Research On Recommendation System Based On Non-negative Matrix Factorization And Clustering Algorithm

Personalized recommendation system through the analysis of user behavior,for the user to select from the massive data that it may be interested in information,in e-commerce,movies and social networking sites and other fields have a wide range of applications.In many of today's recommendation algorithms,collaborative filtering recommendation is the most recognized and widely used method.However,with the increasing number of users of the recommendation system,the traditional collaborative filtering algorithm exposes many problems at the same time,such as data sparsity and scalability.This article has carried out research and practice on related issues.The main tasks and innovations completed are mainly reflected in the following aspects:(1)Explained the research background and domestic and foreign research situation of the recommendation system,made a basic introduction to the personalized recommendation system,and focused on several related technologies in the personalized recommendation system.(2)In order to solve the problem of data sparsity and scalability of traditional collaborative filtering algorithms,non-negative matrix factorization is used to standardize and reduce the dimension of user item scoring matrix.K-means clustering algorithm is used to classify different types of users and improve the scalability of the algorithm.(3)In order to improve the shortcomings of the noise point and random center point in the K-means algorithm which have a great impact on the clustering result,we combine the idea of density and the maximum and minimum distance to propose a K-means clustering algorithm to optimize the initial clustering center.Methods,and experimental verification using UCI data.(4)Combining non-negative matrix factorization with K-means algorithm,a combined recommendation system based on non-negative matrix factorization and clustering algorithm is designed and implemented.Experimental comparison is performed using MovieLens data to verify the effectiveness of the proposed method.