关键词: 协同过滤算法, 用户相似性, KL散度, 共同评分信息, 数据稀疏

Abstract: User similarity based collaborative filtering algorithm is one of most widely used technologies. Most of user similarity algorithms only consider the co-rated items between two users, but ignore other ratings that probably hide valuable information. To evaluate user similarity accurately, a user similarity collaborative filtering algorithm based on Kullback-Leibles (KL) divergence was proposed. The proposed algorithm utilizes both the co-rated items and the influence of other no co-rated items. Since the algorithm makes full use of all rating information, it improves the accuracy and reliability of user similarity. Experiments show that the proposed algorithm outperforms other user similarities. Moreover, it can still measure the user similarity effectively, even if no co-rated items exist. Therefore, the presented algorithm solves the problem of full dependence on co-rated items and gains better flexibility.

Key words: collaborative filtering algorithm, user similarity, Kullback-Leibles divergence, co-rated information, data sparseness