Deep Q-Learning and Particle Swarm Optimization for Bot Detection in Online Social Networks

Social bot is a software program which preforms malicious activities, such as distributing spam content, phishing URLs and creating fake accounts. Moreover, social bots dynamically change their behavior by manipulating their trust value and pretend like a legitimate user in order to avoid detection. Hence, detecting social bots more accurately is an important task. In traditional Q-learning methods, Q-values have to be computed for all learning actions. As a result, convergence of Q-learning will be slow. In this paper, a particle swarm optimization (PSO) method is adopted to improve Q-learning in order to obtain an optimal sequence of learning actions. We define position of particle as a sequence of learning actions with respect to Q-values and the velocity of particle as state transition probability values for choosing a particular action. Further, a Deep Q-Learning algorithm based on Particle Swarm Optimization (DQL-PSO) has been proposed for social bot detection by considering updation strategy of PSO and Q-value function. The experimental results illustrate the efficacy of the proposed algorithm by considering the real-world Twitter dataset.