关键词: 协同过滤, 用户项目评分, 项目属性, pearson 相关系数

Abstract: The algorithm of collaborative filtering is only considered to analyze the user-item evaluation matrix traditionally． The properties of the item or the user are often ignored． In order to solve this problem and improve the accuracy of the algorithm of collaborative filtering recommendation，we apply the attribute categories of the items into the formula for calculating the similarity of items． The specific method is as follows: firstly，get the degree of difference between the item properties by creating the items' attribute categories table．Secondly，apply the degree of difference between the item properties into the pearson correlation formula and calculate the similarity between items． The experiment results show that the recommended MAE of the improved method is smaller and the hit rate is higher compared to the traditional collaborative filtering algorithm．

Key words: collaborative filtering, user program rating data, item properties, pearson correlation coefficient