A risk defense method based on microscopic state prediction with partial information observations in social networks

Abstract

The development of network science has led to an increase in the size and user number of social networks. Messages (e.g., rumors, leaked user information) will quickly spread to social networks and lead to terrible results. Researchers have proposed a number of protection methods in risks’ propagation process, such as blocking pivotal topological nodes, controlling the bridges between social communities, etc. However, these methods mainly focus on static topological characteristics of the networks and rarely take the spatio-temporal diffusion dynamic of risks into consideration. In fact, if the selected controlled nodes or bridges are far enough away from the risk source or have already undergone the risks before, they cannot actually affect the risk propagation process at current time. To solve this problem, we propose a microscopic risk diffusion model and aim to defend against network risks and threats by predicting their dynamic propagation from the microscopic probability perspective and collecting the infection boundary nodes that are currently most likely to be contagious state. Meanwhile, in real life, we often fail to obtain the monitoring data of all network nodes, so we use the sensor observation and assume that there are some short propagation paths that are clear to us. We experimentally demonstrate that the estimations of proposed microscopic diffusion model fairly accurately predict the propagation behaviors of the network risks. Moreover, on average, the proposed risk elimination solution based on microscopic state prediction with partial observations outperforms acquaintance immunization and targeted immunization approaches in terms of defense effects and by approximately 30% in terms of defense cost.