Research On Collaborative Filtering Algorithm Based On Community Division

Collaborative filtering technology is the most successful in personalized recommendation by far. It is mainly based on user-item ratings to calculate the similarity between two users and predict the rates of users by searching the neighbors. As the widly used application, collaborative filtering recommendation can recommend for individuals initiatively and recommended range throughout many fileds. But there are still some problems. The similarity methods of traditional recommendation is too simple to obtain the accurate results. With the increasing number of users and items, large-scale data lead to serious sparse data and reduce the performance of recommendation.In order to improve accuracy of Collaborative filtering recommendation, the paper propose an hybrid similarity function based on item factor and time factor by analyzing indequate traditional similarity methods. In hybrid similarity functions we consider that depending on the different items, user's neighbors should be different and the user-item ratings will change over time. Finally, experimental results represent that the hybrid similarity function can improve the performance of recommendation system.This paper introduces the topological potential theory, and use a new overlapping community detection algorithm based on node location analysis, which divided the node into four kinds: peaks, valleys, slopes position and edge position, and put the users with similar interests into the same community depend on the position of the nodes, and in combination with collaborative filtering recommendation algorithm to calculate the prediction rating on the basis of the user's neighbor set searched from their community. Experimental results show that the method effectively reduces the impact on the prediction accuracy of the sparse matrix, and improves the ability to predict.