Distrust Prediction Based On Heterogeneity Of Social Network Users

The rapid development of Internet has changed the way of human life and habits of life, meanwhile, it also gradually changes social relationship between people. Unprecedented changes have taken place in human social relations. People are not satisfied with the current of the communication between friends in real life, the users who talk to each other in the current social network may have never met, even in a different country or at the edge of the earth, so in this way the users under the different relationship of time and spacer is active in the social network that we live together in. Therefore the current social relation network is unprecedented large and complex. Based on the complex relation of users’ interaction, it is extremely important and difficult to find the trustless relationship in the large and complex social network.In the social network, the trusted and distrusted relationships between online users are critical to seek information available in complex networks. People always used the experience in real social life to find online social relationships. People are more inclined to communicate with online users he trust, and distrusted relationship can often avoid some network cheating, website marketing, etc, misleading false information, so as to find the available needed information. But in reality the symbol discovery in the social network, find distrusted relationship between the users in the adjacency matrix of the relationship between the users is very sparse, so the rare discovery of useful information in vast amounts of user relationship is like a needle in a haystack. So for the difficult problems of the distrusted relationship of the users found in social symbol network users, a new prediction algorithm structure He-Distrust is put forward in this article. The main ideas of he-Distrust are: While applying the idea of optimization theory, not only using matrix decomposition method to find existing extremely huge distrusted relationship in the adjacency matrix, but also by analyzing the heterogeneity between online users to find the influence of the prediction of the distrusted relationship, so as the accuracy of the prediction of the distrusted relationship between users is improved. Two key aspects are included in this architecture,(1) the nonnegative matrix factorization is a minimization problem with nonnegative constraints, through proper transformation and decomposition, high-dimensional raw data vectors are expressed as a set of linear combination of the low dimensional vector, due to the analysis of the structure of the original data attributes which can be used to identify.(2) Researches of epinion users data set found that users with distrusted relationship appeared different heterogeneity in scoring the goods, using the weighted(Mahalanobis distance) to get a normalized degree matrix, the optimal solution in the normalized difference matrix to find the high heterogeneity effect on distrusted relationship, and use it to help predict distrusted relation in social symbolic network.The prediction of relationship between social network mainly divided into the prediction of trusted relationship and distrusted relationship. This paper mainly discusses the influence of the use of user rating heterogeneity producing on the forecast of distrusted relationship. We have described the methods and the difference of existing prediction of distrusted relationship in unsupervised learning, and integrated the Heterogeneity effect in optimization of problem to form He- Distrust architecture. At last, through the application of real user data sets in Epinions, compare and verify the feasibility of proposed method, and then explore the influence of the Heterogeneity between users on the prediction of distrusted relationship in a social network.