Relationship Mining And Influence Analysis For Social Network Group Analysis

With the rapid development of Internet,our society is showing the trend of networking.Users as the core of various social media constitute members of the social network,and the relationship between users as links constitutes the basic structure of the social network.Therefore,analysis of online social network groups is of great significance for maintaining national security and stability and sustaining the long-term stability of society.Researchers have now analyzed and studied social networks from different aspects.Individuals in social networks interact and transmit information based on the link relationship between individuals,and the change of their structural relationship directly affects the breadth and depth of information transmission.Therefore,this thesis makes further research on the technology of relationship mining and group analysis in social network,including link prediction,community discovery and node influence analysis.The main research contents are as follows:Despite the satisfying practicality and effect of traditional link prediction algorithm based on network topology information,it cannot utilize the temporal information in today’s dynamic social networks.The thesis first looks at the link prediction problem as an issue of binary classification,dividing the dynamic network into time slices from which the natural weights in the network are obtained so as to introduce the weighted features based on the network topology to constitute link classification features.Then the moving average model to process the weighted features on multiple time slices is utilized,while the unlinked node pairs are predicted through the classifier.Traditional community detection algorithms such as label propagation algorithm(LPA)are not stable enough.Considering that community detection and clustering have similar ideas,this thesis introduces the integrated clustering into the community detection algorithm,on the basis of which integrated selection algorithm is further incorporated,to propose a new community detection framework based on integrated clustering.The algorithm first generates multiple integration members by running community detection algorithms such as multiple label propagation algorithm.Then it selects diverse and high-quality candidate subsets by the integration selection algorithm and assigns different weights to each integration member using its modularity.Finally,the integration algorithm is used to obtain new community detection results.As for node influence analysis,researchers have proposed many network centrality criteria for analysis.However,a single criterion that all focus on the same aspect can lead to the lack of information.Additionally,most research works tend to turn a blind eye to the information of the community structure in which the nodes are located.Therefore,the thesis proposes an algorithm for node influence analysis that integrates community structure and network topology.The method not only considers the network local topology information,global topology information and community structure information,but also uses the entropy weight method to calculate the weight value of each indicator to reduce the influence of subjective factors.The experimental results show that the method can rank the node influence more comprehensively and accurately than a single centrality criterion.