Collaborative Filtering Algorithm Based On Co-clustering And Matrix Decomposition

The rapid development of Internet and e-commerce has brought a variety of information and information resources to grow exponentially. In the face of serious overload of information. Now e-commerce recommendation system applications increasingly widespread and recommendation algorithm as the core of the recommendation system has also been extensively studied. Especially in collaborative filtering recommendation algorithm, which is one of the most successful recommendation algorithm.It based on an explicit user ratings actions recommended techniques.Along with the rapid growth of users and items in website system, which is facing data sparsity,cold start and scalability of these three major challenges.Users in the use of the process is constantly adding new training data,request recommendation algorithm can quickly and accurately updated. In order to improve the traditional collaborative filtering algorithms problems, this paper proposes a collaborative filtering based on combination of co-clustering and matrix decomposition algorithm.First of all, the two users similar taste just for some projects, the traditional collaborative filtering algorithm often ignores this, but all the items are taken into account. Aiming at this problem, this paper introduced the algorithm based on clustering.The algorithm divide the original matrix into several sub-matrix,the sub-matrix is much smaller than the size of the original scoring matrix,which not only reduce the amount of computation,but also alleviate the problem of data sparsity.After the new data is added, only you need to update a number of sub-matrix in one or more of an unknown matrix score.Then,in the forecast period,choose the method based on matrix decomposition for the final score predicts. In a matrix based on the traditional matrix decomposition of regularization constraint to prevent model fitting phenomenon,can effectively relieve the extensibility.With film recommended with as the carrier.On the basis of the traditional matrix decomposition(SVD),dig deep characteristics of data set,to join the global offset and the impact of social interest change over time,in order to improve the prediction precision.Finally,The clustering and score prediction experiments on the MovieLens datasets,and compared the performance of the new algorithm and traditional clustering methods,in order to illustrate the performance of the merits of the proposed algorithm.Experimental results show that the method is highly efficient.