Making Item Predictions through Tag Recommendations

As opposed to the search engine, social tagging can be considered an alternative technique tapping into the wisdom of the crowd for organizing and discovering information on the Web. Effective tagbased recommendation of information items is a critical aspect of this social information discovery mechanism. While most existing work in the tagging domain makes item recommendations directly after constructing or learning the user profiles, items are not particularly recommendable indeed due to the limiting descriptive ability of the binary values they were assigned on interacting with users. In response to this problem, we propose to recommend the more recommendable tags, which have numerical interactions with users, to refine users’ tag preference first, and then deliver quality item recommendations based on the global relationship between tags and items. Experiments on three realworld social tagging datasets demonstrate the effectiveness of our approach.