A2C-Based Approach for Rumor Containment and Correction in Social Networks with a Time Delay

Social networks possess a dual nature, serving as platforms for rapid dissemination of innovations as well as widespread circulation of rumors. The proliferation of false information through rumors can potentially incite panic and pose risks. Therefore, measures must be taken to curb the spread of rumors. Rather than severing connections to block rumors, a more effective strategy involves promoting the dissemination of accurate information. Typically, the responsibility of monitoring and disseminating truth rests with official or authoritative entities. However, their influence remains relatively limited in comparison to the overall user base of social networks, resulting in a delay in detecting and addressing rumors. In this paper, a Linear Threshold model with a Time Delay (LTTD) is introduced to model the competitive spread of two different types of information within a shared social network. Specifically, leveraging deep reinforcement learning, the paper proposes an approach to minimize the negative effect of rumors (MNER) by training an A2C agent. Empirical findings from real networks demonstrate that the proposed approach surpasses other heuristic algorithms, effectively curtailing the propagation of rumors and promoting the dissemination of truth information.