Hybrid collaborative filtering model for improved recommendation

Abstract

Collaborative filtering (CF) based recommendation system, which can automatically predict unknown preference of a user to certain products and then generate meaningful recommendations using a explicit known ratings matrix, has become one of the most successful approaches in web-based activities such as e-commerce. As users will typically not bother to rate items they bought, data sparsity is one main challenge for CF task. Item-oriented CF algorithm and user-oriented CF algorithm are two state of the art techniques for recommendation system. However, the utilization of singe item similarity matrix or single user similarity matrix always results in poor prediction accuracy because of sparse data. In this paper, a new hybrid collaborative filtering model is proposed by combining item-based CF algorithm and user-based CF algorithm. Both item similarity matrix and user similarity matrix are considered in this hybrid CF model, which is more robust to sparse problem. Experimental results on MovieLens data set show the superiority of our approach over current state of the art methods.