Research On Bipartite Network-based Projection Algorithm In The Application Of Recommendation System

With the promotion of the Internet, the personalized recommendation getunprecedented development and application. Due to recommend shooting high and not berestricted by items type, the bipartite network-based projection recommendation algorithmcapture Widespread attention. Around the bipartite network-based projectionrecommendation algorithm, after further study of its characteristic, do the followingimprovement.First of all, in order to improve the resource allocation recommendation algorithm toodependent on the role of node’s degree, we propose a user preferences-based non-uniformdistribution recommendation algorithm. The proposed algorithm put the ratings which beregarded as user preferences in recommendation algorithm linear fusion to the resourceallocation recommendation algorithm, and according to the difference of user’s rating orthe similarity between users to adjust the distribution coefficient. Finally, the object whichbe rated highly by the user who get the most resources has priority to be incorporated intothe recommendation list.Secondly, for the new user problem in the bipartite network-based projectionrecommendation algorithm, we propose a random walk in trust network recommendationalgorithm, which coalesce the recommendation algorithm based on trust and the heatconduction recommendation algorithm. At the same time this algorithm add the bipartitenetwork-based structure dimensionality reduction and the concept of maximum entropy、six degrees of space. Algorithm divide the trust values into direct trust value and globaltrust value by random walk, finally, through the global trust value for the target userrating predicts.Thirdly, through the experiments on dataset show that the improved bipartiterecommendation algorithm respectively improves the Mean Average Precison, MeanReciprocal Rank and normalize Discounted Cumulative Gain compared to otheralgorithms. That is to say, this algorithm can make the recommendation list hit more andthe hit objects rank in the front of the list. highlights the research significance and application value of the improved recommendation algorithm.