UserRec: A User Recommendation Framework in Social Tagging Systems

Social tagging systems have emerged as an effective way for users to annotate and share objects on the Web. However, with the growth of social tagging systems, users are easily overwhelmed by the large amount of data and it is very difficult for users to dig out information that he/she is interested in. Though the tagging system has provided interest-based social network features to enable the user to keep track of other users' tagging activities, there is still no automatic and effective way for the user to discover other users with common interests. In this paper, we propose a User Recommendation (UserRec) framework for user interest modeling and interest-based user recommendation, aiming to boost information sharing among users with similar interests. Our work brings three major contributions to the research community: (1) we propose a tag-graph based community detection method to model the users' personal interests, which are further represented by discrete topic distributions; (2) the similarity values between users' topic distributions are measured by Kullback-Leibler divergence (KL-divergence), and the similarity values are further used to perform interest-based user recommendation; and (3) by analyzing users' roles in a tagging system, we find users' roles in a tagging system are similar to Web pages in the Internet. Experiments on tagging dataset of Web pages (Yahoo!~Delicious) show that UserRec outperforms other state-of-the-art recommender system approaches.