Rumor restraining based on propagation prediction with limited observations in large-scale social networks

In order to minimize the negative effect of malicious rumors in large-scale OSNs (Online Social Networks), researchers have proposed numerous solutions, such as controlling important user nodes and controlling bridges of social communities. However, these methods rarely take the space-time dynamic of rumor propagation into consideration. The selected controlled user nodes or bridges may be unable to influence the propagation process actually at current time if they are far from the rumor source or they have already undergone the rumor before. In the above mentioned two scenarios, it will be meaningless to control so-called important users and community bridges. In our work, we aim to restrain the rumor by predicting propagation dynamic from the microscopic perspective and collecting the boundary users who are most likely to be contagious at the moment. Moreover, to predict rumor's propagation dynamic practicably and efficiently, we adopt the sensor observation and meanwhile assume that there exist some short propagation paths which are explicit. We experimentally demonstrate that the proposed microscopic model's estimations rather accurately predict the propagation dynamic of the rumor. Moreover, the proposed rumor restraining method outperforms evidently classic Degree Centrality and Target Immunization approaches regarding the immunity scale and the immunity speed.