Research On Sparse Matrix Filling Algorithm For Recommendation System

The source of recommendation system has diversity and particularity.How to efficiently mine user preference information has always been the focus of recommendation system research.However,large-scale increase of users and project information will inevitably bring about the cold start system,the sparseness and user interest offset and other issues.Therefore,filling the scoring matrix by making full use of information such as user,project and user-item scoring matrix can effectively improve the problems of cold start and sparsity of recommendation system.In this thesis,we improve the problems of cold start-up,user-item score matrix high missing value and so on,and put forward the different source clustering integration algorithm.In addition to using the user-item scoring information,we also make full use of user and project information,A sparse matrix filling algorithm for recommendation system is proposed.First,we use different source clustering integration algorithms to integrate the results of user,project and scoring information clustering to get the improved user and project neighbor set.Secondly,when the similarity of users(items)before and after guaranteeing matrix decomposition is consistent,this thesis improves the objective function of the classical regularization matrix decomposition model,and adds the first k nearest neighbor sets of user(item)as the optimization constraint to the objective function in.This method makes it possible to reduce the cold start and sparsity by using the user information and the project information to calculate the similarity of users and projects under the condition of meeting new users,new projects and scoring matrix sparse.Finally,the gradient descent method is used to optimize the solution,and the factor matrices P and Q are solved,and the scoring matrix is filled with the product of the two as the predictive score.In this algorithm,the fusion of users,the matrix decomposition of the project neighborhood model and the improvement of the integration algorithm of different source clustering,theoretically alleviate the problems of cold start and sparsity,and to some extent improve the accuracy of the score matrix filling.In this thesis,experiments are performed on the classic MovieLens dataset and the ecommerce real data set,and the proposed k-CE-MF algorithm is compared with other matrix filling results based on the matrix decomposition model recommendation algorithm.The experimental results show that the proposed sparse matrix filling algorithm for recommendation system can more accurately fill the user-item scoring matrix,and can better alleviate the problem of cold start and sparsity.In addition,in the more abundant and comprehensive user,project information data Set with better results.