Identification of Deceptive Texts Using Cascade Classification

Online reviews of products, hotels, restaurants, and other services play an important role for both sellers and buyers. Through these reviews, potential buyers can get an idea of what to expect before making a purchase. However, not all reviews are authentic, as there are companies that pay their employees to generate reviews that discredit their competitors. To address this task, this study presents a computational method based on cascade classification that first automatically detects the distribution of latent emotions in the text. Subsequently, the emotion distribution vectors, in combination with lexical features, are used to identify signs of deception. Our experimental results demonstrate that the proposed method of this study show good performance on different datasets analyzed. In addition, our method automatically provides information about the emotions that influence the act of deception.