Reinforcement Learning for Combating Cyberbullying in Online Social Networks

In recent decades, social network holds a pivotal role for people’s communication, but it is also a particularly susceptible to cyberbullying due to their rapid information dissemination capabilities. The state-of-art research in cyberbullying mainly focus on cyberbullying detection and literature discover, which encounters significant challenges in aspects like theoretical assurances, time effectiveness, and adaptability to broad contexts. In this paper, we present a resilient framework leveraging deep reinforcement learning (DRL) to tackle the problem of cyberbullying in online social networks. Our approach leverages dynamic graph neural networks to perform network embedding and the double deep Q-network (DDQN) for the parameter learning. To evaluate the effectiveness of our proposed approach, we conducted a comprehensive set of experiments using realistic datasets. The experimental findings demonstrate that our approach outperforms the comparison methods, even we train our model with small randomly generated ER graphs. This shows the strong generalization ability of our proposed model.