Recommendation Algorithm Research Based On Clustering In Bipartite Graph

With the development of information technology and the increase in scale of the Internet,the problem of information overload has become increasingly prominent. Portals and searchengines appear to some extent alleviated the problem, but the information returned by thesetools is popular and unable to meet the individual needs of users. In this context, apersonalized recommendation system came into being. Like search engines, recommendationsystem is also a kind of information filtering tool. It can do personalized recommendationsbased on the user’s background information and specific behaviors. Different users get therecommended results are associated with their own background and behavior, so as toeffectively meet the needs of individual users.Now the recommendation system has been widely used in various fields, the most widelyused one is collaborative filtering recommendation system. But there still have the sparse dataproblem,"cold start" problem and the scalability problem. In recent years, some scholars putclustering algorithm into recommender systems, through clustering can effectively reduce thescale of the data, improve the algorithm scalability and recommendation accuracy. But it justsimple combination of clustering algorithm and collaborative filtering algorithms, do not digdeeper into relationship between clustering and recommendation.Aiming at the shortcomings of traditional clustering based recommendation algorithm,this paper mainly does the following work:(1) This paper combines the bipartite graph mapping and user’s rating information,proposes a weighted mapping method. Use the projection method to compress the user-itembipartite graph, reduce the loss of information in the mapping process and improve thecalculation formula of the similarity matrix.(2) Using clustering algorithm to do clustering on the item, dig out the inner similar itemsand reduce the item space. The traditional item-based clustering algorithm staring from theperspective of the item and select the item related classes to reduce the item space. This article starting from the perspective of the user, narrow space and reduce user’s interest drift, so as toimproving the accuracy.(3) This paper presents a new method of calculation recommended weights, which takefull account of the user’s preference information of interest, dig the interest of the users of theclass and the weight of items in the class.