关键词: 协同过滤, 相似度可信值, 动态K-means, 双向兴趣, 社交网络

Abstract: The rapidly development of SNS makes the dramatic increasement in the number of users. To deal with the difficulty for users to find potential friends quickly, many personalized recommendation technologies emerge. Among them collaborative filtering algorithm is most popular. However, collaborative filtering algorithm also has some drawbacks, like sparsity and low accuracy. In this paper, a bidirectional collaborative filtering algorithm combined with dynamic K-means clustering algorithm is proposed. It applies dynamic K-means clustering algorithm to cluster the similar users into the same cluster, and proposes the concept of similarity credible value to improve the accuracy of similarity. Then it exploits the improved similarity method to compute the neighbors respectively from the taste aspect and attract aspect. It generates the recommendation list based on both the suggestion from taste and attract neighbors. Experimental result demonstrates that this algorithm turns out to be superior to traditional CF algorithm in accuracy of recommendation.

Key words: collaborative filtering, similarity credible value, dynamic K-means, bidirectional interest, social networking