Research On Co-Clustering Based Matrix Factorization Models For Recommendation

The rapid development of Internet has brought massive information to users for their different types of information demands.Meanwhile,it brings a new issue information overload,which causes the difficulties for users to find out the information for their personalized needs quickly.In response to this problem,recommender systems play a pivotal role in alleviating information overload.Collaborative Filtering(CF)is a popular way to build recommender systems,and it has been widely deployed by many e-commerce websites.Collaborative filtering algorithms recommend items by predicting the ratings of the user-item rating matrix.Although the collaborative filtering algorithms have high recommendation accuracy,how to improve the algorithm efficiency has been a key problem following the increasing number of users and items.To this end,in this paper,we propose a matrix factorization model based on co-clustering(MFCC)to address this problem.On the basis of maintaining the high precision of matrix factorization algorithms,the proposed method in this paper can improve the efficiency obviously.Firstly,we adopt a co-clustering algorithm to cluster the user-item rating matrix into several separate sub rating matrices,in which each sub rating matrix has strong relevance within the same group,and each sub rating matrix is irrelevant in different groups.Secondly,we use efficient matrix factorization algorithm to predict ratings.It is worth mentioned that because of the irrelevance of each sub rating matrix,the process of rating prediction can be performed on each sub rating matrix.Finally,the efficiency and effectiveness of the proposed matrix factorization model based on co-clustering method are proved by comparing against various related algorithms.