Research On User Portrait Construction And Recommendation Based On Social Networks

With the rapid development of mobile Internet,social networks and various industries on the Internet penetrate each other,and the social fission effect is obvious.Mining the relationship between users and users in social networks to provide better recommendation services has become a research hotspot.The recommendation algorithm based on social network makes use of users' social relations to alleviate the sparse scoring data problem.However,the matrix factorization recommendation algorithm based on direct social relations has the problem of sparse direct social relations,which affects the accuracy of recommendation results.In view of the above problems,this paper focuses on the impact of indirect social relations on recommendation algorithms,and makes use of community structure and social influence to achieve better recommendation results: firstly,direct use of social network structure information to calculate the influence between users with indirect social relations;secondly,research on user portrait construction in social networks to better mine users' characteristic information;Then,user portraits are used to calculate the influence among users to improve the accuracy of recommendation results.The main research work and contribution of this paper is as follows:(1)Influence calculation and recommendation based on social networkIn order to alleviate the problem of sparse direct friendship,considering that users in the same community will have similarities in some aspects and influence each other,this paper uses social relationships among users in the community to recommend.A matrix factorization recommendation algorithm,SoInf,is proposed to integrate community structure and social influence.The algorithm uses both user rating information and social network structure information to calculate social influence among users and construct recommendation model.When using social network structure information to calculate the influence among users,it combines the social distance between users.Considering that users with large personal influence are more likely to influence other users,the asymmetric influence among users is obtained by combining the personal influence of users with the influence among users.To enhance the recommendation effect,the influence of community information and users is combined into the recommendation algorithm at the same time.Experiments show that the accuracy of this algorithm has been improved to a certain extent compared with the existing algorithms.(2)User portrait construction based on social networksIn order to make better use of the structure information of social network and get more accurate user feature information to improve the recommendation effect,this paper proposes a user portrait construction algorithm DBTNR based on deep neural network.The nodes in the sequence of nodes generated by random walk follow the power law distribution as the words in the document.The algorithm takes the sequence of user nodes generated by random walk as sentences,and uses the model in natural language processing to construct user portraits.Considering the difference of social distance between users,the position information of users in the sequence is integrated into the user portrait construction.Moreover,the deep bidirectional conversion coder pays attention to the left and right nodes of the user nodes at the same time,so as to better obtain the local and global information of the user in the network,so as to improve the accuracy of the user portrait.Experiments prove the accuracy of user portraits constructed by DBTNR algorithm.(3)Influence calculation and recommendation based on user portraitIn this paper,user portraits are used to calculate the influence among users.Combining with community structure information,a matrix factorization recommendation algorithm based on user portraits is proposed,which is called SoUMPF.The algorithm uses the community discovery algorithm to get the user's community information,and then uses user portraits to analyze the difference of implicit features between users in the community,in order to calculate the influence of users.Finally,the influence between community information and users is combined into the matrix decomposition recommendation algorithm.The real data sets verify that the proposed algorithm has higher accuracy and better recommendation effect.