Reinforcement Learning-Based Rumor Blocking Approach in Directed Social Networks
Qiang He, Yingjie Lv, Xingwei Wang, Min Huang, Yuliang Cai
Abstract
Social network platforms (such as Facebook, Wechat, and Weibo) can help people build relationships, transmit information, and make daily communication more convenient. However, in recent times, the rapid spread of misinformation and rumors has been causing public panic. Especially, during the epidemic, the severity of the crisis has been further exacerbated. Therefore, in this article, we study the influence minimization problem and propose a practical framework to address the rumor propagation problem. At first, we formulate the influence minimization problem as the mathematical optimization model. Then, we leverage the multistage competitive linear threshold model to reflect the activation status of network nodes. We propose a practical framework, called CCSQ, to select the seed nodes, which consists of community detection, candidate seed nodes, and the seeding algorithm with the Q-learning method. In particular, we construct the action, reward, and state of the Q-learning-based seeding algorithm to adaptively generate the seed nodes. Experimental results show that the proposed approach achieves smaller rumor propagation than the baseline algorithms.