In collaborative filtering, the correlations among ratings explicitly or implicitly provided by users are used to provide recommendations. Systems of this kind are in widespread use in recommending books, movies, songs, web pages, etc. In his presentation, Patrik Hoeyr from HIIT stated that this technology has yet to be successfully applied to the scientific research literature. Motivation for recommender systems for science include (1) keeping watch and recommending potentially interesting papers (based on topic), and (2) having found a paper, checking whether it is worth reading. An online 'journal club' of expert discussions would be one solution.
Current systems include search engines for academic search (Google scholar, CiteSeer), bibliography management and sharing (BibSonomy, citeulike, Mendeley, Connotea, RefWorks, zotero), and archives (arXiv.org, PubMed). Other systems include whattosee, journalreview.org, getcited, techlens, pubzone, wepapers, and action science explorer.
Problems include that (1) there is not much commercial potential, (2) one has to draw users to a collaborative filtering system, given that there are few other users in the beginning, and (3) privacy issues have to be balanced with authority and trustworthiness. As a solution to the cold start problem, users have to be provided with some other service and make the use very easy. As an experimental efforts, Hoyer presented his project grocko.org. The system provides
- automated topic analysis of abstracts,
- personalized recommendations of newly appeared articles,
- trivial set-up, and
- anonymous discussions and ratings of existing paper
No comments:
Post a Comment