Research On Recommender System Based On User Social Relationship

Nowadays,people's lives are filled with a huge amount of information,and the information overload that comes with it has become the primary problem troubling people.While helping consumers find the goods they need conveniently,the recommender system can also promote the sales of goods,which is essential for both consumers and commodity producers.Accuracy is one of the most important indicators of a recommender system.On the one hand,the recommender system can mine the connection between users and item data itself.On the other hand,it can take advantage of time series,social network information,tag data,hot spot data and so on to improve its quality.As the core of social networks among these information,social relations have attracted more attention with the popularization of the Internet in a high speed.In real life,users are always willing to accept the recommendation of trusted friends,and the extraordinary results will be achieved.What's more,the similar preferences exist among users with similar social relationships,and the interaction of users with deep connections is more easily realized.The recommender system based on users' social relations makes the recommendation of items through social relations.Firstly,the users' social networks are constructed by it in line with the social relations between people,and through the social networks,it recommends the most favorable products for users.In contrast with other recommended algorithms,this kind of algorithm often has higher accuracy and interpretability,which is also more easily to be accepted by users.Many problems still exist though the social relations of users are widely used in recommender systems.Firstly,most social relations are obtained through the authorization of users.This way can reflect the user's subjective will more directly,but it often destroys the privacy of user,and users do not want to accept the time cost to be paid during this period.Under the influence of privacy and time cost,the serious distortion of the model will be caused due to the sparse social networks,which will lead to huge recommendation deviation.Furthermore,the explicit social relations of users mostly only take advantage of trust relationships.The potential relationship between users cannot be well reflected by a single trust relationship,so the accuracy of the recommender system will be seriously restricted.To solve these problems,the following research is conducted in this thesis:(1)Matrix factorization algorithm based on user bias and implicit social relations(NMFPS)is proposed.The algorithm first calculates the user's implicit social relationship value through user similarity and global social relationship,and then introduces the user bias vector to express the user's own rating preference,and then reduces the dimension of the social relationship matrix.User social vector,user bias vector,user feature vector and item feature vector are used to balance the weight of user social relationship and user bias,and finally the optimization algorithm Adam is used to obtain the optimal user vector and item vector.(2)Bias probability matrix factorization algorithm Based on social similarity(TSSPMF)is proposed.The algorithm first calculates the social relationship between users through the0-1 trust relationship,then calculates the social relationship similarity matrix through the user social relationship matrix,and then performs matrix factorization on the social relationship similarity matrix and the social relationship matrix,so that the two matrices share the same set of user eigenvectors,and the two matrices are constrained by the user eigenvectors.At the same time,bias items are added to express the preferences of users and items,so as to better describe the characteristics of users and items.Finally,the probability matrix factorization model is used to fuse the social relationship matrix and the social relationship similarity matrix and solve it iteratively.(3)Design a recommender system application platform based on user social relations(USRS).The platform recommends movies for platform participants through the fusion of explicit and implicit social relationships.Even users who use the platform for the first time can get a good experience,which can significantly alleviate the cold start phenomenon of the recommender system.In addition,the recommender system can also collect implicit social information when users browse products,constantly revise user preferences,and further optimize the user model.The longer the user spends on the platform,the better the recommendation effect.