Addressing Large-scale Reviewer Recruitment on Amazon: A Reviewer-centric Approach to the Fake Review Problem

Abstract

Fake reviews pose a significant challenge to the integrity of e-commerce platforms, undermining consumer trust and causing substantial economic harm. Despite advancements in machine learning-based detection methods, the emergence of organized recruitment schemes for fake reviews through private social media groups has made traditional approaches less effective. This study addresses the problem by proposing a novel reviewer-centric framework to detect and disrupt fake review activities. Using a dataset derived from Amazon and Facebook groups, we construct a reviewer network and develop an explainable Graph Neural Network (GNN) model that exploits relational patterns in reviewer behavior. Our findings demonstrate that incorporating network structures alongside behavioral and review features improves detection accuracy by approximately 10%. Recruited reviewers are found to form homogeneous ego networks, connecting predominantly with other fraudulent actors, whereas genuine reviewers exhibit more diverse connections. We show that removing reviews from identified recruited reviewers significantly reduces the ratings of review-buying products without harming genuine sellers, thereby disincentivizing fraudulent practices. These findings provide actionable insights for platforms, emphasizing the importance of targeting reviewers rather than individual reviews or products, and offer scalable interventions to mitigate the fake review problem while preserving consumer trust and market fairness.

Keywords: market for fake reviews, recruited reviewers, e-commerce, graph neural networks, explainable artificial intelligence, networks