Research On Bipartite Network-Basedprojection Algorithm In The Collaborative Filtering Recommendation

The rapid development of Internet technology provides users with a range of optional information to meet the information needs of users.Personalized recommendation systems based on the historical interest of the user preferences are to help users select items that best meet the required information from the mass.The recommendation algorithm based on networks has been attracting more and more researchers’ attention.because of its high hit rate and no restrictions on the type of project properties.But it depends only on the degree,does not distinguish the preference of the user about the object,recommended limited.which limits the recommendation.Firstly, the bipartite graph network structure considers user’s common neighbors rather than the scores that users rate the items.And these algorithm tend to recommend popular objects,without considering the influence of objects degree,user degree and the weights of the object.To solve these problems,this paper presents a new perspective on characterizing the similarity between users,more generally,nodes of a weighted graph.We proposed an improved recommendation algorithm based on weighted networks. Adjustment of the popular items can make the results more accurate and diverse.To solve recommendation problem for cold-start users,this paper proposes an algorithm that combines the trust propagation based recommendation with the bipartite networks-base recommendation.By recommending a user the latent friends he is interested in,it may help widen her circle of friends,and enhance accuracy accordingly.The algorithm integrates users’ trust and similar users in order to find the nearest neighbors of the target user,which the algorithm uses to compute the weight of neighbors and to form item recommendation.Last,Experimental results on the Movie Lens database show that the improved bipartite recommendation algorithm can perform well in comparison with other methods on accuracy and diversity,while reducing the epidemic of the recommended objects.The result highlights the research significance and application value of the improved recommendation algorithm.