A transformer-based ensemble algorithm for fake review detection in e-commerce systems

E-commerce platforms rely heavily on customer reviews and ratings to build trust and influence consumer behavior. However, some businesses may manipulate these reviews to gain an unfair advantage or to discredit their competitors. This paper introduces a cutting-edge method for detecting fake reviews by harnessing the capabilities of transformer-based models, specifically BERT and GloVe, and incorporating sentiment analysis.

The researchers fine-tuned a pre-trained BERT model using a labeled dataset of reviews, enabling it to recognize complex linguistic patterns often found in fraudulent content. They also integrated GloVe embeddings to enhance the model's understanding of the semantic relationships between words. Additionally, they applied sentiment analysis using an attention-based transformer model to assess the authenticity of the reviews based on the sentiment expressed.

The key insight behind this approach is that genuine reviews will convey true sentiments, whereas fake reviews will contain fabricated emotions. To optimize the detection process, the researchers employed a stacking method, combining the outputs of multiple pre-trained models to generate an ensemble result. They then compared their findings with those of a traditional machine-learning algorithm.

Extensive testing on various datasets yielded impressive results, with the ensemble model achieving an impressive accuracy of 98%. This study was supported by Research Supporting Project Number ORF-2026-553, funded by King Saud University in Riyadh, Saudi Arabia. The research acknowledges equal contributions from Musaed Alhussein, Khursheed Aurangzeb, Waqar Ahmad, Hadiqa Azam, and Muhammad Majid, all from esteemed institutions in Pakistan and Italy.

Correspondence regarding this work should be directed to Hassan Nazeer Chaudhry or Muhammad Majid. This research is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which allows for non-commercial use and sharing, provided proper attribution is given to the original authors and source.