Research On Multi-dimensional Recommendation Based On Bipartite Network

With the development of the Internet, electronic commerce gradually integrate intopeople’s lives and brings more convenience for people, but also make people not knowwhat to do in the face of a large number of goods and services. Information overload hasbecome a very serious problem. As one of the most effective technologies to solve theproblem of information overload, recommendation technology has be extended fromsimple screening product for the user to promoting the sale for businesses and enhancingthe products and services sales’ brand image. In the field of theoretical research, it hasattracted the attention of many scholars. The application of the bipartite network structureis further promote the development of the recommendation technology.In order to further investigate the structural characteristics of bipartite network and itsapplication in the recommendation technology, a multi-dimensional recommendationtechnology framework based on bipartite network is proposed on the basis of in-depthresearch of recommendation technology status and bipartite network theory. Thisframework first identifies bipartite overlapping communities, and then on the base of thecommunities, it combines three kinds of recommendation models, and last it gives the finalrecommendation list. The main researches of this paper are as follows:(1) First of all, beyond the weighted user-item bipartite network, the user nodes andthe item nodes are projected into user network and item network respectively, and to gainthe user preference items’ category, the items’ category are projected from item nodes touser nodes. The three kinds of message above make up three basic recommendationmodels. At last, an adaptive method is used to dynamic calculate weight factors of the threerecommendation models, and the last recommendation list is accurate and diverse.(2) For Affinity Propagation clustering results accurate, dealing with large dataquickly, a bipartite overlapping community identification algorithm based on Affinity Propagation is proposed. The algorithm first identify bipartite link communities by theimproved Affinity Propagation clustering method and then get overlapping nodescommunities from the link communities. It can achieve the identification of localcommunity, and get layered, nested, overlapping communities, while reducing thecomplexity of the community division.(3) By analyzing the community structure obtained by the above algorithm, a bipartitecommunity based multi-dimensional recommendation algorithm is proposed. Thealgorithm realizes the bipartite network projection based multi-dimensionalrecommendation algorithm beyond the inner-community, inter-community and overlappingcommunity and effectively improve the accuracy and the diversity of the recommendation.This paper also design experiments for the methods and technology proposed in thispaper. The analysis of the experimental results shows the effectiveness of the proposedmethod.