Exploiting Tag-Based Personalization for Recommendation on Social Web Frederico Dur

1
Exploiting Tag-Based Personalization for
Recommendation on Social Web Frederico
DurãoAalborg Denmark13.02.2012
Advised by Prof. Peter Dolog
2
Outline

• Introduction
• Motivation, problem statement, research questions
• Literature Review
• User modeling, recommender systems and search
  engine
• Research Overview
• Intuition, model and achievements
• Conclusion and Future Works
• Contributions and looking ahead

3
Motivation

• The World Wide Web was born in the early nineties
  when Tim Berners-Lee had the idea of sharing
  information between scientists from remote
  laboratories Tim Berners-Lee and Mark Fischett,
  2000.
• The Web 2.0 has instituted decentralized content
  creation, thus leading the Web into a more open,
  connected, and democratic environment Tim
  OReilly, 2005.
• Social Web an environment for online
  communication and collaboration in which user
  participation is the primary driver of value
  Breslin, 2009.
• An underlying feature of social web applications,
  Collaborative Tagging allows users to assign
  keywords, known as TAGS, to resources on the Web
  such as photos, videos, and websites.

4
Social Tagging Applications
Tags
5
Taggingltuser, tag, resourcegt

• User
• You, me, a system, someone who assigns a tag to a
  resource
• Tag
• a word, or a set of words that describe a
  resource
• Resource
• text, link, bookmark, image, video, you decide

user
Picture of Maradona
Tags
Tags
Picture of Pele
6
Personalized Tag-based Recommendation
Tag Amount
F1 100
Ferrari 70
BMW 50
Racing 20
Woman 2
Tag Amount
Travel 80
Beach 40
Paradise 50
Holidays 20
Work 1
7
Problems

• The exposure of users to both tags and resources
  creates the immense availability of social data
  on the Web.
• Coping with such amount of information becomes
  critical since users on the Web are heterogeneous
  and have distinct interests.

People are different in taste and preferences
How to make proper recommendations ?
8
Research Questions

• RQ1 How can we learn and rank user preferences
  from tagging information for personalization?
• RQ2 How can we personalize information using
  tag-based user profiles for social web
  applications?
• RQ3 How can social aspects impact the
  performance of tag-based personalization models
  for social web applications?

9
Literature Review

• User Modeling
• An essential activity for maintaining information
  about the users knowledge, beliefs, goals,
  abilities, attitudes, and preferences. Peter
  Brusilovsky, 2001
• Recommender Systems
• Software tools that support users to make
  decisions among various alternatives by
  suggesting items that could be of interest of
  them. Tariq Mahmood and Francesco Ricci, 2009
• Search Engines
• Computer programs that support users to find
  specific piece of information within vast
  collection of documents. Ricardo Baeza-Yates
  and Berthier Ribeiro-Neto, 1999

10
Related Work

• User Modeling

Stereotype-Based - User Models for E-Commerce
Tagging-Based
Highlights
11
Related Work (2)

• Recommender Systems

Collaborative Filtering - Content-Based
- PageRank/FolkRank
Highlights
(ratings) - (item
features/tags) - (links/taggings)
12
Related Work (3)

• Search Engine

User Clicks - Query Log - Query Expansion -
Social Search
Highlights
13
Research Overview
Paper N. Title Research Question
1 A Personalized Tag-Based Recommendation in Social Web Systems 1,2
2 Extending a Hybrid Tag-based Recommender System with Personalization 2
3 A Multi-Factor Tag-Based Personalized Search 1,2
4 Social and Behavioral Aspects of a Tag-based Recommender System 3
5 Recommending Open Linked Data in Creativity Sessions using Web Portals with Collaborative Real Time Environment 2,3
6 Improving Tag-Based Recommendation with the Collaborative Value of Wiki Pages for Knowledge Sharing 1,2,3
14
Paper 1 A Personalized Tag-Based Recommendations
in Social Web Systems

• Assumption
• Resources that share same tags
• are likely to be related to each other.
• Personalization Tag-Based Model
• to recommend unknown documents dmax,u
  for all user , which maximize the
  personalized function persRec as
• Multifactor Model

(Tags football, sports)
(Tags reading, books)
(Tags football, books)
15
Paper 1 A Personalized Tag-Based Recommendations
in Social Web Systems

• Evaluation Goal (Qualitative Assessment of the
  Tag-based Recommender)
• Assess the efficiency of the proposed approach by
  measuring the degree of satisfaction of users
  about the received recommendations.
• Methodology
• 38 participants from 12 countries
• Participant -gt Del.icio.us account -gt 10
  bookmarks at least
• 5542 tags and 1143 bookmarks.
• Experiment Results

Recommendations Accepted Rejected
5/5 59 41
gt 3/5 58 42
16
Paper 2 Extending a Hybrid Tag-based Recommender
System with Personalization

• Assumption
• Syntax variations reduces considerably the
  chances of finding tag similarity.
• Example tags Berlin and Germany are related
  and should be considered in the similarity
  calculus.
• Semantic Similarity Model the semantic factor
• SemanticSimilarity(tag1,tag2)
  WordNet(tag1,tag2) x Ontology(tag1, tag2)
• Example
• SemanticSimilarity(Berlin,Germany)

17
Paper 2 Extending a Hybrid Tag-based Recommender
System with Personalization

• Semantic Group Achievements
• Evaluation / Comparison with previous study
• Example of tag sets similarity calculus

WordNet Ontologies Both
45 24 31
Precision Rates Precision Rates
Recommendations Accepted Items Rejected Items
Pure Syntax-Based 70 57
Semantic Factor 77 40
18
Paper 3 A Multi-Factor Tag-Based Personalized
Search
(2)
(1)

• Assumption
• Multiples indicators of user preference
• can better determine what is more or less
• relevant than a single one.
• Multi-factor Personalized Search Model
• Tag Frequency
• Weighing Factor
  Example
• One factor can be more important that the other.

(3)
Factor Weigh
Users personal tags (1) 0.6
Tags assigned to favorite pages (2) 0.3
Tags assigned to frequent visited pages (3) 0.1
19
Paper 3 A Multi-Factor Tag-Based Personalized
Search

• Evaluation
• MovieLens dataset 1,147,048 ratings and 95,580
  tags applied to 10,681 movies by 13,190 users.
• Comparison of our approach against 6 similarity
  methods
• Cosine Similarity, Matching Coefficient, Dice,
  Jaccard and Euclidean Distance
• Results

61.6 of precision improvement over traditional
text-based information retrieval (non-pers)
and 6,13 of precision gain over cosine
similarity, the best method compared
20
Paper 4 Social and Behavioral Aspects of a
Tag-based Recommender System

• Exploratory Research
• Which social issues interfered in the performance
  recommendations ?
• Focus on users tagging behavior.
• Question 1 Why correctly generated
• recommendations were rejected?
• Low novelty. Items already known.
• Items not really interesting.
• Recommendation written in a language x but
  tagged in English.
• Therefore, recommendations are correctly
  generated.
• Ex Some Danish rejected recommendations written
  in Chinese because they simply could not read
  them.

21
Paper 4 Social and Behavioral Aspects of a
Tag-based Recommender System

• Question 2 How is your purpose while tagging?
• Question 3 Do you tag only for self
  understating?

Social means use of popular terms
Social tags facilitate Tag-based recommendations
22
Paper 5 Recommending Open Linked Data in
Creativity Sessions Using Web Portals with
Collaborative Real Time Environment

• Assumption
• Semantic-based recommendations
• can support participants in a brainstorming
  sessions.
• Method
• Searching semantic relations on the from Linked
  Open Data on the Web.
• Adapting to a creative technique 5W1H - When,
  Where, Who, What, Why and How

Suggestions for Desert
Camel
Sahara
Dunes
Interrogative Property Mapping
Interrogatives
What
When
Who
Where
How
Ontology Property
type
hasYear
foaf
stateOf
toDo
23
Paper 5 Recommending Open Linked Data in
Creativity Sessions Using Web Portals with
Collaborative Real Time Environment

• Evaluation
• Comparison of groups (A, B, C) without semantic
  recommendations against groups with the support
  of recommendations (D, E, F).
• Results
• On average 55,9 of the recommendations
  were rated with highest ratings (4-5) whereas
• 44,1 of the
  recommendations were rated with lowest ratings
  (1-2-3).
• User comments

24
Paper 6 Improving Tag-Based Recommendation with
the Collaborative Value of Wiki Pages for
Knowledge Sharing

• Assumption
• The collaborative value of wiki pages (such as
  Wikipedia) reflects its knowledge sharing
  capacity.
• We compute this social commitment and used it as
  ranking factor.
• Intended for problem solving.
• Method
• WPCV - Wiki Page Collaborative Value.

Which page should be recommended?
Nelson Mandela
Wiki Page1 Wiki Page 2
Fred
Obama
Silvio Berlusconi
- User Knowledge
- Everything you write, read, tag, comment
increases UK.
- User Interactivity
- You knowledge increases the more interact with
your friends.
25
Paper 6 Improving Tag-Based Recommendation with
the Collaborative Value of Wiki Pages for
Knowledge Sharing

• Evaluation Methodology
• 63 participants using a semantic wiki solve a
  given task
• To ?ll out incomplete wiki pages by collecting
  information placed in other pages of the system.
• They were required to navigate through the pages
  using our recommendations to ?nd the needed
  information.
• Two set of recommendations were provided one
  powered by WPCV and the other purely tag-based
  recommendation.
• Results
• Qualitative Assessment
• Subjective Assessment

Improvements of precision, recall and f-measure
at rates of 11, 7 and 12 against PTB.
user participation over time (30 days)
26
Conclusion

• RESEARCH QUESTION 1
• Tag-based models to capture and rank user
  preferences by observing user activity including
  tagging, searching, rating, commenting and social
  networking.
• RESEARCH QUESTION 2
• Multi-factor personalization model to generate
  user oriented information from tag-based user
  profiles.
• RESEARCH QUESTION 3
• Analyses of social and behavioral aspects that
  harm the effectiveness of recommendations.

27
Future Works

• Flexible user models that allow individuals to
  interact with personalization systems.
• Use of a time decay factor for adjusting users
  preference over time.
• Recommendations should invest on diversity to
  avoid redundancy.
• Attenuate the new user problem by moving towards
  a hybrid approaches.
• Personalization is never dissociated of criticism
  since it invades the users privacy.

28
References

• 31 A. Kobsa. Generic user modeling systems.
  User modeling and user-adapted interaction,
  11(1)4963, 2001.
• 33 Giorgio Brajnik and Carlo Tasso. A shell for
  developing non-monotonic user modeling sys-tems.
  Int. J. Hum.-Comput. Stud., 403162, January
  1994. 15
• 34 J. Kay. Reusable tools for user modelling.
  Artificial Intelligence Review, 7241251, 1993.
• 35 Alfred Kobsa and Wolfgang Pohl. The User
  Modeling Shell System BGP-MS. In In User
  Modelling and User-adapted Interaction, pages
  59106, 1995.
• 36 Wolfgang Pohl. Logic-Based Representation
  and Reasoning for User Modeling Shell Systems.
  User Modeling and User-Adapted Interaction,
  9217282, January 1998.
• 38 D. Cooperstein, K. Delhagen, A. Aber, and K.
  Levin. Making Net shoppers loyal. Forrester
  Research, Cambridge, MA June, 1999.
• 39 P. Hagen, H. Manning, and R. Souza. Smart
  personalization. Forrester Research, Cambridge,
  MA, 1999.
• 45 Badrul Sarwar, George Karypis, Joseph
  Konstan, and John Riedl. Analysis of
  recommen-dation algorithms for e-commerce. In
  Proceedings of the 2nd ACM conference on
  Electronic commerce, EC 00, pages 158167, New
  York, NY, USA, 2000. ACM.
• 51 David Carmel, Naama Zwerdling, Ido Guy,
  Shila Ofek-Koifman, Nadav Harel, Inbal Ronen,
  Erel Uziel, Sivan Yogev, and Sergey Chernov.
  Personalized social search based on the users
  social network. In CIKM 09 Proceeding of the
  18th ACM conference on Information and knowledge
  management, pages 12271236, New York, NY, USA,
  2009. ACM.
• 53 Markus Strohmaier. Purpose tagging
  capturing user intent to assist goal-oriented
  social search. In Proceeding of the 2008 ACM
  workshop on Search in social media, SSM 08,
  pages 3542, New York, NY, USA, 2008. ACM.

29
References (2)

• 74 Pavan Kumar Vatturi, Werner Geyer, Casey
  Dugan, Michael Muller, and Beth
  Brownholtz.Tag-based filtering for personalized
  bookmark recommendations. In CIKM 08 Proceeding
  of the 17th ACM conference on Information and
  knowledge mining, pages 13951396, New York, NY,
  USA, 2008. ACM.
• 75 Shiwan Zhao, Nan Du, Andreas Nauerz, Xiatian
  Zhang, Quan Yuan, and Rongyao Fu. Improved
  recommendation based on collaborative tagging
  behaviors. In IUI 08 Proc. of the 13th Intl.
  conference on Intelligent user interfaces, pages
  413416, New York, NY, USA, 2008. ACM.
• 76 Toine Bogers. Recommender Systems for Social
  Bookmarking. Phd thesis, Tilburg University,
  Tilburg, The Netherlands, dec 2009. 24, 25,
• 77 Robert Jaschke, Leandro Marinho, Andreas
  Hotho, Schmidt-Thie Lars, and Stum Gerd. Tag
  recommendations in social bookmarking systems. AI
  Commun., 21231247, December 2008. 25, 35,
• 78 Marek Lipczak, Yeming Hu, Yael Kollet, and
  Evangelos Milios. Tag Sources for Recommen-dation
  in Collaborative Tagging Systems. In Folke
  Eisterlehner, Andreas Hotho, and Robert
• Jschke, editors, ECML PKDD Discovery Challenge
  2009 (DC09), 497 of CEUR-WS.org, pages 157172,
  September 2009. 25, 35
• 79 A. Byde, H. Wan, and S. Cayzer. Personalized
  tag recommendations via tagging and content-based
  similarity metrics. In Proceedings of the
  International Conference on Weblogs and Social
  Media, 2007.
• 80 Valentina Zanardi and Licia Capra. Social
  ranking uncovering relevant content using
  tag-based recommender systems. In RecSys 08
  Proceedings of the 2008 ACM conference on
  Recommender systems, pages 5158, New York, NY,
  USA, 2008. ACM.
• 81 C. Musto, F. Narducci, M. de Gemmis, P.
  Lops, and G. Semeraro. Star a social tag
  recom-mender system. In Proc. the ECML/PKDD 2009
  Discovery Challenge Workshop, pages 215227.
  Citeseer, 2009.
• 82J. Palau, M. Montaner, B. L opez, and J.L.
  De La Rosa. Collaboration analysis in
  recom-mender systems using social networks.
  Cooperative Information Agents VIII, pages
  137151, 2004.

30
References (3)

• 95 Taher H. Haveliwala. Topic-sensitive
  PageRank. In WWW 02 Proceedings of the 11th
  international conference on World Wide Web, pages
  517526, New York, NY, USA, 2002. ACM.
• 96 Bin Tan, Xuehua Shen, and ChengXiang Zhai.
  Mining long-term search history to improve search
  accuracy. In KDD 06 Proceedings of the 12th ACM
  SIGKDD international conference on Knowledge
  discovery and data mining, pages 718723, New
  York, NY, USA, 2006. ACM.
• 97 Xuehua Shen, Bin Tan, and ChengXiang Zhai.
  Implicit user modeling for personalized search.
  In CIKM 05 Proceedings of the 14th ACM
  international conference on Information and
  knowledge management, pages 824831, New York,
  NY, USA, 2005. ACM.
• 98 Paul Alexandru Chirita, Claudiu S. Firan,
  and Wolfgang Nejdl. Personalized query expansion
  for the web. In SIGIR 07 Proceedings of the
  30th annual international ACM SIGIR conference on
  Research and development in information
  retrieval, pages 714, New York, NY, USA, 2007.
  ACM. 30
• 99 F. Durao, K. Bayyapu, G. Xu, P. Dolog, and
  R. Lage. Using Tag-Neighbors for Query Expansion
  in Medical Information Retrieval. In Information
  Science and Applications (ICISA), 2011
  International Conference on, pages 19. IEEE,
  2011. 30
• 100 Jaime Teevan, Meredith Ringel Morris, and
  Steve Bush. Discovering and using groups to
  improve personalized search. In WSDM 09
  Proceedings of the Second ACM International
  Conference on Web Search and Data Mining, pages
  1524, New York, NY, USA, 2009. ACM.
• 101 Ahu Sieg, Bamshad Mobasher, and Robin
  Burke. Web search personalization with
  ontological user profiles. In CIKM 07
  Proceedings of the sixteenth ACM conference on
  Conference on information and knowledge
  management, pages 525534, New York, NY, USA,
  2007. ACM.
• 102 Jaime Teevan, Susan T. Dumais, and Daniel
  J. Liebling. To personalize or not to
  personalize modeling queries with variation in
  user intent. In SIGIR 08 Proceedings of the
  31st annual international ACM SIGIR conference on
  Research and development in information
  retrieval, pages 163170, New York, NY, USA,
  2008. ACM. 30
• 103 Zhicheng Dou, Ruihua Song, and Ji-Rong Wen.
  A large-scale evaluation and analysis of
  personalized search strategies. In WWW 07
  Proceedings of the 16th international conference
  on World Wide Web, pages 581590, New York, NY,
  USA, 2007. ACM.
• 104 J. Gemmell, A. Shepitsen, M. Mobasher, and
  R. Burke. Personalization in Folksonomies Based
  on Tag Clustering. In Proceedings of the 6th
  Workshop on Intelligent Techniques for Web
  Personalization and Recommender Systems, July
  2008.

31
Thanks

• To my family,
• To my supervisor,
• To the PhD committee members,
• To my MI friends,
• To my IS friends,
• To my KiWi friends,
• To my all Casiopeia personnel,
• To my always helpful secretaries,
• To my work colleagues,
• To my close friends in Aalborg,
• To my M-Eco friends,
• To your patience with me ?

32
Publications

• Nothing more than my obligation ?
• http//vbn.aau.dk/en/persons/frederico-durao(f26ea
  ca1-ec6a-4315-85d4-9c38fa167956)/publications.html

33
Tag-Based Personalization

• Tagged resources and user tag-based profile are
  collected.
• Resource tags are compared against user tag-based
  profile.
• Tag-based personalized are generated.