Minimizing the influence of rumors during breaking news events in online social networks

Abstract

The malicious rumors have tremendously attracted a more substantial number of researchers to join the fight against the propagation of these types of information in online social networks (OSNs). The spread of rumors has a severe impact on society, which can creates political conflicts, shape public opinion and weakens their trust in governments; therefore, it must be stopped as soon as it is detected. This paper investigates the problem of minimizing the influence of malicious rumors that emerge during breaking news, which are characterized by the dissemination of a large number of malicious information over a short period. Therefore, we introduce the problem of multi-rumors influence minimization (MRIM) in OSNs and propose a solution to it. To this end, we design a multi-rumor propagation model named the HISBMmodel that captures the propagation process of multi-rumors in OSNs. Moreover, we present a new formulation of an individual’s opinion toward a rumor based on a Markov chain representation, which adds a layer of realism to the proposed model. Subsequently, we propose a dynamic blocking period (DBP) approach as a solution for the MRIM problem. The main objective is to minimize both the spread and the influence of these rumors in OSNs. The proposed method selects and blocks nodes that most likely to spread a large number of rumors and support them. Different from existing methods, the proposed solution does not block nodes for an unlimited period, but this period is estimated according to the high activity of a node in an OSN. The survival theory has been exploited in this work to provide a solution formulated from the perspective of probabilistic inference of networks. Consequently, an algorithm has been proposed based on a likelihood principle to select the target nodes, which guarantees a (1−1∕e)-approximation of the optimal solution. The experimental results show that the HISBMmodel could capture the propagation of multi-rumor propagation more accurately than classical models and provides metrics to assess the impact of rumors efficiently. Moreover, the results show the outstanding performance of the proposed approach compared to the other solution in the literature. The experimental results show that in the worst-case, the DBP achieves on an average 37.66% reduction on the impact of rumors, compared to 18.46% obtained by the second-best method. However, in the best-case the performance of the proposed method reached 93.38% where second-best method achieved only 57.65% on an average. Besides, even though when the number of rumors is high, the DBP could achieve on an average 68.01% reduction on the impact of rumors.