Research On Collaborative Filtering Model Based On Clustering Matrix Approximation

Recommendation system has made great contributions to helping find information what people like most.However,in the user-product rating matrix,there are problems of data sparsity and scalability,which pose great challenges to the accuracy and computational efficiency of the recommendation system.Collaborative filtering recommendation method based on clustering matrix approximation of explicit infornation considers the local characteristics of users' interest in items by clustering users and items,at the same time,the internal structure of the rating matrix is found and dense matrix blocks are formed.Using the real rating data of users'ratings,not the filling data,reduces the interference of noise data and improves the quality of recommendation.Recommendation within dense matrix blocks reduces the overall data input of the recommendation system and improves the computational efficiency.Collaborative filtering recommendation method based on clustering matrix approximation of implicit information treats the users' rating behavior differently,regarding the purchasing behavior as the users'strong preference for the items.Based on the processed data,it searches for the most similar user-commodity dense matrix block.In the dense matrix blocks,the preference labeling data is restored to the original user's rating data,and carry out matrix approximation.The processing method of implicit information makes use of the potential information in the user's purchasing behavior,which contains the user information that can,t be reflected by explieit information.It is likely that the information are't aware of the users themselves,so their recommendation effect is different.Finally,this paper carries out experiments on the real Movielens data set,and uses RMSE evaluation index to evaluate the recommendation system.Recommendation experiments show that the collaborative filtering recommendation method based on clustering matrix approximation can effectively improve the accuracy of collaborative filtering recommendation and reduce the computational complexity.