Deceptive Opinion Spam Detection Using Neural Network

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 140–150, Osaka, Japan, December 11-17 2016. Deceptive Opinion Spam Detection Using Neural Network Yafeng Ren and Yue Zhang Singapore University of Technology and Design, Singapore renyafeng@whu.edu.cn, yue zhang@sutd.edu.sg Abstract Deceptive opinion spam detection has attracted significant attention from both business and re- search communities. Existing approaches are based on manual discrete features, which can capture linguistic and psychological cues. However, such features fail to encode the semantic meaning of a document from the discourse perspective, which limits the performance. In this paper, we empirically explore a neural network model to learn document-level representation for detecting deceptive opinion spam. In particular, given a document, the model learns sentence representations with a convolutional neural network, which are combined using a gated recurrent neural network with attention mechanism to model discourse information and yield a documen- t vector. Finally, the document representation is used directly as features to identify deceptive opinion spam. Experimental results on three domains (Hotel, Restaurant, and Doctor) show that our proposed method outperforms state-of-the-art methods. 1 Introduction Online reviews on products and services are extensively used by consumers and businesses for conduct- ing decisive purchase, making product design and altering marketing strategies. As a result, deceptive opinion spam (e.g. deceptive reviews) arouses increasing attention (Streitfeld, 2012). Opinion spam is a type of review with fictitious opinions, deliberately written to sound authentic (Jindal and Liu, 2008; Ott et al., 2011). It can be difficult for human readers to distinguish them from truthful reviews. In a test by Ott et al. (2011), the average accuracy of three human judges is only 57.33%. It can be expensive to detect opinion spam manually over large user-generated texts. Hence, machine learning methods for automatically detecting deceptive opinion spam can be useful. The objective of the task is to identify whether a given document is a spam or not. The majority of existing approaches follow the seminal work of Jindal and Liu (2008), employing classifiers with supervised learning. Most studies focus on designing effective features to enhance the classification performance. Typical features represent linguistic and psychological cues, but fail to effectively represent a document from the viewpoint of global discourse structures. For example, Ott et al. (2011) and Li et al. (2014) represent documents with Unigram, POS and LIWC (Linguistic Inquiry and Word Count) (Newman et al., 2003) features. Although such features give the strong performance, their sparsity makes it difficult to capture non-local semantic information over a sentence or discourse. Recently, neural network models have been used to learn semantic representations for NLP tasks (Le and Mikolov, 2014; Tang et al., 2015), achieving highly competitive results. Potential advantages of using neural networks for spam detection are three-fold. First, neural models use dense hidden layers for automatic feature combinations, which can capture complex global semantic information that is difficult to express using traditional discrete manual features. This can be useful in addressing the limitation of discrete models mentioned above. Second, neural …

Content truncated to 500 words