The Research Of Recommendation Algorithm Basedon Bipartite Network And Its Application On Spark Platform

In recent years,the Internet has grown vigorously.With the continuous popularization of social information,the global data has grown at an alarming rate every year.The overloading of information makes it difficult for people to find information that they really need in a large amount of data.The birth of the recommendation system provides users with personalized informa tion service or content,allowing users to get reasonable recommendations under the premise of uncertainrequirements.The bipartite network recommendation algorithm is a recommendation algorithm that only considers the connection between the users and the objects,and derives from the energy propagatio n theory in dynamic physics.There are two of the most representativebipartite network recommendation algorithm.O ne is material diffusion algorithm,which is the biased recommendation of popular items,with high recommended accuracy.And the other isheat conduction algorithm which is bias to the minority recomme nded items with a higher diversity.However,the material diffusion algorithm has the problem of low diversity,while the heat conduction algorithm has the shortcomings of low accuracy.After analyzing the advantages and disadvantages of the two most representative bipartite graphsalgorithm,the research of this paper is to improve the recommendation accuracy of material diffusion,and to further enhance the diversity of diffusion algorithm based on bipartite network.The main research work is as follows:1)By combining the Jaccard coefficient with the one-way diffusion resource transfer formula by using the idea of similarity fusion,the JacN BI algorithm is obtained.This algorithm overcomes the problem that the calculation of the resource transfer matrix in material diffusion fails to consider the problem of decreasing accuracy due to the number of objects.The experimental results show that Jac N BI algorithm has improved the accuracy and recall rate at the same time,and the diversity of Hamming distance is also improved.2)In this paper,the non-directional diffusion algorithm is used to calculate the resource transfer matrix between the items.The causal relationship leads to the deviation and irrationality of the actual results.The formula of the resource transfer matrix of the JacN BI algorithm was incorporated into the bidirectional diffused computing model,and the new bidirectional diffusion algorithm JacC BI was obtained.The results show that the JacC BI algorithm is improved on the basis of JacN BI,which not only improves the accuracy and recall rate,but also enhances the diversity of the JacC BI algorithm.3)Finally,the paperpresents the application of the improved two part network recommendation algorithm on the Spark platform and the prototype of the movie recommender system,which provides a valuable reference for the actual recommendation system development.