Detecting Arabic Fake Reviews in E-commerce Platforms Using Machine and Deep Learning Approaches

Abstract

With the high spread of technology and e-commerce platforms, especially after the pandemic of COVID-19, customers increasingly rely on product reviews to assess their online choices. However, the usefulness of online reviews can be hindered by fake reviews. Therefore, detection of fake reviews is needed. Unfortunately, the number of studies on the automatic detection of fake reviews is limited. This paper is one of the very few works attempting to detect fake reviews written in Arabic and, to the best of our knowledge, the first paper to evaluate the deep learning architecture for this challenging task.   Most reported studies focused on English reviews with little attention to other languages. Thus, this research paper aims to experiment with Arabic fake reviews and investigate how they can be automatically detected using machine and deep learning approaches. Due to the unavailability of the Arabic fake reviews dataset, we used the Amazon e-commerce dataset after translating them into Arabic; first, we have evaluated some traditional algorithms, including logistic regression, decision tree, K-nearest neighbors, and support vector machine (SVM), and compared the results with other state-of-the-art approaches such as Gradient boosting classifier, Random Forest classifier and deep learning structures; AraBERT. Among the traditional methods, the results showed that SVM achieved the highest accuracy of 87.61%. However, AraBERT significantly outperformed the SVM and achieved 93.00 % accuracy in detecting Arabic fake reviews.