News Recommendation System Based On Improved SVD Collaborative Filtering Algorithm

With the advent of the mobile Internet era,mobile phones have become the main tool for shopping,reading and social activit ies.Every day people use the phone to browse a lot of news information,but the face of massive information is often unable to start,most of the time wasted in looking for information of their own interest.The news recommendation system uses personalized recommendation techniques to help users quickly find information of int erest from massive news and information.Collaborat ive filtering is the most widely used recommendation algorithm in the recommended system.The principle is to predict the unknown items according to the neighbor data of the target users,and then complete the personalized recommendation.This paper discusses the classification,implementat ion principle and evaluat ion index around the collaborat ive filter ing,and discusses the existing problems in collaborative filter ing,such as sparse data and cold start problem.This paper introduces the collaborat ive filter ing based on SVD in detail,uses the latent factor model to alleviate the problem of scoring prediction caused by sparse matrix,and further introduces SVD++ and TrustSVD based on SVD.The implicit feedback information and trust network are added respectively,which makes the accuracy of the recommendation result greatly improved and the defects in the trust model has been improved by adding a trust weight and a trust bias factor.At the end of this paper,we design and implement the news rec ommendation system based on android,including the background management system and android client,and the related technology and function modules involved in the system are introduced in detail.The system builds the trust model and the user model according to the user's social network and behavior data,uses the score matrix and the trust matrix to complete the individual recommendation.The recommendation engine adopts different recommendation strategies according to the different states of the users,making the recommendation result more diversification.