Collaborative personalized tweet recommendation

612 views

Published on

0 Comments
3 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
612
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
0
Comments
0
Likes
3
Embeds 0
No embeds

No notes for slide

Collaborative personalized tweet recommendation

  1. 1. Collaborative PersonalizedTweet RecommendationKailong Chen, Tianqi Chen, GuoqingZheng, Ou Jin, Enpeng Yao, Yong YuShanghai Jiao Tong Univ.To Appear in SIGIR 2012 WUME Reading Group Liangjie Hong
  2. 2. Outline• Problem• Contribution & Assumptions• Model• Dataset & Experiments
  3. 3. The Problem
  4. 4. Contributions• Topic latent factors to capture users’ interests• Social latent factors to model social relations• Incorporate Explicit features• Build on traditional collaborative ranking approach
  5. 5. Assumptions• Users’ retweeting actions reflect their personal judgement of informativeness and usefulness.• Users who have retweeted similar statuses in the past are likely to retweet similar statuses in the future.
  6. 6. Starting from scratchTo predict the response, the simplest MF model: bias latent factors
  7. 7. Starting from scratchLearning pair-wise preferences
  8. 8. Starting from scratchObjective function
  9. 9. Incorporating FeaturesTopic Decomposition
  10. 10. Incorporating FeaturesSocial Relations
  11. 11. Incorporating FeaturesCombine Topic + Social
  12. 12. Incorporating FeaturesExplicit Features
  13. 13. Incorporating FeaturesExplicit Features
  14. 14. Incorporating FeaturesFeatures• Relation features ▫ Co-follow scores ▫ Mention scores  # of times user u has mentioned publisher p ▫ Friend  binary indicator  u <-> p
  15. 15. Incorporating FeaturesFeatures• Relevance features ▫ Relevance to status history ▫ Relevance to retweet history ▫ Relevance to hash tags
  16. 16. Incorporating FeaturesFeatures• Content-meta features ▫ Length of tweet ▫ Hash tag count ▫ URL count ▫ Retweet count
  17. 17. Incorporating FeaturesFeatures• Publishers’ Authority features ▫ Mention count ▫ Followee count ▫ Follower count ▫ Status count
  18. 18. Parameter Estimation• Stochastic gradient descent• Down-sample negative samples
  19. 19. Dataset• 8059 users in the base with all statuses ▫ Select 1048 users: > 15 followees ▫ one (+) vs. four (-) ▫ 490 tweets on average per user ▫ 409680 (4/5) in training ▫ 102457 (1/5) for testing
  20. 20. Dataset
  21. 21. Evaluation metrics• Redefined MAP
  22. 22. Method Comparison• Chronological• Retweeted Times• Profiling• LDA• RankSVM• JointMF ▫ [Yang et al, WWW 2011]• CTR ▫ This paper.
  23. 23. Results
  24. 24. Results
  25. 25. Results
  26. 26. Conclusion• Model: a modified feature based MF model• Novelty ▫ The task ▫ Topic decomposition ▫ Feature integration

×