Learning to recommend with social relation ensemble

Recommender systems with social networks (RSSN) have been well studied in recent works. However, these methods ignore the relationships among items, which may affect the quality of recommendations. Motivated by the observation that related items often have similar ratings, we propose a framework integrating items' relations, users' social graph and user-item rating matrix for recommendation. Experimental results show that our approach performs better than the state-of-art algorithm and the method with only users' social graph ensemble in terms of MAP and RMSE.