1 / 21

Scientific Paper Recommendation Emphasizing Each Researcher’s Most Recent Research Topic

Scientific Paper Recommendation Emphasizing Each Researcher’s Most Recent Research Topic. Kazunari Sugiyama 8 th January, 2010. Introduction. The number of published scientific papers continues to grow. Users of digital library suffer from finding papers relevant to their information needs.

alina
Download Presentation

Scientific Paper Recommendation Emphasizing Each Researcher’s Most Recent Research Topic

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Scientific Paper Recommendation Emphasizing Each Researcher’s Most Recent Research Topic Kazunari Sugiyama 8th January, 2010

  2. Introduction • The number of published scientific papers continues to grow. • Users of digital library suffer from finding papers relevant to their information needs. • Recommendation systems are promising approach to address each user’s interest. • Mid-level or senior researchers • Several different research interests based on several years experience • Junior researchers • Quite small publication list(too short to construct user profile)

  3. Related Work • Improvement in Ranking of Digital Library • ISI impact factor (ISI IF) • Papers with high impact and low impact are treated equally. • Its ranking are biased towards popularity. • Improved approach • “Focused PageRank” [Sun and Giles, ECIR’07] • “FutureRank” [Sayyadi and Getoor, SIAM-Data Mining, ‘09] • Weighted PageRank, Y-factor (product of ISI IF and weighted PageRank) [Bollen et al., Journal of Scientometrics ‘06] • “Scientific gems” [Chen et al., Journal of Informetrics ‘07]

  4. Related Work • Recommendation Systems in Digital Library • Recommend citations [McNee et al., CSCW’02] • Recommend papers by combining collaborative filtering and content-based filtering [Torres et al., JCDL’04] • Recommend paper s ranking-oriented collaborative filtering [Yang et al., JCDL’09]

  5. Related Work • Construction of Robust User Profile in Recommendation Systems • Content-based approach • Frequent patterns obtained by click-history [Kim et al., ICADL’08] • News recommender system [Das et al., WWW’07], [Chu and Park, WWW’09] • Long-term search history [Shen et al., SIGIR’05], [Tan et al., KDD’06], [White et al., SIGIR’09]

  6. Proposed Method • System Overview • Construction of User Profile • Junior Researchers • Mid-level or Senior Researchers • Construction of Feature Vectors for Candidate Papers to Recommend

  7. System Overview (2) Compute similarity between (1) Construct user profile from each researcher’s past papers and Candidate papers to recommend to Researcher (3) Recommend papers with high similarity

  8. Junior Researchers’ Published Papers [No published papers in the past] (‘09) Relation between reference papers and References (‘09) (‘02) (‘07) (‘06)

  9. Weighting Schemes for Junior Researchers’ Published Papers • Linear Combination (LC) • Similarity between the most recent paper and others (SIM) • Reciprocal of the difference between published year of the most recent paper and that of other papers (RPY)

  10. Mid-level or senior researchers’ published papers new old (‘05) (‘09) (‘02) (‘03) Relation between citation or reference papers and (‘06) (‘07) (‘09) References (‘05) (‘01) (‘04) (‘03)

  11. Weighting Schemes for Mid-level or Senior Researchers’ Published Papers • Linear Combination (LC) • Similarity between the most recent paper and others (SIM) • Reciprocal of the difference between published year of the most recent paper and that of other papers (RPY) • Forgetting factor (FF)

  12. System Overview (2) Compute similarity between (1) Construct user profile from each researcher’s past papers and Candidate papers to recommend to Researcher TF-IDF (3) Recommend papers with high similarity

  13. Experiment • Experimental Data • DBLP papers for each researcher • ACL Anthology • 597 papers published in 2000 - 2006

  14. Experiment • Evaluation Measure • Normalized Discounted Cumulative Gain (NDCG) • NDCG@5, NDCG@10 • Mean Reciprocal Rank (MRR)

  15. Experimental Results • Junior Researchers • Mid-level or Senior Researchers

  16. Recommendation Accuracy for Junior Researchers

  17. [NDCG@5] [NDCG@10] * * [MRR] * * : statistically significant for p < 0.05

  18. Recommendation Accuracy for Mid-level or Senior Researchers

  19. [NDCG@5] [NDCG@10] * * * [MRR] * ** : statistically significant for p < 0.01 * : statistically significant for p < 0.05

  20. [NDCG@5] [NDCG@10] + + * [MRR] + + : statistically significant for p < 0.05

  21. Conclusion • Recommendation system of scientific papers for junior researchers, and mid-level or senior researchers • Junior researcher • User profile constructed using the most recent paper and its pruned reference paper gives the best recommendation accuracy. • Threshold of pruning: 0.2 • NDCG@5: 0.521, NDCG@10: 0.459, MRR: 0.624 • Mid-level or senior researcher • User profile constructed using papers published within 3 years and its pruned citation and reference papers gives the best recommendation accuracy. • Threshold of pruning : 0.4 • NDCG@5: 0.540, NDCG@10: 0.518, MRR: 0.812

More Related