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

Reviews and ratings are essential components of e-business feedback. Since positive reviews entice customers for any business, websites tend to manipulate reviews to promote their business or demote their opponents. This paper presents a novel approach for identifying fake reviews by leveraging the power of transformer-based models, specifically BERT (Bidirectional Encoder Representations from Transformers), and the semantic embeddings from GloVe (Global Vectors for Word Representation). In this study, we fine-tune a pre-trained BERT model using a labelled review dataset, training it to capture intricate linguistic patterns indicative of fraudulent content. Simultaneously, we incorporate GloVe embeddings to enhance the model’s understanding of semantic relationships between words. Furthermore, we employed sentiment analysis over text to extract the truthfulness of the reviews using an attention-based transformer model. The premise of this work lies in the fact that the user with a genuine review would have true sentiments in it, while the fake review contains fabricated sentiments. We used the review text, and after applying pre-processing steps, we applied attention-based sequence to the sequence model, furthermore, we used transformers for sentimental analysis. We proposed a stacking method to combine all pre-trained models to get the ensemble result. We have compared our results with a traditional machine-learning algorithm. We have tested our algorithms on various datasets and achieved an accuracy of 98%.