1. A Probabilistic Approach to Personalized Tag Recommendation Meiqun Hu, Ee-Peng Lim and Jing Jiang School of Information Systems Singapore Management University

2. Social Tagging Social tagging allows users to annotate resources with tags. – organize tags are keywords, serving as (personalized) index terms that group relevant resources – store online storage gives mobility and convenience to access – share published bookmarks can be viewed by other users – explore to leverage collective wisdom to find interesting resources Image credit @ logorunner.com

3. Personalized Tag Recommendation Personalized tag recommendation aims to recommend tags to the query user for annotating the query resource. Recommendation eases the tagging process. ?

4. Why Personalize Recommendations? Tag recommendation should be personalized. – users exhibit individualized choice of tag terms e.g., language preference – personalized index for personal consumption and consistency

5. Problem Formulation and A Basic Method Problem Formulation: p(t|r q,u q ) A Basic Method: freq-r, to recommend top frequent tags – assuming that the more people have used this tag, the more likely it will be used again – current state-of-the-art in many social tagging sites, e.g., – fails to personalize the recommendations for the query user

6. Three Scenarios Scenario 1: foto is an infrequent tag for the resource. Scenario 2: foto is has not been used for the resource, but has been used by the user for annotating other resources in the past. Scenario 3: foto has not been used for the resource but has been used by others when annotating other resources.

7. Collaborative Filtering Method A Method based on Collaborative Filtering: knn, to select top k-nearest neighbors and recommend tags used by these neighbors for annotating the resource – assuming that there are like-minded users who have annotated the same resource – classic collaborative filtering, without ratings – addresses scenario 1, but – fails scenario 2,3

8. Personomy Translation Method To translate the resources tags to the users personal tags (trans-u) – to learn p(foto|u q, photo) – addresses scenario 2, but – fails scenario 3, if u q has never used foto

9. To Address Scenario 3 borrow translation

10. A PROBABILISTIC FRAMEWORK 1.Personomy Translation 2.A Framework 3.Measuring User Similarity

11. Borrowing Translations To learn p(foto|u,photo) and sim(u,u q ) borrow translation

12. Personomy Translation To learn p(foto|u q,photo) [Wetzker et al. 2009]

13. Measuring Similarity between Users sim(u,u q ) – assuming that users are similar if they perform similar translations – users are profiled by sets of translation probabilities, e.g., p(foto|u,photo),…, p(image|u,photo) p(netz|u,web),…, p(internet|u,web) – we adopt distributional divergence to measure (dis)similarity between users JS-divergence, L1-norm, such as in [Lee 1997]

14. Distributional Divergence between Users sim (photo) (u,u q ) sim (web) (u,u q ) … sim(u,u q )

15. Remark on the 3 Scenarios This framework is able to address all three scenarios – addresses scenario 1 by allowing self-translation, e.g., p(photo|u,photo) – addresses scenario 2 by allowing self-similarity, e.g., sim(u q,u q ) – addresses scenario 3 by enabling borrowed translations

16. EXPERIMENTS 1.Data Collection 2.Experimental Setup 3.Recommendation Performance

17. Dataset from BibSonomy trainvalidationtest time framestart ~ DEC-08JAN 09 ~ JUL 09JUL 09 ~ DEC 09 |R|22,389667258 |U|1,18513657 |T|13,276862525 |A|253,6152,6041,262 |P|64,120775279 average posts per user53.6955.6994.895 average tag tokens per user3.9553.3604.523 average distinct tags per user61.83313.19114.667 Note: time order: train validation test users in test set must have been appeared in validation set.

18. Experimental Setup Methods to compare – trans-n1, trans-n2 k: {5,10,20,50,100,200,300,400,5 00} js-divergence, l1-norm : {1,2,4,8} for js-divergence : {1,2,4,8,12,16} for l1-norm – trans-u1, trans-u2 – knn-ur, knn-ut k: {5,10,20,50,100,200,300,400,5 00} – interpolating with freq-r Evaluation metric – pr-curve at top 5 – macro-average for users Parameter optimization – macro-average f1@5 – global vs. individual settings

19. Recommendation Performance Global Setting

20. Recommendation Performance Individual Setting

21. Recommendation Case Study userresourcetags assignedtop 5 recommendations trans-u1trans-n1 920a45…57f2008, bookmarking, folksonomy, social, spam, folksonomies, tagorapub, web20, 20, integpub, systems, tagger, web diplomathesis captcha folksonomybackground closelyrelated folksonomy tagging social web20 web 1119d16…b50it, news, technology, blog, feed, technologie kultur online radio kunst cd news web20 blog software technology 3217467…655annotation, ontology, knowledge, semantic sql erd eclipse tagging folksonomy ontology web20 semantic scenario 3 tags

22. Conclusion We propose a probabilistic framework for solving the personalized tag recommendation task, which incorporate personomy translation and borrowing translation from neighbors. We devise to use distributional divergence to measure similarity between users. Users are similar if they exhibit similar translation behavior. We find the proposed methods give superior performance than translation by the query user only and classic collaborative filtering.