RESEARCH ON PERSONALIZED RECOMMENDER SYSTEMS BASED ON MATRIX FACTORIZATION

Abstract

With the development of social network, there is a large amount of variable information has been made by social network users. We can mine these social data from social network, and find the preference latent relationship between user and items. We would make a model for the user and give a recommended items list to user with a suitable recommender algorithm. That is a variable research subject. So our research would achieve a personification recommender system based on matrix factorization. The research will deal with large-scale user-item ratings matrix. In order to improve the recommender systems’ performance we study the social relationship and the implicit feedback of the user. We add a social regularization, demographic information configuration term and users’ consumer records as item’s latent factor bias terms in the matrix factorization optimization function. Through experiments we recommend more accurate results than CF algorithm and SVD algorithm.