Neural networks for deceptive opinion spam detection: An empirical study

The products reviews are increasingly used by individuals and organizations for purchase and business decisions. Driven by the desire of profit, spammers produce synthesized reviews to promote some products or demote competitors products. So deceptive opinion spam detection has attracted significant attention from both business and research communities in recent years. Existing approaches mainly focus on traditional discrete features, which are based on linguistic and psychological cues. However, these methods fail to encode the semantic meaning of a document from the discourse perspective, which limits the performance. In this work, we empirically explore a neural network model to learn document-level representation for detecting deceptive opinion spam. First, the model learns sentence representation with convolutional neural network. Then, sentence representations are combined using a gated recurrent neural network, which can model discourse information and yield a document vector. Finally, the document representations are directly used as features to identify deceptive opinion spam. Based on three domains datasets, the results on in-domain and cross-domain experiments show that our proposed method outperforms state-of-the-art methods.