关键词: 电子商务，推荐系统，协同过滤，全面均值，个性化

Abstract: The users’similarity computation is a key step of Collaborative Filter (CF) algorithm. The All-Average method and classified recommendation improved algorithm based on probabilistic analysis of users' discrete explicit rating vector were proposed to solve the problem of CF sparsity and other practical problems. Experimental result shows the improved method enhances the precision and quality of CF prediction.

Key words: Electronic commerce, Recommendation system, Collaborative filter, All-average, Personalized