Personalized Recommendation Based On Social Network Mining

With the development of science and technology and the advent of the information age,the information is growing explosively.Massive information makes people unable to find valuable information in a large number of data quickly,which leads to the phenomenon of information overload.The recommender system can solve the information overload problem.Recommender system can provide recommendations for users according to their information and historical behavior records,and can predict the ratings of the items that are not rated by users.The traditional recommender system is faced with data sparsity and other problems,which can be well alleviated by social information.At the same time,the rise of social sites makes it possible to use social information for recommendation.This paper mainly studies how to describe users' interests better and improve the performance of the recommender system by mining the social relations among users deeply.This paper first introduces the theory and method of traditional recommender system.Then we introduce the problems and challenges and research based on social recommendation.In this paper,we propose a dual-regularized matrix factorization using ratings and social information.On the basis of matrix factorization,we adds two regular terms,social regular term and nearest neighbor regular term,which constrain the user's preference to follow his social friends' preference and his nearest neighbors' preference simultaneously.Then we proposed a recommendation algorithm based on social friends and trust propagation.In this paper,people who can give advice to users are called the user's consultant group.Through the trust propagation of the consultant group,the relationship between users is transferred.Finally,the performance of the proposed algorithm is verified by multiple data sets.The experimental results show that the proposed algorithm is superior to other algorithms in the accuracy of rating predicting,and can provide more accurate recommendations for new users who have few historical ratings.