Rumor Control Based on the RO- DQN Algorithm

Social problems can arise from the spread of malicious rumors on social networks such as Facebook, Twitter, and others. Once a rumor gains momentum and spreads quickly through a network, it is difficult to keep it under control. Discovering ways to minimize the spread of rumors within a social network is a major problem in information diffusion. A powerful approach to tackling the control rumor problem could be a deep reinforcement learning strategy that includes a wide range of components. In this regard, RO-DQN has been proposed as a stepwise discrete-time optimization method for rumor control to address the rumor influence minimization problem. The RO-DQN model aims to identify and select blocker nodes at each time step so that it can respond to the changing state of the network. This means that the strategy can be adapted based on the current situation in the network rather than relying on a static approach that does not take into account ongoing developments. Furthermore, sequential decision making is used, which involves a step-by-step optimization process. The main contribution of this work is the development of a new reward function that rewards the node that has a high centrality measure. The effect of the function is reduced by the number of already infected nodes connected to the target node. It is worth noting that the centrality measure is averaged from the page rank and the betweenness of the node. The other contribution is the proposal of a propagation model for modeling edge weight that considers the number of already infected nodes that have direct edges with the target node. The performance of the model is measured by the infection rate, and the model has been evaluated using real-world datasets: Facebook, Zachary Karate Club, Dolphins, and American Football. The model has made progress in performance compared to related work.