Online consumer review spam detection based reinforcement learning and neural network

Despite cutting-edge approaches for detecting spam reviews, there is still a lack of precision to classify new reviews since spammer strategy/behavior rapidly evolves and goes back over time. Most existing studies are based on static models that cannot detect spammers with new spam behaviors. This study tackles this challenge for the first time and proposes a Dynamic Spam Detection System (DSDS) to improve the detection model over time dynamically. The DSDS system integrates the neural network (NN) with reinforcement learning (RL) to identify spammer assaults in offline and online modes. In offline mode, the RL aims to identify the best NN architecture for the offline/static consumer review dataset. In the online mode, the DSDS system has the merit of handling the limited dataset problem by automatically adding new reviews to the offline dataset. Moreover, a novel feature selection-based algorithm is proposed to explore new spammer behaviors from the new reviews. The experiments that were conducted rigorously using well-known datasets demonstrated the ability of the system to detect new spam behaviors as well as to effectively classify online reviews by achieving a high accuracy rate and a low false-positive rate of 94.23%, and 0.0026%, respectively, for the YelpChi dataset, and of 97% and 0.0021, respectively, for the Amazon dataset. Moreover, a comparison with the state-of-the-art approaches on the same datasets proves the contribution of the proposed system.