Opinion Spam Detection in Online Reviews Using Neural Networks

Abstract

People of the Internet era usually rely on online reviews to make decisions about an online purchase, a hotel booking or a car rental and much more, since people believe that making decisions based on other’s opinions lead to making the right choice. As writing fake reviews come with monetary gain, the opinion spam activities have increased dramatically on online review websites. Thus opinion spam in reviews has become a big challenge to people to make purchase decisions as well as damages the reputation of review websites. Hence the deceptive opinion spam detection is an essential task in the field of natural language processing. Most of the existing research on opinion spam detection uses the traditional bag-of-words model to represent the review text features and apply standard machine learning models such as Support-Vector Machines or Naive Bayes as classifiers. There is only a few state of the art methods, which have utilized neural network based methods for spam detection. With the recent advancements in deep learning, there are successful applications of Convolutional Neural Networks (CNN) for natural language processing problems which have achieved improved performance. In this study, a CNN model is developed to detect opinion spam using the features extracted from the pre-trained GloVe: Global Vectors for Word Representation model. Moreover, some word and character level features used in existing research work, are extracted from the text and concatenated with a feature set extracted by the convolutional layers of the model to improve the performance. The proposed model found to outperform the state-of-the-art method and the inclusion of additional features improved the performance further.