AI-Generated Fake Review Detection

Abstract

Online reviews of e-commerce platforms have long been recognized as a major factor that influences a consumer’s purchasing decisions. However, the emergence of generative artificial intelligence (AI) has accelerated the proliferation of fake online reviews, which can significantly reduce consumer trust in these platforms. This study proposes a novel supervised learning approach to help platforms effectively detect AI-generated fake reviews. In the approach, we first construct three types of variables to distinguish between human-written genuine reviews and AI-generated fake reviews. Then, we introduce an outlier detection method based on cumulative probability density to calculate the probability that a fake review generated by AI. Finally, we train several well-known classification models using the cumulative probability density values of reviews computed above to obtain classifiers that can accurately detect AI-generated fake reviews. Numerical experiments demonstrate that the proposed method can produce more accurate detections of AI-generated fake review than several state-of-the-art baseline methods. We contribute to the related literature by the exploitation of the statistical theory, which posits that outliers, as small probability events, are typically located at the tails of feature distributions, a principle effectively employed in detecting AI-generated fake reviews.

Keywords: online reviews, generative AI, review manipulation, anomaly detection, supervised learning.