Social network based applications such as Facebook, Mys-pace and Linkedln have become very popular among Internet users, and a major research problem is how to use the social network information to better infer users' preferences and make better recommender systems. A common trend is combining the user-item rating matrix and users' social network for recommendations. However, existing solutions add the social network information for a particular user without considering the different characteristics of the products to be recommended and the neighbors involved. This paper proposes a new approach that can adaptively utilize social network information based on the context characterized by products and users. This approach complements several existing social network based recommendation algorithms and thus can be integrated with existing solutions. Experimental results on Epinions data set demonstrate the added value of the proposed solution in two recommendation tasks: rating prediction and top-K recommendations.
展开▼
