Research On Collaborative Recommendation Based On Bipartite Network

Collaborative recommendation technology is one of the most effective methods to deal with the problem of information overload and it has been greatly concerned by numerous researchers. However, with the development of new network technologies, collaborative recommendation appears dynamic complexity and its key issues are still prominent, such as scalability, data scarcity, dynamics and so on. In recent years, the research on complex network has developed and many new theoretical models and methods have been proposed, which provides new ideas and methods to solve above key issues.The bipartite network is an important manifestation of complex network, it could describe topological structures of collaborative recommendation systems. Therefore, we deeply studied related contents on bipartite network and collaborative recommendation, and then proposed a framework of collaborative recommendation based on the bipartite network. After clarifying feasibilities and advantages of this framework, we defined its main goals and key steps in view of solving those key problems of collaborative recommendation.The main researches of this paper are as follows:Firstly, we proposed a projection method based on the user-resource topological structure. This method could accurately extract the relationships between users or resources. Utilizing these relationships, we implemented the user-based and resource-based collaborative recommendation algorithm respectively, which could effectively improve the recommendation accuracy and diversity simultaneously.Then, we proposed a community division algorithm based on propagation, which could naturally obtain overlapping and hierarchical node communities after obtaining link communities. Furthermore, this method could be executed synchronously and updated dynamically. By introducing a parameter, this method could achieve multi-resolutions and make the original method more flexible. Finally, take advantage of the above methods, we proposed a community-based collaborative recommendation algorithm. This algorithm could indeed achieve dynamic neighborhoods selecting. It uses communities as the neighborhoods and adjusts their correlations on the basis of communities memberships and corresponding relationships so that the recommendation could be more flexible and effective.We chose two data sets:Southern women and MoiveLens, and made numerical experiments. The experiment results verify that our methods is feasible and effective. And we also put forward the problems to be studied further.