Research On Key User Identification Method Based On Influence In Social Networks

Social network is composed of users and relationships between users, whatever researchon social networks is related to the users. With the rapid development of Internet technology,the user scale of social networks that poses a serious obstacle for research and developmenton social networks becomes larger. Identify key users of social networks can effectively solvethe problem, but, the current identification methods have defects that single identificationbasis, low accuracy and can not identify key users more comprehensive.Currently, there are few methods to identify key users in social networks, and mostidentify key users through centrality analysis based on users’ topology. One calculationmethod of centrality measure is degree centrality, it abstracts users as nodes, and set thestatistics on adjacent nodes as degree centrality measure, namely the degree of node (In graphtheory, the degree of node is the node’s adjacent node number.). The bigger degree of node,the more important the node is. So the corresponding user is more possible to be key user.This method is based on the power-law distribution of node——the proportion of nodeswhich have big degree is very small, the proportion of nodes which have small degree is quitelarge, and so nodes with big degree play a greater role and are more important than the nodeswith small degree in the same social networks. Although this method can identify key usersquickly, the identification accuracy of key users with relatively small is not high.The method of this paper makes full use of social networks user’s features, divides socialnetworks into several subnets by cluster algorithm, and calculates user influence in everysubnet, then identify key users by user influence. The core of the method proposed in thispaper is user influence calculation which consists of four parts: quantify user feature, activeuser’s link intensity and influence algorithm, calculate similarity between active user andinactive user, inactive user’s influence algorithm. In these parts, active user’s link intensityand influence algorithm and inactive user’s influence algorithm are the best important. Activeuser’s link intensity and influence algorithm is based on HITS, it replaces webpage with user,sets trust relationships between active users as link relationships, and constructs transfermatrix by user feature instead of webpage adjacency relationship. So the authority and hub calculated by HITS corresponding to active user’s link intensity and influence. The base ofinactive user’s influence algorithm is Linear Threshold Model (LTM). The Improved LinearThreshold Model (ILTM) has advantages as follows: avoiding errors brought by randomlyselected active users； replacing random threshold with similarity between active user andinactive user which is more consistent with the real situation of mutual influence betweenusers； showing more fine-grained inactive user’s influence calculation method bycomprehensively considering active user’s link intensity and similarity between active userand inactive user, which makes inactive user’s influence more accurate. After getting all theusers’ influence, comparing user influence with influence threshold to identify key users,whose influence bigger than the influence threshold.Apply the proposed method and current method to real social networks——Epinions, theexperimental result shows that, the proposed method can identify key user better, and getbetter results in accurate, stability.