Rumor Source Mining Based On The Complex Network

With the popularization of the Internet, the information is exploding, and the speed of information transmission is very easy to form a cascade effect, so it is very important to control the abnormal information spreading under the complex networks, including social networks. The abnormal information in the network, especially in the network environment, has seriously affected the legitimate interests of the parties and leads to social panic, so it is necessary to control the spread of this kind of rumor. If we can dig out the source of information diffusion, it has important practical significance for the control of the spread of the rumor. The difficulty of mining is how to find the location of the information source, in the situation which the path and the relationship between nodes is unknown and the time information of nodes is in the unknown state.The existing mining methods are divided into two types, one is based on the spread of sub graph mining, the method requires that the complete communication sub graph of the network, which is based on the observation point. The other is based on the observation point.Based on the analysis of the existing model, the paper studies the following aspects: first, we can get the IRC(based on maximum a posteriori) model, which is based on the maximum of the probability distance. Second, In-depth analysis method based on complex network centrality metrics, looking for more stable lifting scheme DRC model(based on complex network closeness centrality computation model) in general. Third, the model of the observation point deployment scenario is deeply studied, and the effect of the model TRC(based on the time label mining model) is not too dependent on the observation point deployment, but based on the similarity of the time difference between the spread time and the theory. Fourth, the existing CMS(based on community detection) model is studied, and we propose ICMS model which is based on the model of the transmission probability distance. The stable region of the infected area is generated by multiple iterations. Then, the range of the infection is better and we can also get low mean error.The results of this paper are as follows: firstly, the improved IRC model can better express the actual propagation distance, which is more accurate than the original one. Second, the effect of DRC model is greatly improved than ECC(measure based on the eccentric distance calculation in complex network centrality) model. Third, in the actual application of the scene in the observation point of the deployment of mining scene, the DRC model has the right to observe the situation, the paper proposed the TRC model based on the time label can be a good solution to this problem. Fourth, the ICMS model proposed in this paper is better than the original CMS model based on adaptive clustering.