predictors of users' willingness to personalize web search: fqas2013
DESCRIPTION
This paper presents a user survey-based analysis of the correlation between the users' willingness to personalize Web search and their social network usage patterns. The participants' responses to the survey questions enabled us to use a regression model for identifying the relationship between SNS variables and willingness to personalize Web search; the obtained results show that there is a strong relationship between willingness for personalized Web search and social network usage patterns.TRANSCRIPT
FQAS 2013
Predictors of Users' Willingness to Personalize Web Search
Arjumand YounusJoint PhD Candidate @ National University of Ireland, Galway and
University of MilanoBicocca, Italy
Colm O'RiordanNational University of Ireland, Galway
Gabriella PasiUniversity of MilanoBicocca, Italy
FQAS 2013
Outline
1.Introduction: Privacy concerns with personalized web search
2.Proposed analysis for users willing to personalize web search
1. Methodology
3. Study Results
1.Prediction model for web search personalization willingness
4.Future Work
5.Conclusions
FQAS 2013
1. Introduction: What are the Concerns Regarding
Web Search Personalization?
FQAS 2013
Concerns with web search personalization
Personalized Web search has emerged as a promising solution to improve the search quality.
However, privacy remains a big concerniGoogle shutting down in November, 2013
Why are users concerned about privacy with Web search personalization?
Accumulation of search history: query logs, clickthrough data etc.
FQAS 2013
New form of interaction on the web
“Web users now on Facebook longer than Google”CNN, September 2010
FQAS 2013
The value of new form of data
Social data is being exploited as a new means for Web search personalization
Bookmarks
Facebook status updates
Tweets
FQAS 2013
The value of new form of data
Social data is being exploited as a new means for Web search personalization
Bookmarks
Facebook status updates
Tweets
This however does not solve the privacy concern
FQAS 2013
2. Proposed Direction for Analysis of Users Willing to Personalize Web
Search
FQAS 2013
Correlation between social network usage patterns and web search personalization
willingness We investigate if the social network usage
patterns can provide an indication of Web search personalization willingness
Consider user A highly active on social networks communicating thoughts on range of topics
Consider user B less active sharing thoughts on limited topics
FQAS 2013
Correlation between social network usage patterns and web search personalization
willingness We investigate if the social network usage
patterns can provide an indication of Web search personalization willingness
Consider user A highly active on social networks communicating thoughts on range of topics
Consider user B less active sharing thoughts on limited topics
It would be interesting to investigate correlations of behaviors of users A and B and their openness to Web
search personalization
FQAS 2013
2.1. Methodology
12
User survey methodology
Design of a user survey to investigate Social network usage patterns of users
Privacy concerns users have with respect to Web search personalization
Various SNS tools investigated in detail along with different characteristics of SNS usage
Large-scale user survey in two parts
First part: 380 respondents from various countries
Second part: 113 respondents from various countries
13
Information about survey respondents (1/2)
Gender N (%)
Male 235 (61.8%)
Female 145 (38.2%)
Location N (%)
Europe 206 (54.2%)
America 21 (5.5%)
Asia 153 (40.3%)
Age N (%)
10-20 0 (0%)
21-30 259 (68.2%)
31-40 87 (22.9%)
41-50 19 (5%)
Above 50 15 (3.9%)
14
Information about survey respondents (2/2)
SNS Tool Details N (%)
Facebook Presence 356 (93.7%)
Twitter Presence 241 (63.4%)
Google+ Presence 239 (62.9%)
LinkedIn Presence 272 (71.6%)
Bookmarking Sites Presence
60 (15.8%)
SNS Usage Details N (%)
Facebook as Most Used
325 (85.5%)
Twitter as Most Used
106 (27.9%)
Google+ as Most Used
30 (7.9%)
LinkedIn as Most Used
17 (4.5%)
15
Measures and variables investigated (1/4)
Personalized Web Search Results Considered as Useful
N (%)
YesNo
188 (49.5%)192 (50.5%)
Awareness about Personalization Feature of Web Search Engines
N (%)
YesNo
275 (72.4%)105 (27.6%)
Awareness about Web Search Engines using Search History Data
N (%)
YesNo
337 (88.7%)43 (11.3%)
Comfortable with Web Search Engines using Search History Data
N (%)
YesNo
196 (51.6%)184 (48.4%)
16
Measures and variables investigated (2/4)
Frequency of Posting on Facebook MEAN (SD)
YesNo
2.99 (0.95)
Frequency of Facebook Likes MEAN (SD)
YesNo
3.22 (0.94)
No. of Facebook Friends N (%)
Less than 100100 – 200200 – 300300 – 400400 – 500More than 500
62 (17.4%)88 (24.7%)85 (23.9%)50 (14.0%)28 (7.9%)43 (12.1%)
17
Measures and variables investigated (3/4)
Survey respondents who used Twitter asked to provide Twitter handle through which we extracted
Number of mentions
Number of retweets
Number of topics in tweets of survey respondents
18
Measures and variables investigated (4/4)
Ever Used Social Networks for Information-Seeking N (%)
YesNo
272 (71.6%)108 (28.4%)
Q&A Activity on Social Networks Considered as Useful
N (%)
YesNo
187 (49.2%)193 (50.8%)
Frequency of Asking Questions on SNS MEAN (SD)
1.99 (0.95)
Frequency of Considering Answers on SNS More Reliable than Search Engines
MEAN (SD)
2.38 (1.03)
FQAS 2013
3. Study Results
FQAS 2013
Analysis of personalization willingness (1/6)
WP AP AH WH
Male 1.635** 2.643** 6.318*** 1.480*
American 0.001 0.004 4.478 0.002
Asian 0.001 0.003 0.004 0.001
European 0.001 0.004 0.001 0.001
Age 1.069 0.997 0.981 0.841
Note *p<.05, **p<.01, ***p<.001
WP: Willingness for Web search personalizationAP: Awareness of personalizationfeature in Web search enginesAH: Awareness of fact that search engines use search history data for personalizationWH: Comfort level with fact that search engines use search history data for personalization
FQAS 2013
Analysis of personalization willingness (2/6)
WP AP AH WH
Facebook Presence
0.982 0.561 0.860 1.844
Twitter Presence
1.544* 1.692 3.651*** 1.519*
Google+ Presence
1.816*** 1.427 1.364 1.626**
LinkedIn Presence
0.940 2.475* 2.031** 1.150
Bookmarking Sites Presence
1.289 2.535 1.153 1.771**
Note *p<.05, **p<.01, ***p<.001
FQAS 2013
Analysis of personalization willingness (3/6)
WP AP AH WH
High Usage of Facebook
1.599 0.736 0.488 1.222
High Usage of Twitter
1.166* 1.574 3.986** 1.353
High Usage of Google+
3.042*** 1.565 0.637 2.292**
High Usage of LinkedIn
1.193 2.069 1.127 0.971
Note *p<.05, **p<.01, ***p<.001
FQAS 2013
Analysis of personalization willingness (4/6)
WP AP AH WH
Facebook Usage Frequency
0.898 1.204 1.051 1.231
Facebook Posting Frequency
1.637*** 0.450* 0.893 1.246
Facebook Liking Frequency
0.920 1.776 0.922 0.924
No. of Facebook Friends
0.873* 1.031 1.181 0.899
Note *p<.05, **p<.01, ***p<.001
FQAS 2013
Analysis of personalization willingness (5/6)
WP AP AH WH
Twitter Mentions
1.000 0.983 0.997 0.998*
Twitter Retweets
1.001 0.982 0.999 0.999
No. of Topics in Tweets
0.997 0.969 1.036 1.012
No. of Tweets 0.999 1.015* 1.001 1.001*
Note *p<.05, **p<.01, ***p<.001
FQAS 2013
Analysis of personalization willingness (6/6)
WP AP AH WH
Prefers Q&A Activity on SNS 1.000 0.983 0.997 0.998*
Considers Q&A Activity on SNS Useful
1.001 0.982 0.999 0.999
Frequency of Q&A Activity on SNS
0.997 0.969 1.036 1.012
Frequency of Considering Responses from SNS More Useful than Search Engines
0.999 1.015* 1.001 1.001*
Note *p<.05, **p<.01, ***p<.001
FQAS 2013
3.1. Prediction Model for Web Search Personalization Willingness
27
Prediction model details
GoalTo see if prediction accuracy would be sufficient for a real personalized Web search system
To explore the value of various types of information in the process of automatically determining willingness for Web search personalization
Second set of user survey data (113 respondents) used as test data
Data collected in first phase (380 respondents) used as training data
Support vector machines utilized for prediction model
28
Prediction model results
FQAS 2013
4. Future Work
30
Implications
Users' privacy concerns are a significant challenge within the domain of Web search personalization
We can use inferences from social network usage patterns to address the question of when to personalize and when not to personalize
This work serves as first step in direction of understanding target audience of personalized search systems
We aim to incorporate more aspects of user dimensions (via social traces left by users) in personalized search algorithms
FQAS 2013
5. Conclusions
32
Summary
Utilized survey methodology to gather relevant data and investigating correlations between users' social network usage patterns and their openness to opt for Web search personalization
SNS features such as user's presence/absence and amount of usage activity on particular social networking platforms along with his Q&A activity on social networks provide valuable insights
Significant implications for design of future personalized search and social search applications
FQAS 2013
Thanks for your attention!
Questions?
Arjumand Younus
Email: [email protected]: @ArjumandYounus