The Analysis And Prediction Of User Relationship In Social Network

Online social network takes an important role in people’lives and work. Social network analysis has become a significance research area and the results have been widely used in different scientific research, business and politics. In a social network the relationship between users generally refers to the bidirectional association, which is the basic channel of information sharing and user interaction. User relationships may highly influence the formation and evolution of social networks. So user relationship analysis is regarded as a basic problem of social network, which tries to find the intimacy between users and predict whether there is a relationship between users in future. User relationship analysis has been widely used in many fields. For example, in social security, relationship analysis can be used to identify a criminal network structure, according to the known part of the criminal information; in the field of electronic commerce, relationship analysis can be used for recommendation, precise advertising and so on.The existing methods on user relationship analysis mostly adopt the structure of social network, such as the common friends or paths between users. Besides, user attribute information are also used to predict the possible relationship between users. Since node attributes may be on subjective evaluations and usually not assumed to be as reliable as network topology, some methods use user attributes and the network structure for link predication. But they do not consider the sensitivity of the attributes. Since different attributes may have different influence on the relationship of users, how to integrate these data to improve the accuracy of user relationship prediction become an important research problem. There are three challenges:complex social behavior modeling and feature extraction; user preference based user attribute and behavior analysis; and the fusion of multiple metrics.In this paper, we propose the hybrid link inference metrics, which integrate user behavior evaluation into the attribute and structure based measurements. To have a deep understanding of user behavior, we introduce the concept of latent factor to catch the intrinsic correlations between social purpose and behaviors. We consider user preference from two aspects:one is to calculate the attribute importance separately for each user; another is to take into account the bilateral wishes when evaluating a link beyond the traditional one way or an overall measurement. To semantically combine these measurements together to infer a link, we propose two combination modes, independent fusion and interdependent fusion, and apply the information theory to quantify the sensitivity of considered elements. Experiments are performed on several real data sets and the results show that our metrics have better performances than previous methods.