Abstract: Collaborative filtering is the most widely used recommendation technology in the personalized recommendation system. However, the rapid increase of the amount of users and data make the score matrix of user's preference information become more and more sparse, and the collaborative filtering algorithm encounters a bottleneck. The calculation of similarity is the most important step in collaborative filtering algorithm. In order to improve the accuracy of the similarity of sparse matrix, this paper improves the traditional collaborative filtering algorithm. We cluster the item set first, and then use the Slope One algorithm to fill matrix after clustering, finally introduce the degree of preference of each cluster for user as the weight. The experimental results show that the improved collaborative filtering algorithm can effectively alleviate the sparse problem of scoring matrix, so as to improve the quality of the recommendation system.

Key words: collaborative filtering, sparse matrix, similarity, Slope One algorithm