Opinion Spam Detection with Attention-Based LSTM Networks

Zeinab Sedighi, Hossein Ebrahimpour-Komleh, Ayoub Bagheri, Leila Kosseim

Abstract: Today, online reviews have a great influence on consumers’ purchasing decisions. As a result, spam attacks, consisting of the malicious inclusion of fake online reviews, can be detrimental to both customers as well as organizations. Several methods have been proposed to automatically detect fake opinions; however, the majority of these methods focus on feature learning techniques based on a large number of handcrafted features. Deep learning and attention mechanisms have recently been shown to improve the performance of many classification tasks as they enable the model to focus on the most the important features. This paper describes our approach to apply LSTM and attention-based mechanisms for detecting deceptive reviews. Experiments with the Three-domain data set [15] show that a BiLSTM model coupled with Multi-Headed Self Attention improves the F-measure from 81.49% to 87.59% in detecting fake reviews.