Personalized Recommendation Algorithm Based On Sparse Constrained Non-negative Matrix Factorization

With the development and popularization of the Internet industry,it is very convenient for people to obtain and publish all kinds of information,but with the rapid increase of information data,users will face various problems such as information overload.In this huge information base,most of the information is not of interest to users,so it is simply looking for a needle in a haystack for users to find useful information or products they like.Therefore,the emergence of personalized recommendation system solves this problem,which enables users to find their favorite products more easily and quickly,and at the same time provides users with corresponding personalized services.Nowadays,collaborative filtering algorithms and some classic recommendation algorithms are commonly used in personalized recommendation systems,and they are recognized by everyone.However,in the actual application process,these algorithms do not consider the sparsity and diffusion of data,which leads to the reduction of recommendation performance of recommendation systems with the continuous updating and increasing of users and data.In this paper,the related problems have been deeply studied and explored,and the following works have been accomplished through practice:(1)This paper expounds the research background and significance of personalized recommendation system,introduces the research status of non-negative matrix factorization(NMF),personalized recommendation system and K-means clustering algorithm at home and abroad in detail,and focuses on collaborative filtering algorithm,one of the most widely used recommendation algorithms in personalized recommendation system.(2)In order to solve the problems of sparsity,scalability and cold start in the classical algorithm of recommendation system,non-negative matrix decomposition is used to reduce the dimension of matrix,and K-means clustering algorithm is used to classify users to improve the scalability of the algorithm.(3)In order to solve the influence of discrete data in K-means algorithm on clustering results,the K-means algorithm is improved and the data is preprocessed to further improve the calculation accuracy of the model.(4)Combining non-negative matrix factorization with sparse constraints and K-means clustering algorithm,SNMF-K model is proposed,and users are divided by similarity of user preferences to get clustering results.(5)A new combined recommendation algorithm is proposed,and a comparative experiment is carried out with Movielens data set,which verifies the improvement of the recommendation performance of the proposed recommendation algorithm.