Identifying Authentic Reviews on Amazon Dataset using Transformer-Based Models

Reviews and ratings significantly impact businesses, either boosting their reputation or harming competitors. To manipulate these outcomes, some businesses falsify product reviews. This paper introduces a transformer-based model that utilizes GloVe (Global Vectors for Word Representation) semantic embeddings to detect fraudulent reviews. Our approach captures subtle linguistic patterns indicative of fake reviews by fine-tuning a pre-trained GloVe model on a labelled dataset. The GloVe embeddings improve the model’s grasp of word semantics, while an attention-based neural network classifies reviews as genuine or fraudulent. Our model has surpassed traditional machine learning methods, achieving 98% accuracy across various datasets.