Research On Individualized Recommendation Model And Algorithm In Social Network

With the rapid development of social networks the rating information from Internet hasbeen enormously increased. It becomes increasingly difficult for users to obtain informationneeded from such a vast data, and the search engine would usually bring the same searchresult for all users. In order to meet the demand from user’s personalized service, variousrecommender systems are appearing constantly. There are many limits to the presentrecommendation algorithms, and traditional collaborative filtering recommendationalgorithms cannot effectively filter out items and products to adapt to users’demands becauseof the lack of social information from social networks which would lead to relatively poorrecommendation quality.According to the problems mentioned above, this article focuses on the study abouteffects of factors including trust relationships among uses, time factors and geographicalpositions etc. on personalized recommendation algorithms based on social networks. A newpersonalized recommendation model is established by integrating many kinds of contextinformation to improve user satisfaction and to achieve personalized recommendation of highquality. The main research contents and research contributions of this article are as follows.A deep analysis and discussion on technologies of matrix factorization have been carriedout in this paper in light of the defects of traditional collaborative filtering recommendationalgorithms, as well as an analysis of effects of feature vector dimensions on recommendationquality and efficiency of probability matrix factorization algorithm (PMF). Usually the PMFalgorithm only made recommendations based on the user-item rating matrix, withoutconsideration of the possible dynamic changing of users’interest over time, which will lead toinaccurate recommendations. Consequently in this article the TPMF model is put forward,which is a probability matrix factorization algorithm integrated with the time factor. And ithad been proved empirically the better extensibility and higher recommendation accuracy ofTPMF to solve the problem of data sparsity.According to the issues of personalized recommendation based upon trust in socialnetworks, the recommendation model CETrust is propose, which comprehensively took intoaccount the following factors, including direct and indirect trust between users, themechanism of trust propagation and the similarity between users and so on. CETrustintegrated trust relationships and user similarities into the process of probability matrixfactorization, and carried out analysis of latent factor feature between the same preference of the chosen trust users and that of the target users. Experiments showed the obviousadvantages of CETrust in that it can improve user satisfaction remarkably compared withother matrix decomposition algorithms and trust-based personalized recommendationmethods.In order to further improve the accuracy of personalized recommendation algorithms insocial networks, a new personalized recommendation model called TrustSeqMF is built inwhich the trust propagation mechanism, the time sequence information and users-item ratingmatrix information are fused into the probability matrix factorization model. TrustSeqMF isable to learn user’s feature vector by trust relationships even if that user has not rated any item,because it will learn the latent feature vectors both of users and of items, considering thetimes factors and dealing with trust relationships. TrustSeqMF can better solve the cold startproblem and improve recommendation accuracy compared with the existing algorithms. Theanalysis of time complexity shows that TrustSeqMF can easily be extended to applicationscenarios with larger data sets.With the popularity of mobile social networks at present, this paper also puts forward acontext-aware based recommendation model in mobile social networks--CMSR, whichanalyzes the latent social relations between users in mobile social networks and makesposition predictions that users might be interested in by means of combining users’ socialinformation such as time and place into the recommendation algorithm. The feasibility ofCMSR has been verified empirically, as well as the improvement of recommendationaccuracy in the mobile social networks.The followings are the innovations of this article:1. The influence of time factors and trust relationships on personalized recommendationalgorithms in social networks is analyzed and integrated into the correspondingrecommendation algorithms, which has increased the accuracy and extensibility ofrecommendation algorithms.2. In order to deal with the problems of personalized recommendations in mobile socialnetworks, a personalized recommendation method is proposed which considers bothgeographical position and time information simultaneously, to make the recommendationresults more in line with the actual needs of mobile users.3. A kind of personalized recommendation algorithm integrated with multi-source socialinformation of social networks is put forward, and the superiority of the proposed algorithmhas been tested empirically. In this article, based on the discussion of the research contents and innovation pointsmentioned above, the effects of various social factors on the precision of personalizedrecommendation algorithms in the social networks have been analyzed in detail, and therecommendation models integrated with diverse social factors has been built and applied toactual recommended scenarios. The experimental results show that the proposed algorithmscan achieve better recommendation precision and provide effective help for further researchon social networking personalized recommendations.