关键词: 多相似度，协同过滤推荐算法，User-based, MAE

Abstract: Conventional user-based collaborative filtering algorithm measures the similarity of two user's favor of any types of items through the single rating similarity. However, daily experience tells us that people usually have different degree on their favor of different types of objects,and obviously the single rating similarity cannot accurately describethis difference. Aiming at this problem, we deeply analyzed the characteristic of user-based collaborative filtering recommendation algorithm, and proposed a collaborative filtering recommendation algorithm based on user's multi-similarity,which describes different similarity of two user's favor of different types of items through the computation of their multiple independent rating similarity for different types of items. The experimental results show that the proposed algorithm, which computes predicted ratings of unrated items on the basis of user's multi-similarity, can effectively improve the accuracy of predicted ratings and enhance the quality of recommendation.

Key words: Multi-similarity, Collaborative filtering recommendation algorithm, User-based, MAE