Recommending Items via Interest-Similar Cluster Identification in Online Social Networks

Social network has played a more and more important role in recommender system as users are strongly influenced by their friends. But the friend lists of a user are always hidden for privacy concern. Even her friend lists are accessible, there still exist many potential friends who have similar interest with her but she doesn´t know. How to predict the friendship she knows and does not know when the friend lists are hidden? How to identify her interest? To address the above questions, we first propose a Friendship prediction approach mainly based on the analysis of her group information. Secondly, we propose a graph model which simultaneously models user´s friendship and interest. Then we propose an unbiased random walk strategy for item recommendation via the graph model. Finally, we evaluate our approach on large scale real world data from Cite Like and last.fm data set, the results show that the performance of our algorithms is very good in implementation.