Do the Explanations Make Sense? Explainable Fake Review Identification and Users’ Perspectives on Explanations

Customer reviews and feedback play a crucial role in shaping purchase decisions on e-commerce platforms like Amazon, Zalando, and eBay. However, a major concern is the prevalence of fake or spam reviews, often posted by sellers to deceive potential customers and manipulate product perceptions. Machine learning (ML) models are widely used to detect fraudulent reviews, but their decisions can be difficult to interpret due to their complexity—often functioning as black boxes. In this paper, we propose an explainable framework for fake review detection that not only achieves high precision in identifying fraudulent content but also provides interpretable explanations. To assess the effectiveness of these explanations, we conduct an empirical user evaluation to determine which information is most valuable in understanding model decisions. Initially, we develop fake review detection models using deep learning (DL) and transformer-based architectures, including XLNet and DistilBERT. Wethen apply Layer-wise Relevance Propagation (LRP) to generate explanations by mapping word contributions to the predicted class. Experimental results on two benchmark fake review detection datasets demonstrate that our models achieve state-of-the-art performance, outperforming several existing methods. Furthermore, we conduct a user study with 12 participants to evaluate the comprehensibility and usefulness of LRP-generated explanations. The findings from this study provide key insights into how explanations can be improved to enhance transparency and user trust in fake review detection systems.