Research On User Multi-attribute Matrix Factorization Recommendation Algorithm

With the rapid development of Internet technology,the amount of information on the Internet is increasing,making it difficult for people to find content of interest in a huge amount of information.The recommendation system is one of the effective tools to help people solve this problem.The recommendation algorithm based on matrix factorization is one of the hot topics in the current recommendation algorithm.The algorithm decomposes the user's scoring matrix into user feature matrix and item feature matrix,which has the advantages of strong expansibility and high prediction accuracy.However,there is still a poor prediction accuracy on sparse data sets,cold start and other issues.This thesis is analyzed the recommendation algorithm of matrix factorization and make the following improvements to the problems existing in the algorithm.Firstly,for the problem that the prediction accuracy caused by data sparse is not good,a clusterbased non-negative matrix factorization recommendation algorithm is proposed.By calculating the Pearson similarity of the user,the algorithm increases the weight coefficient and reduces the similarity calculation.Secondly,the relationship between users is mined,the user's clustering is realized,and the cluster is integrated into the nonnegative matrix factorization.To increase the recommended information.The proposed algorithm is experimentally verified on the data set Movielens.The results show that the cluster-based non-negative matrix factorization recommendation algorithm is superior to the traditional non-negative matrix factorization algorithm in the recommended performance indicators Root Mean Square Error and Mean Absolute Error.Third,the traditional matrix factorization recommendation algorithm only focuses on the relationship between users and items.This thesis proposes a matrix factorization recommendation algorithm combining attributes.Through the analysis of the user's own attributes,the users are associated with each other according to the user attributes,and the user attribute information is integrated into the matrix factorization algorithm model,and the user's personality and commonality are simultaneously considered.The verification results are performed on the data set Movielens.It is shown that the matrix factorization recommendation algorithm incorporating user attributes compares the matrix factorization algorithm using only the scoring matrix.It is better than the latter in the recommended performance indicators Root Mean Square Error and Mean Absolute Error,and the prediction accuracy is improved,Verify the effectiveness of mitigating cold start issues.