Research On Recommendation Algorithm Based On Matrix Factorization

The recommendation algorithm based on matrix factorization is a popular recommendation algorithm.It can not only decompose the high-dimensional user-project scoring matrix into low-dimensional matrix by training,but also improve the recommendation accuracy of the recommendation system with the increase of scoring data.However,there are many problems in the recommendation algorithm based on matrix factorization.Next,this paper will study the matrix factorization recommendation algorithm in depth.At present,most of the recommendation algorithms based on matrix factorization exist: the implicit feedback information of users and projects is considered to be single,and the feedback information is directly used without correlation measurement;Calculate the problem of inaccurate correlation between users and projects;And data sparsity problem and so on.This paper puts forward the corresponding solution.The main contents of this paper are:(1)A personalized recommendation algorithm that combines project and user implicit feedback is proposed.That is,the matrix factorization based on the implicit feedback of the project and the matrix factorization based on the implicit feedback of the user are combined with a certain weight,and the introduced user and project feedback information is measured by a correlation function.Improved the problem of inaccurate matrix factorization prediction scores caused by user and item implicit feedback information considerations and inaccurate use of feedback information.(2)In order to further improve the accuracy of matrix factorization for predictive scoring,this paper has also improved the similarity measurement method.The similarity calculation not only considers the user-project score,but also includes the attribute information of the project.(3)This paper proposes a method of data filling based on popular attribute labels to improve the data sparse problem of matrix factorization.This not only enhances the interpretability of the filled data,but also improves the accuracy of the prediction score of thematrix factorization.This paper uses MovieLens data set to carry out experimental verification from multiple angles.Experiments show that the algorithm proposed in this paper effectively improves the accuracy of prediction score.