Research On Collaborative Filtering Algorithm Combining User Trust With Item Preference

Continued prevalence of the electronic commerce system make some compa nies welcome new opportunities.But with all kinds of items and information in site persistently updating,categories and amount constantly increasing,it is m ore trouble and sparing them more time that users search for some their really wanted information.Collaborative Filtering provide a new train of thought to solve this question.Recommended system analysis and predict the users' inter est to recommend the items to users,advantages and disadvantages of its algor ithm have direct influence on business and users in the e-commerce site.This a rticle will improve collaborative filtering algorithm that is more applied in reco mmendation system.(1)Based on the question that the traditional method of Collaborative Filtering can not mine the deeper relationship between users,referring to trust relationship in social network sites,this article will add the trust relationship to building method.This trust relationship is divided to personal credibility and trust degree between users.In order to calculate personal credibility and trust degree between users by using this trust model,the method in this paper analysis of the rates of system.Then,the method in this paper screen trusted neighbors being refered by predicting users' rates.(2)In traditional similiary algorithm,Whatever target items algorithm face,target users and their neighbors similiary being used by algorithm is same.But in reality,for different items,preference between the same two people is not same.So the method in this paper add the preference factor to traditional similiary to improve similiary method.Then,improving similiary method combining trust degree screen improve predicting accuracy.(3)Based on the question that Collaborative Filtering cannot recommend new items,the method in this paper refer to users having preference for kinds of items.Trust relationship combating new item method that refer to some neighbors' preference on item attribute predict the rates of new items.In order to reduce negative effects of the sparsity,the article incorporate userreliability and improve similarity.Then,the users' preference on item attribute is used to recommend new items.Finally,in order to verify the validity of the method,Movielens data set was used to compute the mean absolute error of kinds of algorithms.After verification,compared with traditional algorithm,the improving method is more accuracy under the sparsity and cold start condition.