Abstract:  Collaborative filtering algorithm, which can recommend the items appeal to users from mass of data through studying the user’s preferences is widely used in electronic commerce. However, collaborative filtering algorithm suffers from decreasing accuracy and inefficiency in scalability, data sparsity, and cold start. In order to solve there problems, the concept of user attribute similarity is introduced in this paper, and the user can be divided into appropriate user clusters to predict the user’s ratings for a project by using K-means clustering algorithm. Furthermore, through fusing the recommendation algorithm based on user attributes and the collaborative filtering algorithms based on the project by using the method of mixed weights, a collaborative filtering algorithm synthesizing the attributes of user is proposed. Through experiment by using MovieLens data sets, we verify that the proposed algorithm has extensibility. Simultaneously, it can ease cold start problem and improve the prediction accuracy of recommendation algorithm in some degree.

Key words: collaborative filtering, K-means clustering, user attributes, cold start