Research On Personalized Recommendation Based On The Bipartite Network And Community Partition

Personalized recommendation can meet users with different needs and provide users with high-precision personalized service. It can increase the adhesion to the user and bring benefits for the developers of network and the system. But the personalized recommendation is not very satisfactory at the accuracy rate and the recall rate(coverage rate) aspect. This paper aims to improve the effect of personalized recommendations.First of all, it find that the bipartite network can better describe the network and the relationships between the network nodes through detail analyses of complex networks, community division and personalized recommendation. Hence, we select bipartite network as the research object, and make personalize recommendation by using the two type nodes of bipartite network and the relationships between the nodes.Secondly, it clusters the user nodes and item nodes of bipartite network and generates communities by using the improved Fuzzy C-Means(FCM) algorithm. It reduced the sparsity of data and has get a better division results. The improved FCM algorithm performance in the improved Subtractive Clustering for initialization, improved distance function(including 0-1 matrix distance function and weighted matrix distance function) and the best solution selection procedure. Then we made a test on the actual data of a bipartite network use the improved FCM algorithm and get a good division results.Then, we put forward a personalized recommendation model based on bipartite network, and respectively clustering for the user nodes and item nodes in the 0-1 matrix and weighted matrix. Then we make the popularity sort by using the improved Page Rank algorithm in the communities’ nodes after clustering. At last we make personalized recommendations through the popularity sort.Finally, we build a user-movie bipartite network by the Movie Lens data and verified the personalized recommendation model based on the bipartite network. The results were compared with the results of classic collaborative filtering algorithm, and comparative results show that personalized recommendation algorithm based on bipartite graph network has a better effect on the accuracy rate, the recall rate(coverage rate) and F1.