Skip to content

Latest commit

 

History

History
103 lines (72 loc) · 4.52 KB

bibsonomy-ecml-pkdd-2009.md

File metadata and controls

103 lines (72 loc) · 4.52 KB

Bibsonomy Dataset (2009 ECML PKDD Challenge)

https://www.kde.cs.uni-kassel.de/ws/dc09/

This dataset has tag-assignment information for resources hosted on BibSonomy.

The resources are divided into two parts.

Part 1): Bookmarks

Information on bookmarks made by users on the website. Attributes include:

  • url
  • description
  • extended description
  • date added

Part 2): Articles (bibtex)

Normal bibtex fields, such as:

  • title
  • abstract
  • description
  • authors

Dataset statistics:

  • # Bookmarks (full dataset): 235,328
  • # Articles (full dataset): 143,050
  • # Tags (full dataset): 93,756
  • # Bookmarks (post-core 2): 14,443
  • # Articles (post-core 2): 7,946
  • # Tags (post-core 2): 13,276

Tag Prediction/Recommendation NOT on Post-core 2 (Task 1 of ECML PKDD)

I.e., only for resources,users and tags which have LESS than 2 assignments

F1 (micro-averaged) / Mean-F1 Strategy Reported by Notes
0.18740 Extracted features from resource titles and urls Lipczak et al., 2009
0.18001 Extracted tags from titles, web pages (in the case of urls) and previous tags given to that resource. Then each of these recommendations is weighted and a final prediction is given. Mrosek et al., 2009
0.17975 Extract tag suggestions from the resource texts, from tags previously assigned to that resource and from tags given by that user to other resources, then learns weights to combine these into a final recommendation. Ju and Hwang 2009 Personalized Predictions
0.14151 CF-based Zhang et al 2009
0.1408 Keywords and Association Rules Wang et al. 2009
0.11886 FDT (Feature-Driven Tagging) Similar to Feature Engineering Si et al .2009 Not Personalized

Tag Prediction/Recommendation on Post-core 2 (Task 2 of ECML PKDD)

These are all @5

F1 (micro-averaged) / Mean-F1 Strategy Reported by Notes
0.35594 PITF (Pairwise Interaction Tensor Factorization) Rendle and Schmidt-Thieme 2009 Personalized
~0.35 FM (Factorization Machine) Rendle 2010 Personalized
0.34791 FasTag + Blsc (a method to tune the number of recommended tags) Gueye et al. 2014 Personalized
0.33185 Relational Classification Marinho et al. 2009
0.32461 Content-based Lipczak et al. 2009
0.32230 Content-based Zhang et al. 20009
0.3154 SimRate Zhang et al. 2014 Personalized, Bookmarks only
0.315 FasTag (Similar to User-Based CF) Gueye et al 2014 Personalized
0.308 maxarg(t) of p(t u,i) Gueye et al. 2014
0.3075 Diffusion Rank Si et al. 2009 Personalized
0.305 STRec Gueye et al .2014 Personalized
0.3018 FDT (Feature-Driven Tagging) Similar to Feature Engineering Si et al. 2014 Personalized
0.288 PopRes Rendle and Schmidt-Thieme 2009 Not Personalized
0.285 FolkRank Gueye et al. 2014 Personalized
0.2536 PopRes (Most popular tags by resource) Zhang and Yu 2014 Not personalized, Bookmarks only
0.2382 Cosine on TAS features Zhang and Yu 2014 Personalized, Bookmarks only

Tag Prediction/Recommendation on Post-core 2 (Results on Training set (via CV))

F1 (micro-averaged) / Mean-F1 Strategy Reported by Notes
0.4110 PopRes,PopUser Si et al. 2014 Personalized, bibtex only
0.3959 Content-based kNN Si et al. 2014 Personalized, bibtex only
0.351 PopRes Rendle and Schmidt-Thieme 2009 Not personalized
0.3018 FDT Si et al. 2009 Not personalized, bookmark only
0.2646 Search-based KNN (Mishne 2006) Si et al. 2009 Not-personalized, bookmark only
0.2537 Search-based KNN (Mishne 2006) Si et al. 2009 Not-personalized, bibtex only
0.2478 FDT Si et al. 2014 Not personalized, bibtex only

References

  • Lipczak et al. 2009: Tag Sources for Recommendation in Collaborative Tagging Systems

  • Mrosek et al. 2009: Content- and Graph-based Tag Recommendation: Two Variations

  • Ju and Hwang 2009: A Weighting Scheme for Tag Recommendation in Social Bookmarking Systems

  • Si et al. 2009: Content-based and Graph-based Tag Suggestion

  • Zhang et al. 2009: A Collaborative Filtering Tag Recommendation System based on Graph

  • Gueye et al. 2014: A Social and Popularity-Based Tag Recommender

  • Zhang and Yu 2014: Hybrid Personalized Tag Recommendation Algorithm Design and Evaluation