895=

119886119895120572

119896120572

if 119896120572= 0

0 otherwise(4)

where 119896120572denotes the number of users who have collected

object119874120572on bipartite network Denote 11990410158401015840

119894(119905) and 11990410158401015840

120572(119905) by the

probability of user 119880119894and object 119874

120572on bipartite network at

time 119905 respectively Therefore we have

11990410158401015840

119894(119905 + 1) =

119899

sum

120572=1

119886119894120572

119896119894

11990410158401015840

120572(119905) if 119896

119894= 0

0 otherwise

11990410158401015840

120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

11990410158401015840

119895(119905) if 119896

120572= 0

0 otherwise

(5)

Similar to random walk on social network the initial proba-bility for target user119880

119894is given by 11990410158401015840

119894(0) = 1 But the difference

is the fact that there are two different nodes on bipartitenetwork and the initial probability 11990410158401015840

119895(0) = 0 and 11990410158401015840

120572(0) = 0

for all the other user119880119895and object 120572 In the odd time step and

119905 ge 3 the probability of 11990410158401015840120572(119905)means the probability of target

user 119880119894selecting uncollected object 119874

120572 Therefore we can

obtain the recommendation list according to this probabilityfor target user

Biased Random Walk on Coupled Social Network Let 119875 bethe119872 times119872 transition probability matrix of a coupled socialnetwork where119872 = 119898 + 119899 In order to solve the user cold-start problem suppose that a random walker at user 119880

119894goes

to their neighbors (leaders) on directed social network withprobability 120582 isin (0 1) and to their neighbors on bipartitenetworkwith probability 1minus120582Whatrsquosmore a randomwalkerat object119874

120572goes to all users who collect object119874

120572with equal

probability Thus the target user finds the potential objectsnot only through other users with similar collecting intereston bipartite network but also through their friends (leaders)on directed social network Denote 119904

119894(119905) and 119904

120572(119905) by the

probability of walker user119880119894and object119874

120572on coupled social

network at time 119905 respectively Therefore we have

119904119894(119905 + 1) =

120582 sdot

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) + (1 minus 120582)

sdot

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

119898

sum

119895=1

119887119894119895

119896out119894

119904119895(119905) if 119896

119894= 0 119896

out119894= 0

119899

sum

120572=1

119886119894120572

119896119894

119904120572(119905) if 119896

119894= 0 119896

out119894= 0

0 otherwise

4 The Scientific World Journal

119904120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

119904119895(119905) if 119896

120572= 0

0 otherwise(6)

That is to say initially we assign the target user one unitof resource Then 120582 (0 le 120582 le 1) proportion of the resourceis evenly distributed to the userrsquos social neighbors throughthe directed links (social network) and 1 minus 120582 proportion isdistributed to collected objects through the undirected links(bipartite network) In (6) when 119896out

119894= 0 then 120582 = 0

it means that user 119880119894has no outlinks in social network

therefore heshe will distribute all of hisher resources tobipartite network Similarly when 119896

119894= 0 then 120582 = 1 user

119880119894will distribute all of hisher resources to social network

The initial score for target user 119880119894is given by 119904

119894(0) = 1

119904119895(0) = 0 and 119904

120572(0) = 0 for all the other user 119880

119895and object

120572 Thus we can obtain the recommendations by ranking thescore 119904

120572of all objects at time 119905 for target user At time 119905 = 2

the recommendations are obtained only from social networkthat is hisher social leaders At time 119905 = 3 and 120582 = 0 therecommendations are obtained only from bipartite networkand it is the pure MD algorithm

Thus the probability that a random walker arrives atthe object at time 119905 is recognized as the possibility thatthe target user purchases this object We call this algorithmbiased random walk (BRW) For the example in Figure 1 thetransition probability matrix 119875 for coupled social network isgiven in the following equation

119875 =

1198801

1198802

1198803

1198804

1198805

1198741

1198742

1198743

1198744

1198745

1198801119880211988031198804119880511987411198742119874311987441198745

((((((((

(

0 12 0 0 0 14 14 0 0 0

14 0 14 0 0 14 14 0 0 0

16 0 0 16 16 0 14 14 0 0

0 0 0 0 12 0 0 14 14 0

0 0 12 0 0 0 0 0 0 12

12 12 0 0 0 0 0 0 0 0

13 13 13 0 0 0 0 0 0 0

0 0 12 12 0 0 0 0 0 0

0 0 0 1 0 0 0 0 0 0

0 0 0 0 1 0 0 0 0 0

))))))))

)

(7)

Consider 120582 = 05 and 119905 = 2 then 119875(1198803 1198745) = 00833

which means users 1198803and 119874

5are reachable within 2 steps

with 00833 probability through the coupled social networkOn the other hand without social network the random walkdistance on the original bipartite network 11987510158401015840(119880

3 1198745) = 0 for

an arbitrary time 119905 because1198803and119874

5are not reachable from

each other in bipartite network

4 Data and Metrics

41 Data Sets To evaluate our algorithmrsquos performance tworeal data sets are analyzed in the experiments The datasets are from httpwwwepinionscom and httpwwwfriendfeedcom both of which provided user-objectscollecting information and user-user social relationshipThe Epinions data set was collected by Paolo Massa ina 5-week crawl (NovemberDecember 2003) from thehttpwwwepinionscom website [36] and the Friendfeeddata set was collected by Fabio Celli et al from httpwwwfriendfeedcom (September 6 2009 to September19 2009) [42] We extract a smaller data set by randomlysampling the whole records of user activities in both Epinionsand Friendfeed data sets 4066 users 7649 objects 154122collected links and 217071 social links in total were foundin the Epinions data set Friendfeed contains 4148 userswho collected 5700 objects 96942 collected links and386804 social links Table 1 shows the basic statistics for

two representative data sets Denote |119880| |119874| and 119873119877by

the number of users objects and ratings respectivelySparsity = 119873

119877(|119880| times |119874|) denotes the data sparsity of

user-objects network

42 Metrics To test our algorithmrsquos performance each infor-mation network is randomly divided into two parts thetraining set consists of 90 entries and the remaining entriesconstitute the testing set The training set is treated asknown information used for generating recommendationswhile the training set is regarded as unknown informationused for testing the performance of the recommendationresults To evaluate the proposed algorithm we employedfive different metrics that characterize not only the accuracyof recommendations but also the diversification which aredefined as follows

(1) Precision [43] Precision represents the probability that theselected objects appeared in the recommendation list whichis shown as

Precision119894=119873119894

119903119904

119871 (8)

where Precision119894represents user 119906

119894rsquos precision 119873119894

119903119904denotes

the number of recommended objects that appeared in the119880119894rsquos

testing set and 119871 represents the length of recommendation

The Scientific World Journal 5

Table 1 Properties of the tested data sets

Data sets Users Objects Collecting links Social links SparsityEpinions 4066 7649 154122 217017 5 times 10

minus3

Friendfeed 4148 5700 96942 386804 41 times 10minus3

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0

120582

08

07

06

05

04

03

02

t

(a) Epinions

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0120582

t

07

06

05

04

03

02

01

(b) Friendfeed

Figure 2 Ranking score values on Epinions and Friendfeed data sets (color online)

list By averaging over all usersrsquo precisions we can obtain thewhole recommender systemsrsquo precision as

Precision = 1119898

119898

sum

119894=1

Precision119894 (9)

where119898 represents the number of users Obviously a higherprecision means a higher recommendation accuracy

(2) Recall [43] Recall represents the probability that therecommended objects appeared in userrsquos collected list shownas

Recall119894=119873119894

119903119904

119873119894119901

(10)

where Recall119894represents user 119906

119894rsquos recall and119873119894

119901is the number

of objects collected by user 119906119894in the testing set Averaging

over all individualsrsquo recall we can obtain the recall of thewhole recommender system

(3) F-Measure [43] Generally speaking for each user recall issensitive to 119871 and a larger 119871 generally gives a higher recall buta lower precision The F-measure that assigns equal weightfor precision and recall is defined as

119865-measure119894=2 sdot precision

119894sdot recall

119894

precision119894+ recall

119894

(11)

By averaging over all usersrsquo119865-measure we can also obtainthe whole systemrsquos 119865-measure

(4) HD [17] HD is a metric to measure the diversity ofusersrsquo recommendation lists It uses the Hamming distance

to measure the difference of recommendation lists betweenusers 119906

119894and 119906

119895 which is defined as

HD119894119895(119871) = 1 minus

119876119894119895(119871)

119871 (12)

where 119876119894119895(119871) is the number of commonly recommended

objects shown in top-119871 locations of users 119906119894and 119906

119895rsquos recom-

mendation list Averaging over all pairs of usersrsquo HD119894119895(119871) we

can obtain theHDof the recommender algorithmObviouslyhigher HD means higher diversity of users

(5) Ranking Score (119903) [44] Generally the recommendersystem aims to generate a ranking list for the target userrsquosuncollected objects through the prediction score In therecommender systems one of the most used metrics toevaluate the algorithmrsquos performance is ranking score whichmeasures the usersrsquo satisfaction of the ranking list and isdefined as follows

119903119894120572=119871119894120572

119873119894

(13)

where 119871119894120572is the position of uncollected object 120572 in user 119880

119894rsquos

ranking list and 119873119894is the length of the user 119880

119894rsquos ranking list

By averaging all linksrsquo ranking score value we can obtain thewhole systemrsquos ranking score value 119903 A small 119903 means therecommender system puts the userrsquos favorite objects in a topplace in the recommender list hence the smaller 119903 is thebetter an algorithmrsquos performance will be

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

4 The Scientific World Journal

119904120572(119905 + 1) =

119898

sum

119895=1

119886119895120572

119896120572

119904119895(119905) if 119896

120572= 0

0 otherwise(6)

That is to say initially we assign the target user one unitof resource Then 120582 (0 le 120582 le 1) proportion of the resourceis evenly distributed to the userrsquos social neighbors throughthe directed links (social network) and 1 minus 120582 proportion isdistributed to collected objects through the undirected links(bipartite network) In (6) when 119896out

119894= 0 then 120582 = 0

it means that user 119880119894has no outlinks in social network

therefore heshe will distribute all of hisher resources tobipartite network Similarly when 119896

119894= 0 then 120582 = 1 user

119880119894will distribute all of hisher resources to social network

The initial score for target user 119880119894is given by 119904

119894(0) = 1

119904119895(0) = 0 and 119904

120572(0) = 0 for all the other user 119880

119895and object

120572 Thus we can obtain the recommendations by ranking thescore 119904

120572of all objects at time 119905 for target user At time 119905 = 2

the recommendations are obtained only from social networkthat is hisher social leaders At time 119905 = 3 and 120582 = 0 therecommendations are obtained only from bipartite networkand it is the pure MD algorithm

Thus the probability that a random walker arrives atthe object at time 119905 is recognized as the possibility thatthe target user purchases this object We call this algorithmbiased random walk (BRW) For the example in Figure 1 thetransition probability matrix 119875 for coupled social network isgiven in the following equation

119875 =

1198801

1198802

1198803

1198804

1198805

1198741

1198742

1198743

1198744

1198745

1198801119880211988031198804119880511987411198742119874311987441198745

((((((((

(

0 12 0 0 0 14 14 0 0 0

14 0 14 0 0 14 14 0 0 0

16 0 0 16 16 0 14 14 0 0

0 0 0 0 12 0 0 14 14 0

0 0 12 0 0 0 0 0 0 12

12 12 0 0 0 0 0 0 0 0

13 13 13 0 0 0 0 0 0 0

0 0 12 12 0 0 0 0 0 0

0 0 0 1 0 0 0 0 0 0

0 0 0 0 1 0 0 0 0 0

))))))))

)

(7)

Consider 120582 = 05 and 119905 = 2 then 119875(1198803 1198745) = 00833

which means users 1198803and 119874

5are reachable within 2 steps

with 00833 probability through the coupled social networkOn the other hand without social network the random walkdistance on the original bipartite network 11987510158401015840(119880

3 1198745) = 0 for

an arbitrary time 119905 because1198803and119874

5are not reachable from

each other in bipartite network

4 Data and Metrics

41 Data Sets To evaluate our algorithmrsquos performance tworeal data sets are analyzed in the experiments The datasets are from httpwwwepinionscom and httpwwwfriendfeedcom both of which provided user-objectscollecting information and user-user social relationshipThe Epinions data set was collected by Paolo Massa ina 5-week crawl (NovemberDecember 2003) from thehttpwwwepinionscom website [36] and the Friendfeeddata set was collected by Fabio Celli et al from httpwwwfriendfeedcom (September 6 2009 to September19 2009) [42] We extract a smaller data set by randomlysampling the whole records of user activities in both Epinionsand Friendfeed data sets 4066 users 7649 objects 154122collected links and 217071 social links in total were foundin the Epinions data set Friendfeed contains 4148 userswho collected 5700 objects 96942 collected links and386804 social links Table 1 shows the basic statistics for

two representative data sets Denote |119880| |119874| and 119873119877by

the number of users objects and ratings respectivelySparsity = 119873

119877(|119880| times |119874|) denotes the data sparsity of

user-objects network

42 Metrics To test our algorithmrsquos performance each infor-mation network is randomly divided into two parts thetraining set consists of 90 entries and the remaining entriesconstitute the testing set The training set is treated asknown information used for generating recommendationswhile the training set is regarded as unknown informationused for testing the performance of the recommendationresults To evaluate the proposed algorithm we employedfive different metrics that characterize not only the accuracyof recommendations but also the diversification which aredefined as follows

(1) Precision [43] Precision represents the probability that theselected objects appeared in the recommendation list whichis shown as

Precision119894=119873119894

119903119904

119871 (8)

where Precision119894represents user 119906

119894rsquos precision 119873119894

119903119904denotes

the number of recommended objects that appeared in the119880119894rsquos

testing set and 119871 represents the length of recommendation

The Scientific World Journal 5

Table 1 Properties of the tested data sets

Data sets Users Objects Collecting links Social links SparsityEpinions 4066 7649 154122 217017 5 times 10

minus3

Friendfeed 4148 5700 96942 386804 41 times 10minus3

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0

120582

08

07

06

05

04

03

02

t

(a) Epinions

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0120582

t

07

06

05

04

03

02

01

(b) Friendfeed

Figure 2 Ranking score values on Epinions and Friendfeed data sets (color online)

list By averaging over all usersrsquo precisions we can obtain thewhole recommender systemsrsquo precision as

Precision = 1119898

119898

sum

119894=1

Precision119894 (9)

where119898 represents the number of users Obviously a higherprecision means a higher recommendation accuracy

(2) Recall [43] Recall represents the probability that therecommended objects appeared in userrsquos collected list shownas

Recall119894=119873119894

119903119904

119873119894119901

(10)

where Recall119894represents user 119906

119894rsquos recall and119873119894

119901is the number

of objects collected by user 119906119894in the testing set Averaging

over all individualsrsquo recall we can obtain the recall of thewhole recommender system

(3) F-Measure [43] Generally speaking for each user recall issensitive to 119871 and a larger 119871 generally gives a higher recall buta lower precision The F-measure that assigns equal weightfor precision and recall is defined as

119865-measure119894=2 sdot precision

119894sdot recall

119894

precision119894+ recall

119894

(11)

By averaging over all usersrsquo119865-measure we can also obtainthe whole systemrsquos 119865-measure

(4) HD [17] HD is a metric to measure the diversity ofusersrsquo recommendation lists It uses the Hamming distance

to measure the difference of recommendation lists betweenusers 119906

119894and 119906

119895 which is defined as

HD119894119895(119871) = 1 minus

119876119894119895(119871)

119871 (12)

where 119876119894119895(119871) is the number of commonly recommended

objects shown in top-119871 locations of users 119906119894and 119906

119895rsquos recom-

mendation list Averaging over all pairs of usersrsquo HD119894119895(119871) we

can obtain theHDof the recommender algorithmObviouslyhigher HD means higher diversity of users

(5) Ranking Score (119903) [44] Generally the recommendersystem aims to generate a ranking list for the target userrsquosuncollected objects through the prediction score In therecommender systems one of the most used metrics toevaluate the algorithmrsquos performance is ranking score whichmeasures the usersrsquo satisfaction of the ranking list and isdefined as follows

119903119894120572=119871119894120572

119873119894

(13)

where 119871119894120572is the position of uncollected object 120572 in user 119880

119894rsquos

ranking list and 119873119894is the length of the user 119880

119894rsquos ranking list

By averaging all linksrsquo ranking score value we can obtain thewhole systemrsquos ranking score value 119903 A small 119903 means therecommender system puts the userrsquos favorite objects in a topplace in the recommender list hence the smaller 119903 is thebetter an algorithmrsquos performance will be

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World Journal 5

Table 1 Properties of the tested data sets

Data sets Users Objects Collecting links Social links SparsityEpinions 4066 7649 154122 217017 5 times 10

minus3

Friendfeed 4148 5700 96942 386804 41 times 10minus3

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0

120582

08

07

06

05

04

03

02

t

(a) Epinions

2 3 4 5 6 7 8 9 10

09

08

07

06

05

04

03

02

01

0120582

t

07

06

05

04

03

02

01

(b) Friendfeed

Figure 2 Ranking score values on Epinions and Friendfeed data sets (color online)

list By averaging over all usersrsquo precisions we can obtain thewhole recommender systemsrsquo precision as

Precision = 1119898

119898

sum

119894=1

Precision119894 (9)

where119898 represents the number of users Obviously a higherprecision means a higher recommendation accuracy

(2) Recall [43] Recall represents the probability that therecommended objects appeared in userrsquos collected list shownas

Recall119894=119873119894

119903119904

119873119894119901

(10)

where Recall119894represents user 119906

119894rsquos recall and119873119894

119901is the number

of objects collected by user 119906119894in the testing set Averaging

over all individualsrsquo recall we can obtain the recall of thewhole recommender system

(3) F-Measure [43] Generally speaking for each user recall issensitive to 119871 and a larger 119871 generally gives a higher recall buta lower precision The F-measure that assigns equal weightfor precision and recall is defined as

119865-measure119894=2 sdot precision

119894sdot recall

119894

precision119894+ recall

119894

(11)

By averaging over all usersrsquo119865-measure we can also obtainthe whole systemrsquos 119865-measure

(4) HD [17] HD is a metric to measure the diversity ofusersrsquo recommendation lists It uses the Hamming distance

to measure the difference of recommendation lists betweenusers 119906

119894and 119906

119895 which is defined as

HD119894119895(119871) = 1 minus

119876119894119895(119871)

119871 (12)

where 119876119894119895(119871) is the number of commonly recommended

objects shown in top-119871 locations of users 119906119894and 119906

119895rsquos recom-

mendation list Averaging over all pairs of usersrsquo HD119894119895(119871) we

can obtain theHDof the recommender algorithmObviouslyhigher HD means higher diversity of users

(5) Ranking Score (119903) [44] Generally the recommendersystem aims to generate a ranking list for the target userrsquosuncollected objects through the prediction score In therecommender systems one of the most used metrics toevaluate the algorithmrsquos performance is ranking score whichmeasures the usersrsquo satisfaction of the ranking list and isdefined as follows

119903119894120572=119871119894120572

119873119894

(13)

where 119871119894120572is the position of uncollected object 120572 in user 119880

119894rsquos

ranking list and 119873119894is the length of the user 119880

119894rsquos ranking list

By averaging all linksrsquo ranking score value we can obtain thewhole systemrsquos ranking score value 119903 A small 119903 means therecommender system puts the userrsquos favorite objects in a topplace in the recommender list hence the smaller 119903 is thebetter an algorithmrsquos performance will be

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

6 The Scientific World Journal

Table 2 Algorithmic performance for Epinions data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0172 0036 0099 0046 0673UCF 0186 0033 0090 0041 056RW 0171 0036 01 0046 0652

Table 3 Algorithmic performance for Friendfeed data set with recommendation list 119871 = 20

Method 119903 Precision Recall 119865-measure HDMD 0116 003 0140 0041 09405UCF 012 0029 00902 00386 08772RW 0108 003 0141 0041 09250

5 Results

Figure 2 shows the ranking score values on Epinions andFriendfeed data sets From the figure we can see that the bestperformance is achieved at time 119905 = 3 At time 119905 = 2the recommendations are obtained only from social networkand when 120582 = 0 it will generate random recommendationresults since the ranking score value 119903 is much bigger thanothers When 120582 = 0 the resource will spread only on bipartitenetwork therefore objects get scores in odd time steps onlyand user get scores in even time steps only In addition theranking score will fluctuate up and down alternately withtime 119905 That is because when 120582 gt 0 the recommendationsare obtained from social interest in odd time step and fromboth social interests and collecting preferences in even timestep With the increase of time 119905 in even and odd time steprespectively the ranking score becomes worse due to theexistence of the redundant correlations [45]

The best ranking score performance occurs at time119905 = 3 that is when we consider the social interest inthe recommender systems it will improve the performanceof recommender systems Figure 3 shows the experimentalresults of precision recall F-measure HD with recommen-dation list 119871 = 20 and ranking score 119903 on Epinions andFriendfeed data sets at time 119905 = 3 120582 = 0 gives the pureMD algorithm It can be found that when the parameter 120582reaches the optimal value the precision recall 119891-measureand 119903 almost simultaneously reach themaximumvalue exceptthat of HD Tables 2 and 3 show the results of biased randomwalk (BRW) compared with the mass diffusion (MD) anduser-based CF (UCF) on Epinions and Friendfeed data setsrespectively We can see that BRW algorithm has a higherranking-accuracy than other algorithms and almost similaraccuracy-precision with MD but lower diversity-precisionthanMDalgorithm It is because the probability of reciprocitylinks 119903

119871= 119871harr1198711015840 is large in the social network (Epinions data

set is 4547 and Friendfeed data set is 6272) where 119871harr isthe number of bidirectional links and 1198711015840 is the number of alllinks in social network Because it is easier for the randomwalker to go from one user to another user in social networkthe recommendations obtained from social network will besimilar among friends

Generally speaking the small degree users are the vastmajority in the systems (Figure 4 shows the use degreedistribution in the training set on Epinions and Friendfeeddata sets We find that there are 2306 and 615 userswith degrees smaller than 10 on Epinions and Friendfeed datasets resp) That is to say increasing the small degree usersrsquoperformance could result in performance improvement ofthe whole system In Figure 5 we show the effect of userdegrees that is in the training set versus ranking score Fromthe figure we can see that the MD and UCF almost have thesame ability for small degree users and ourmethod has betterperformance thanMDandUCF algorithmMeanwhile it canbe seen that our method considering the social interest intothe recommender system has a better performance for bothlarger and smaller degree users In otherwords it can alleviatethe user cold-start problem

6 Conclusion and Discussion

In a real online recommender system for new users or userswith less collections it is difficult to obtain recommendationsbecause of lack of enough information However if theyare active in the social network the system can obtain therecommendations from their friends or social leaders In thisway the social networks can help us to solve the user cold-start problem

In this paper we proposed a recommendation algorithmvia biased random walk on a two-layer coupled networkuser-object bipartite network and user-user social networkExperiment results on two real data sets indicate that socialinterest and userrsquos preference can be combined together in adelicate way to improve the accuracy metric of recommenda-tion systems Compared with two other baseline algorithmsour algorithm achieves the best precision measure and hasthe best ability of accurately recommending objects to thesmall degree users effectively alleviating the user cold-startproblem

This paper only provides a simple method to incorporatethe social interest into the recommender systems by randomwalk on coupled social-information network while a couple

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World Journal 7

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

0 02 04 06 08 1

120582

024

022

02

018

016

r

013

0125

012

0115

011

0105

01

r

FriendfeedEpinions

004

0035

003

0025

Prec

ision

0035

03

0025

002

Prec

ision

07

065

06

055

HD

1

095

09

085

08

HD

Figure 3 The precision and HD when recommendation list 119871 = 20 and 119903 in the Epinions and Friendfeed data sets Each result is obtained byaveraging over 10 independent runs each of which corresponds on a random division of training set and testing set

of issues remain open for future study (i) The structure andevolution of coupled social networks are still unclear to usbut we believe they will be helpful for designing effective rec-ommendation algorithms (ii)The current algorithm assumesthat a random walker goes to his friend on social networkand his collected objects on bipartite network with the sameprobability we conjecture that an appropriately adjusted

weight assignment will further improve the algorithmicperformance

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

8 The Scientific World Journal

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

10minus1

10minus2

10minus3

10minus4

p(k

)

100 101 102 103

k

Epinions

2036

Friendfeed

615

Figure 4 The user degree distribution of training set on Epinions and Friendfeed data sets

045

04

035

03

025

02

015

01

005

0

r

100 101 102 103

k

0 5 10 15 20

k

05

04

03

02

01

0

r

BRWMDUCF

(a) Epinions

BRWMDUCF

100 101 102 103

k

0 5 10 15

k

07

06

05

04

03

02

01

0

r

07

06

05

04

03

02

01

0

r

(b) Friendfeed

Figure 5 Ranking score values venus degree 119896 on Epinions and Friendfeed data sets (color online) The red line blue line and green lineindicate the performance of BRW MD and UCF respectively The inset figure amplifies that ranking score versus the degree of users from 1to 15

Acknowledgments

The authors acknowledge Jun-Lin Zhou for helpful discus-sions This work was partially supported by the NaturalScience Foundation of China (Grant nos 61103109 11105024and 61300018) and the Special Project of Sichuan YouthScience and Technology Innovation Research Team (Grantno 2013TD0006)

References

[1] A Edmunds and A Morris ldquoProblem of information overloadin business organizations a review of the literaturerdquo Interna-tional Journal of Information Management vol 20 no 1 pp 17ndash28 2000

[2] L Lu M Medo C H Yeung Y Zhang Z Zhang and T ZhouldquoRecommender systemsrdquo Physics Reports vol 519 no 1 pp 1ndash49 2012

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World Journal 9

[3] L C Freeman ldquoCentrality in social networks conceptual clari-ficationrdquo Social Networks vol 1 no 3 pp 215ndash239 1978

[4] Y Ye J Yin and Y Xu ldquoSocial network supported processrecommender systemrdquo The Scientific World Journal vol 2014Article ID 349065 8 pages 2014

[5] F Fu L Liu and L Wang ldquoEmpirical analysis of online socialnetworks in the age ofWeb 20rdquo Physica A Statistical Mechanicsand Its Applications vol 387 no 2-3 pp 675ndash684 2008

[6] M A Nowak ldquoFive rules for the evolution of cooperationrdquoScience vol 314 no 5805 pp 1560ndash1563 2006

[7] G Szabo andG Szabo ldquoEvolutionary games on graphsrdquo PhysicsReports vol 446 no 4ndash6 pp 97ndash216 2007

[8] S Fortunato ldquoCommunity detection in graphsrdquo Physics ReportsA vol 486 no 3ndash5 pp 75ndash174 2010

[9] M N K Boulos and S Wheeler ldquoThe emerging Web 20 socialsoftware an enabling suite of sociable technologies in healthand health care educationrdquo Health Information and LibrariesJournal vol 24 no 1 pp 2ndash23 2007

[10] A I Schein A Popescul L H Ungar and D M PennockldquoMethods and metrics for cold-start recommendationsrdquo inProceedings of the 25th Annual International ACM SIGIR Con-ference on Research and Development in Information Retrievalpp 253ndash260 ACM 2002

[11] E Vozalis and K G Margaritis ldquoAnalysis of recommendersystems algorithmsrdquo in Proceedings of the 6th Hellenic EuropeanConference on Computer Mathematics and Its Applications(HERCMA 03) vol 2003 Athens Greece 2003

[12] F Radicchi and A Arenas ldquoAbrupt transition in the structuralformat ion of interconnected networksrdquo Nature Physics vol 9pp 717ndash720 2013

[13] M deDomenico A Sole-Ribalta E Cozzo et al ldquoMathematicalformulation of multilayer networksrdquo Physical Review X vol 3Article ID 041022 2013

[14] S V Buldyrev R Parshani G Paul H E Stanley and S HavlinldquoCatastrophic cascade of failures in interdependent networksrdquoNature vol 464 no 7291 pp 1025ndash1028 2010

[15] M Givoni and D Banister ldquoAirline and railway integrationrdquoTransport Policy vol 13 no 5 pp 386ndash397 2006

[16] T Zhou J Ren M Medo and Y Zhang ldquoBipartite networkprojection and personal recommendationrdquo Physical Review EStatistical Nonlinear and Soft Matter Physics vol 76 no 4Article ID 046115 2007

[17] T Zhoua Z Kuscsik J Liu M Medo J R Wakeling and YZhang ldquoSolving the apparent diversity-accuracy dilemma ofrecommender systemsrdquo Proceedings of the National Academy ofSciences of the United States of America vol 107 no 10 pp 4511ndash4515 2010

[18] P Resnick N Iacovou M Suchak P Bergstrom and J RiedlldquoGrouplens an open architecture for collaborative filtering ofnetnewsrdquo in Proceedings of the ACM Conference on ComputerSupported Cooperative Work pp 175ndash186 ACM 1994

[19] J B Schafer D Frankowski J Herlocker and S Sen ldquoCollabo-rative filtering recommender systemsrdquo inThe adaptive Web pp291ndash324 Springer New York NY USA 2007

[20] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[21] J L Herlocker J A Konstan A Borchers and J RiedlldquoAn algorithmic framework for performing collaborative fil-teringrdquo in Proceedings of the 22nd Annual International ACM

SIGIR Conference on Research and Development in InformationRetrieval pp 230ndash237 1999

[22] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-basedcollabo rative filtering recommendation algorithmsrdquo in Pro-ceedings of the 10th International Conference on World WideWeb pp 285ndash295 ACM 2001

[23] M Deshpande and G Karypis ldquoItem-based top-N recommen-dation algorithmsrdquo ACM Transactions on Information Systemsvol 22 no 1 pp 143ndash177 2004

[24] J S Breese D Heckerman and C Kadie ldquoEmpirical analysis ofpredicti ve algorithms for collaborative filteringrdquo in Proceedingsof the 14th Conference on Uncertainty in Artificial Intelligence(UAI rsquo98) pp 43ndash52 Morgan Kaufmann Madison Wis USAJuly 1998

[25] G Linden B Smith and J York ldquoAmazoncom recommen-dations item-to-item collaborative filteringrdquo IEEE InternetComputing vol 7 no 1 pp 76ndash80 2003

[26] M J Pazzani and D Billsus ldquoContent-based recommendationsystemsrdquo inThe Adaptive Web pp 325ndash341 Springer 2007

[27] R Burke ldquoHybrid web recommender systemsrdquo inThe AdaptiveWeb pp 377ndash408 Springer New York NY USA 2007

[28] C Palmisano A Tuzhilin and M Gorgoglione ldquoUsing contextto improve predictivemodeling of customers in personalizationapplicationsrdquo IEEE Transactions on Knowledge and Data Engi-neering vol 20 no 11 pp 1535ndash1549 2008

[29] D C Nie M J Ding Y Fu J L Zhou and Z K Zhang ldquoSocialinterest for user selecting items in recommender systemsrdquoInternational Journal of Modern Physics C vol 24 no 4 ArticleID 1350022 2013

[30] Z Zhang T Zhou and Y Zhang ldquoTag-aware recommendersystems a state-of-the-art surveyrdquo Journal of Computer Scienceand Technology vol 26 no 5 pp 767ndash777 2011

[31] K Pearson ldquoThe problem of the random walkrdquo Nature vol 72no 1865 p 294 1905

[32] Z Huang H Chen andD Zeng ldquoApplying associative retrievaltechniques to alleviate the sparsity problem in collaborativefilteringrdquoACMTransactions on Information Systems vol 22 no1 pp 116ndash142 2004

[33] A Zeng A Vidmer M Medo and Y C Zhang ldquoInformationfiltering by similarity-preferential diffusion processesrdquo Euro-physics Letters vol 105 Article ID 58002 2014

[34] P Sarkar and A W Moore ldquoRandom walks in social networksand their applications a surveyrdquo in Social Network DataAnalytics pp 43ndash77 2011

[35] A W Yu N Mamoulis and H Su ldquoReverse top-k searchusing random walk with restartrdquo in Proceedings of the VLDBEndowment vol 7 2014

[36] P Massa and P Avesani ldquoTrust-aware recommender systemsrdquoin Proceedings of the ACMConference on Recommender Systems(RecSys rsquo07) pp 17ndash24 ACM Valley Calif USA October 2007

[37] I Esslimani A Brun and A Boyer ldquoFrom social networks tobehavioral networks in recommender systemsrdquo in Proceedingsof the International Conference on Advances in Social NetworkAnalysis and Mining (ASONAM rsquo09) pp 143ndash148 IEEE July2009

[38] F E Walter S Battiston and F Schweitzer ldquoA model ofa trust-based recommendation system on a social networkrdquoAutonomous Agents and Multi-Agent Systems vol 16 no 1 pp57ndash74 2008

[39] C H Lai D R Liu and C S Lin ldquoNovel personal and group-based trust models in collaborative filtering for document

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

10 The Scientific World Journal

recommendationrdquo Information Sciences vol 239 pp 31ndash492013

[40] B Yin Y Yang and W Liu ldquoExploring social activeness anddyna mic interest in community-based recommender sys-temrdquo in Proceedings of the Companion Publication of the 23rdInternational Conference on World Wide Web Companion pp771ndash776 International World Wide Web Conferences SteeringCommittee 2014

[41] CWei R Khoury and S Fong ldquoWeb 20 Recommendation ser-vice by multi-collaborative filtering trust network algorithmrdquoInformation Systems Frontiers vol 15 no 4 pp 533ndash551 2013

[42] D Crandall D Cosley D Huttenlocher J Kleinberg and SSuri ldquoFeedback effects between similarity and social influencein online communitiesrdquo in Proceedings of the 14th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD 08) pp 160ndash168 August 2008

[43] J L Herlocker J A Konstan L G Terveen and J T RiedlldquoEvaluating collaborative filtering recommender systemsrdquoACM Transactions on Information Systems vol 22 no 1 pp 5ndash53 2004

[44] T Zhou L L Jiang R Q Su and Y C Zhang ldquoEffect of initialconfiguration onnetwork-based recommendationrdquoEurophysicsLetters vol 81 no 5 Article ID 58004 2008

[45] T Zhou R Q Su R R Liu L L Jiang B H Wang and YZhang ldquoAccurate and diverse recommendations via eliminatingredundant correlationsrdquo New Journal of Physics vol 11 ArticleID 123008 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014