Tweet Recommendation with Graph Co-Ranking


Published on

Published in: Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Tweet Recommendation with Graph Co-Ranking

  1. 1. Tweet Recommendation with Graph Co-Ranking Rui Yan, Mirella Lapata, Xiaoming Li ACL 2012 Reader: 東京大学 相澤研究室 藤沼祥成
  2. 2. Motivation• 3 problems related to tweet recommendation – Linkage of following and retweeting – Interest the user – Personalization and diversity
  3. 3. Related Work• Collaborative Filtering [Hannon et al. 2010]• Selecting tweets including URLs [Chen et al. 2010] – And so on…• Co-Ranking Framework: Scientific impact and modeling the relationship between authors and their publications [Zhou et al., 2007].
  4. 4. What is Proposed in this Paper• Adapting Co-Ranking framework to Tweet recommendation• Including personalization
  5. 5. Graphs Tweet-Author GraphTweet Graph Author Graph
  6. 6. Co-Ranking Algorithm• Simultaneously rank tweets and their authors – a tweet is important if it associates to other important tweets – A user is important if the associate to other important users, and they write important tweets
  7. 7. Components of Co-Ranking• Popularity (PageRank [Brin and Page 1998])• Personalization (PersRank) – Modifying PageRank• Diversity (DivRank [Mei et al. 2010]) – Avoid assigning only high scores to closely connected nodes – Popular nodes get popular
  8. 8. Popularity: PageRank• (1-μ): stick to the random walk• μ: Jump to any vertex chosen uniformly at random• m: ranking scores of for the vertices in Tweet graph
  9. 9. Personalization (1/2)• Used Latent Dirichlet Allocation to construct the matrix D• Dij: Probabilitiy of tweet mi belongs to topic tj• Image of D Tweets 𝐷11 ⋯ 𝐷1𝑛 Topics ⋮ ⋱ ⋮ 𝐷 𝑚1 ⋯ 𝐷 𝑚𝑛
  10. 10. Personalization (2/2)• r: ri = the probability for a user to respond to tweet mi• Estimate t: topic interest vector by maximum likelihood
  11. 11. Diversity: DivRank• Transition probabilities change over time• Favors popular nodes as time goes by• After z iterations, M is
  12. 12. CoRank: Figure
  13. 13. Actual Steps• Step 1 Walk from the author• Step 2 Walk from the tweet Ensuring convergence
  14. 14. Co-Ranking Algorithm• Coupling parameter λ• If λ=0, no coupling between Tweet graph and Author graph• In experiment, λ = 0.6
  15. 15. Transition Matrix in Author Graph• It is defined as
  16. 16. Transition Matrix in Tweet Graph• Tweet Graph is defined as•• mi a term vector is weighted as tf・idf
  17. 17. Transition Matrix in Tweet- Author Graph• MU:• UM:• : tweet mi is authored by uj
  18. 18. Data Set• 9,449,542 users – Tracing the edges of 23 users’ followers and followees until no new user is added• 3/25/2011 to 5/30/2011• 364,287,744 tweets
  19. 19. Evaluation• Automatically – Golden: A tweet is retweeted or not• Human-based Judgement – 23 users – Whether they will retweet or not – Calculating the mean
  20. 20. Baselines• Randomly ranked (Random)• Longer tweets ranked higher (Length)• Many retweets ranked higher (RTnum)• RankSVM algorithm (RSVM) [Duan et al. 2010]• Decision Tree Classifier (DTC) [Uysal and Croft 2011]• Weighted Linear Combination (WLC) [Huang et al. 2011]
  21. 21. Criteria• Normalized Discounted Cumulative Gain• Mean Average Precision
  22. 22. Normalized Discounted Cumulative Gain• Highly relevant documents are more valuable• The lower the ranked position of the relevant document is, the less valuable it is for the user Normalized parameter Gradually reduces the obtained from ideal document score ranking
  23. 23. Normalized Discounted Cumulative Gain Rank Tweet 1 A 2 B 3 C AとFが共にリツイートさ れている時、Fが低くラ 4 D ンク付けされている為、 5 E Fにペナルティを付ける 6 F Normalized parameter Gradually reduces the obtained from ideal document score ranking
  24. 24. Mean Average Precision• Average of the precision of top k documents Precision at ith tweet Number of reposted Retweeted or not tweets
  25. 25. Mean Average Precision Rank Tweet 1 A If F is retweeted, 2 B precision increases. If not, precision 3 C decreases 4 D 5 E 6 F Number of reposted Retweeted or not tweets
  26. 26. Up to top ranked 5 Results tweets• Automatic Evaluation• Manual Evaluation
  27. 27. Evaluation of Components• Automatic Evaluation• Manual Evaluation
  28. 28. Conclusion• Relatively improved 18.3% in DCG and 7.8% in MAP over the best baseline• Improved due to using the tweets and their authors• Succeeded to recommend interesting information that lies outside the user’s followers• Future: Include credibility and recency