RLOSD: Representation learning based opinion spam detection

Abstract

Nowadays, by vastly increasing in online reviews, harmful influence of spam reviews on decision making causes irrecoverable outcomes for both customers and organizations. Existing methods investigate for a way to contradistinction between spam and non-spam reviews. Most algorithms focus on feature engineering approaches to expose an accommodation of data representation. In this paper we propose a decision tree-based method to reveal deceptive reviews from trustworthy ones. We use unsupervised representation learning along with traditional feature selection methods to extract appropriate features and evaluate them with a decision tree. Our model takes data correlation into consideration to opt suitable features. The result shows the better performance in detecting opinion spam, comparing most common methods in this area.