Research On Influence Maximization And Diffusion Model In Social Networks

In recent years, with the booming development of software and hardware, and the popularization of personal computer and Internet, many online social platforms such as WeChat based on friends relationship, Renren based on schoolmate relationship and Weibo based on follow relationship, are so popular that people spend almost all spare time with them.These platforms output huge data every day and bring unprecedented chances for the analysis of social networks, as a result, lots of researchers are attracted to research and analyze the social network structure and diffusion laws. And how to choose the most influential TOP-K nodes and how to choose social network diffusion models, are the best choices for researchers.In this paper, a new algorithm is proposed to improve existing algorithms for social network influence maximization firstly. Secondly, we analyze the Independent Cascade Model and Linear Threshold Model in detail, and a new diffusion model in social networks is proposed by combining the law of forgetting and the phenomenon that people have different reactions when getting the information firstly and getting the information again in the later time. There are the main innovations below.(1) LDDC(Linear-Decrescence Degree Centrality) based on Three Degrees of Influence Rule. According to Three Degrees of Influence Rule, influence is working effectively within three degrees, and once beyond, the influence approaches zero almost. So LDDC evaluates nodes' real influence by calculating the potential influence within three degrees, and the potential influence decreases to a times when it diffuses from origin node to nodes two degrees away, then decrease to ? times again when it diffuses to nodes three degrees away,there 0<?, ?<1. After calculating the value of LDDC, we evaluate the performance in four social network datasets from three points of view.(2) The Hybrid Diffusion Model. There are some facts in real human relationship networks: whether people accept the information or not depends on the information itself when they firstly know; If they refuse, whether people accept the information every time after the first failure depends on the accumulated influence from those who tried to convince and failed previously and the one who tries at present while the accumulated influence will decrease as the time goes on because of the law of forgetting. A new diffusion model named The Hybrid Diffusion Model, which has strong similarity with diffusions laws,is proposed by absorbing the advantages of Independent Cascade Model and Linear Threshold Model. At last,we use two methods to validate the effectiveness in Wikipedia vote dataset.