关键词: 个性化推荐系统, 协同过滤, 基于项目的用户相似性, 时间权值

Abstract: Collaborative filtering is the most widely used and the most successful technology in the personalized recommendation system so far.However，existing collaborative filtering algorithms have taken the user’s interests in different time into equal consideration，which leads to the lack of effectiveness in the given period of time.At the same time，they have been suffering from low recommendation accuracy.Based on two crucial steps：Computing the user’s nearest neighbor and predicting item ratings，this paper proposes an improved collaborative filtering algorithm，which computes user similarity based on the relation between items and adds time weight for computing item ratings，to get more appropriate neighbors and make the click interests approaching the gathering time have bigger weight in recommendation process.Experimental results show that the improved algorithm can provide up to date recommendations and give better prediction in accuracy.

Key words: personalized recommendation system, collaborative filtering, user similarity based on item, time weight