introduction to recommender systems: why, how, and what

47
Introduction to Recommender Systems Dr. sc. Amancio Bouza Jan. 27th, 2017

Upload: amancio-bouza

Post on 08-Feb-2017

435 views

Category:

Data & Analytics


1 download

TRANSCRIPT

Introduction toRecommender SystemsDr.sc.AmancioBouzaJan.27th,2017

WhoamI

Dr.sc.AmancioBouzaRecommenderSystemExpert,Intrapreneur

Let’sconnect:https://ch.linkedin.com/in/amanciobouza

Followme:https://twitter.com/AmancioBouza

3

1Mio$Whatare

4

1Mio$PrizeforbeatingNetflix’sRecommenderSystemby10%

https://en.wikipedia.org/wiki/Netflix_Prize

5

Why?

Worldhaschanged:OnlineRetailervsB&MRetailer

6

Amazon:2.3Mio* B&M:40k-100k*

[Brynjolfsson etal.,2003];*numberofbooks

forUSbookmarketinYear2000only…imaginetoday!

7

$1.03BinSocialWelfare[Brynjolfsson &.Smith,2000]

8

Whatdoesitmean?

9

[Brynjolfsson &.Smith,2000]

UnlimitedOnlineProductCataloguesandFilteringAlgorithmshave

7-10xHigherImpactonSocialWelfarethanPriceCompetitioninCompetitiveMarkets

10

SearchEngine:

ConsumersearchesrightProductRecommenderSystem:

ProductfindsrightConsumer

ParadigmShift

11

WhatarePreferences?

12

PreferencescanbedescribedwiththeUtilityFunction(Microeconomics)

13

14

ContentFiltering

UseMachineLearningtoLearnanIndividual’sPreferences

15

[Bouzaetal.,2009],[Bouza,2012]

16

->Good->Bad

RepresentPreferences,e.g.,asDecisionTree

17

[Bouza,2012]

Let’sbepragmatic:MachineLearningModelapproximatesUtilityFunction

18

[Bouza,2012]

19

“Let’srateaniPodNano(green)onAmazonandcheckourrecommendations”,2008

20

Well…Doyou call this ausefull recommendation?Basedonapersonaltruestoryin2008

21

CollaborativeFiltering

BasisforCollaborativeFiltering:User-ItemMatrix

22

23

CollaborativeFiltering:Peoplewhosharedsimilarpreferencesinthepast,willsharesimilarpreferencesinthefuture

CollaborativeFiltering

24

Peoplewhosharesimilarprefernces inthepastcontinuetodosointhefuture.

25

My favourites:“BLACKOUT” and “ZERO”

from M. Elsberg

Peoplewhohavesimilarpreferencesinthepast,continuetodosointhefuture.

26

My favourites:“BLACKOUT” and “ZERO”

from M. Elsberg

What about“HELIX”?

27

But…

28

WhataboutbothIndividuals’PreferencesbesideItemsbothrated?

OnlyItemsbothratedprovidesinformation

??

???

?

? ?

29

Hypothesis-BasedCollaborativeFiltering[Bouza,2008-2012]

SimilarityofHypothesizedPreferencesakaMachineLearningModels

30

[Bouza,2009]

31

DailyChallengesinDataPreparation

What’sthemovietitleinChina?

32

TranslatedMovieTitle:

He’saGhost

http://thoughtcatalog.com/nico-lang/2013/09/51-hilariously-bad-translations-of-movie-titles-that-are-better-than-the-original/

33

Don‘t worry,no spoiler ;)

HypothesizedPartialPreferences(HPP)

34

35

HypothesizedPartialPreferencesSimilarityMatrix

36

[Bouza,2012]

SimilarityoftwoMachineLearningModels

37

[Bouza,2012]

Hypothesis-BasedCollaborativeFilteringoutperformsothers

38

[Bouza,2012]

39

40

OtherApplicationsof

MachineLearningModelSimilarity

ApplyingHypothesis-BasedCollaborativeFilteringtoCross-ProjectDefectPrediction

41[Zenger&Bouza,2011] http://www.merlin.uzh.ch/publication/show/4535

BuildingaTexasHold’em PokerBotbasedonaHypothesis-BasedCollaborativeFilteringApproach

42[Kaul &Bouza,2011] http://www.merlin.uzh.ch/publication/show/2488

GroupRecommendationsinGoogleWaveConversionviaChatBotandRecommenderAPI

43[Bouza,2009] http://blog.cpoet.net/providing-movie-recommendations-to-groups-in-google-wave/

Yourturn!Whereareyougoingtoapplythismagicformula?

44

Getit

FreeDownload:https://www.linkedin.com/pulse/how-ais-collaborate-retrieve-recommendations-everybody-amancio-bouza?trk=pulse_spock-articles

BuyBookonAmazon:110$https://www.amazon.com/dp/1105585085

45

MoreRecommendationApplications:http://blog.cpoet.net/

Findmoreapplicationandideasthatmasterandbachelorstudentsimplementedhere:http://blog.cpoet.net/what-makes-a-location-really-interesting-and-how-to-exploit-it-to-improve-location-recommendations/http://blog.cpoet.net/applying-collaborative-filtering-to-cross-project-defect-prediction/http://blog.cpoet.net/building-an-agent-for-texas-holdem-poker-based-on-a-recommender-system/http://blog.cpoet.net/how-ratings-and-trust-inferrencing-establish-healthyow-ratings-and-trust-inferencing-establish-h-online-markets/http://blog.cpoet.net/considering-seasonal-effects-in-location-recommendations-with-recommendation-clouds-metaphor/http://blog.cpoet.net/proposal-using-collaborative-filtering-to-create-a-win-win-win-situation-and-engage-people-with-ubs-has-been-awarded/http://blog.cpoet.net/applicability-of-social-network-graph-patterns-to-recommender-systems/http://blog.cpoet.net/providing-movie-recommendations-to-groups-in-google-wave/http://blog.cpoet.net/personal-private-movie-recommender-system-at-the-semantic-web-challenge/http://blog.cpoet.net/202/http://blog.cpoet.net/personal-cross-site-movie-recommender-system-implemented-as-mozilla-firefox-add-on/http://blog.cpoet.net/probabilistic-partial-user-model-similarity-for-collaborative-filtering/http://blog.cpoet.net/semtree-ontology-based-decision-tree-algorithm-for-recommender-systems-at-the-iswc-2008/

46

Let’sconnectandexchangeideas

Dr.sc.AmancioBouzaRecommenderSystemExpert,Intrapreneur

Let’sconnect:https://ch.linkedin.com/in/amanciobouza

Followme:https://twitter.com/AmancioBouza