Detecting Fake Reviews with Generative Adversarial Networks for Mobile Social Networks

Abstract

With the growth of mobile social networks (MSNs), crowdsourced information could be used for recommendation to mobile users. However, it is quite vulnerable to Sybil attacks, where attackers post fake information or reviews to mislead users for business benefits. To address this problem, existing detection models mainly use graph-based techniques or extract features of users. However, these approaches either rely on strong assumptions or lack generalization. Therefore, we propose a novel Sybil detection model based on generative adversarial networks (GANs), which contains a feature extractor, a domain classifier, and a Sybil detector. First, the feature extractor is proposed to identify the rich information in the review text with the neural network model of TextCNN. Second, the domain classifier is implemented by a neural network discriminator and is able to extract common features. Third, the Sybil detector is utilized to discriminate the fake review. Finally, the minimax game between the domain classifier and Sybil detector forms a GAN and enhances the overall generalization ability of the model. Extensive experiments show that our model has a high detection accuracy against Sybil attacks.