Handling Imbalance in Fraudulent Reviewer Detection based on Expectation Maximization and KL Divergence

Online review fraud and review manipulation hurt the profits of stakeholders and undermine the value of online reviews. For this reason, it is critical to detect online review fraud and fraudulent reviewers effectively for the development of e-commerce. Extent studies propose various fraud detection techniques to detect fraudulent reviewers. However, most of these studies do not handle the data imbalance problem in fraudulent reviewer detection. To fill this research gap, this paper proposes a novel approach to detect fraudulent reviewers in handling the data imbalance based on Expectation Maximization (EM) and Kullback–Leibler (KL) divergence (called EMKL). We first use the expectation maximization algorithm to model the latent topic distributions of reviewers on the review features. Then, we adopt the Kullback–Leibler divergence to measure the similarities of reviewers based on their topic distributions to detect fraudulent reviewers. The experiment on Yelp dataset shows that the EMKL approach has a good performance in detecting fraudulent reviewers. In addition, the proposed EMKL method performs better than the performance of state-of-the-art techniques.