Research On Influence Mining Method In Social Network

With the popularization and development of the network, the virtual society penetrates into people's lives gradually. There are many large social websites such as Facebook,Twitter,Sina Weibo, etc.People interact with each other through these virtual networks. The influence between individuals in social networks can affect user's behavior and social dynamics, so the mining and analysis of influence between individuals becomes one of the key research issues.And further studies on the influence can improve the efficiency and accuracy of many applications such as viral marketing, recommendation or information retrieval systems.However, most current researches on social influence are focused on verifying the existence of social influence, and there are no or little research on mining directly or indirectly influence strength in social networks. These lead us cannot give quantitative influence intensity between the members of the social network nodes. To solve this problem, we study the methods about influence mining in social network, the main contributions and innovations of this research include:Firstly,this thesis proposes a topic finding method called ToFiM,which can learns the influenced and influential users' topic distribution using generative model LDA, ToFiM provides the foundation for the subsequent influence mining methods.Secondly, this thesis studies a theme-oriented influence mining method called ToiMM.Which can mine the direct/indirect influence strength between nodes with different attributes and relationships between different nodes in social networks based on the formal descriptions and related definitions of the influence mining problem. ToiMM includes subtasks: 1) use influenced users' and influencing users' topic distribution to estimate the direct influence; 2)based on the direct influence strength we learned before, we estimate the indirect influence strength by using two different influence diffusion and aggregation mechanisms: conservative and non-conservative propagations. In addition,the algorithms analysis is given and an application example is given to explain the using process of ToiMM..Thirdly, several experiments are designed and implemented to verify the effectiveness of the methods.The experiment results on Twitter and Citation prove the feasibility and correctness of the ToiMM method.Finally, based on the research on the previous two influence mining methods, a probabilistic generate model BBModel and the corresponding user buy-back behavior prediction methods are proposed, and the experiments result prove the feasibility and effectiveness of the proposed model.