A Multi-type Classifier Ensemble for Detecting Fake Reviews Through Textual-based Feature Extraction

Abstract

The financial impact of online reviews has prompted some fraudulent sellers to generate fake consumer reviews for either promoting their products or discrediting competing products. In this study, we propose a novel ensemble model—the Multi-type Classifier Ensemble (MtCE)—combined with a textual-based featuring method, which is relatively independent of the system, to detect fake online consumer reviews. Unlike other ensemble models that utilise only the same type of single classifier, our proposed ensemble utilises several customised machine learning classifiers (including deep learning models) as its base classifiers. The results of our experiments show that the MtCE can adequately detect fake reviews, and that it outperforms other single and ensemble methods in terms of accuracy and other measurements for all the relevant public datasets used in this study. Moreover, if set correctly, the parameters of MtCE, such as base-classifier types, the total number of base classifiers, bootstrap, and the method to vote on output (e.g., majority or priority), can further improve the performance of the proposed ensemble.