Signed Network Preserving Recommender System Model

With the rapid development of the internet,the huge increase in the amount of information on the network makes the problem of information overload increasingly serious.In order to meet the individual needs of users,the recommendation system emerges as the times require.The conventional collaborative filtering(CF)recommendation system analyzes users' purchasing behavior to understand their preferences,builds user-item rating matrix according to the ratings of the items they have purchased,and then recommends the users based on the assumption that similar users have similar preferences.However,this approach is facing many problems,such as the problem of data sparseness.Due to the huge scale of users and items in the network,the user's rating matrix for items will be sparse.Therefore,it is difficult for the system to provide users with the most appropriate recommendations according to the existing information,which seriously affects the performance of recommender system.Trust-aware recommender system(TARS)appears to alleviate this problem to a certain extent,it mainly uses the trust transitivity to establish the trust relationship between users,combines user-item rating matrix between users and items to recommend.However,this method does not consider negative link information in the network.A limited number of negative links often contain more information than positive links,and the existence of negative links reduces the trust of other users to current users.Therefore,the value of negative links should not be ignored.In this paper,we introduce the negative link information and propose a new recommendation system model based on the signed network.This recommendation system,which combines positive and negative relationships in signed social networks,can effectively improve the accuracy of the recommended results.The model integrates user model with trust and distrust networks,analyzes from the perspective of attributes,extracts the social attributes and structural attributes related to users' positive and negative links,integrates the attributes by the method of logistic regression,assigns the related attributes and calculates intensity of trust among users to identify trustworthy users and then aggregate those users' suggestions to provide useful recommendations to new users.Our experiments were carried out on the real datasets Epinions and Slashdot.Compared with the existing recommendation algorithms,our experiments validated the effectiveness of our proposed method,it has a good performance in terms of accuracy,coverage and so on.Compared with the conventional CF method,the proposed method improves the accuracy by 15%,increases the F value by 9%,and increases the coverage rate by 6%.Thus,this article combined with positive and negative information in the network recommended system to provide users with better recommended results.