Research And Improvement Of Recommendation Algorithm Based On Clustering And Matrix Factorization

The rapid development of the Internet has caused the human society to face serious information overload problems.The recommendation system is a powerful tool to solve this phenomenon.However,the traditional collaborative filtering recommendation algorithm has problems such as data sparsity and interest drift of users.The former will greatly affect the accuracy of similarity calculation,and the latter will make the recommendation result not perfectly fit the users' requirements.Aiming at the problems above,two solutions based on clustering and matrix factorization models is proposed in this paper.Based on the clustering model,a collaborative filtering recommendation algorithm based on double clustering on users and items DCCF(Double Clustering-based Collaborative Filtering)is proposed in this paper.DCCF clusters the score matrix from two sides including users and items to decrease the influence of the data sparsity.And an improved similarity calculation method P-J-T(Pearson-Jaccard-Time)correlation coefficient which combines the time information with the Pearson correlation coefficient and the Jacard coefficient is presented in order to reduce the impact of interest drift of users and improve the accuracy of similarity calculation further.Based on the matrix factorization model,a matrix factorization algorithm based on the user similarity and item similarity USIS MF(User Similarity and Item Similarity-based Matrix Factorization)is proposed in this paper.This algorithm puts user similarity and item similarity into the loss function.In this way,the similarity relationships between users and items will not be changed when the matrix is factorized.At the same time,the model will be more refined and the effect of matrix factorization will be improved.In addition,two methods handling with prediction scores are proposed.Mapping can solve the problem that the prediction scores exceeds the system scoring interval.And the prediction round to neighbor algorithm PRN can reduce the calculation error besides improving the accuracy of recommendation.In this paper,simulation experiments are done to study the influence of parameters on the two algorithms and validate the effectiveness of these algorithms.The results prove that the two algorithms proposed in this paper can effectively reduce the impact of data sparsity and interest drift of users besides improving accuracy of the recommendation.