A collaborative filtering recommendation algorithm based on user clustering with preference types

Abstract

The traditional collaborative filtering algorithm is extensively applied in the field of personalized recommendation, but it still faces the challenges of data sparsity and scalability problems which result in low quality and efficiency of recommendation. To address the problems, a hybrid collaborative filtering recommendation algorithm is proposed based on user preference type clustering. First, by analyzing the relationship between users and item categories, we construct the user item category preference matrix. On this basis, user clustering is carried out and users with similar preference types are clustered into the same user groups. Then, to search for the nearest neighbors of target user, similarity of users is calculated in the cluster it belongs to by considering both the user ratings and their preferred item categories. Finally, an improved Slope One algorithm is proposed and applied to the nearest neighbor set to predict item ratings and then generate recommendations. The experiments indicate the proposed hybrid collaborative filtering algorithm can improve recommendation performance in the case of a great many users.