Optimized Hybrid Deep Learning for Enhanced Spam Review Detection in E-Commerce Platforms

Abstract

Spam reviews represent a real danger to e-commerce platforms, steering consumers wrong and trashing the reputations of products. Conventional Machine learning (ML) methods are not capable of handling the complexity and scale of modern data. This study proposes the novel use of hybrid deep learning (DL) models for spam review detection and experiments with both CNN-LSTM and CNN-GRU architectures on the Amazon Product Review Dataset comprising 26.7 million reviews. One important finding is that 200k words vocabulary, with very little preprocessing improves the models a lot. Compared with other models, the CNN-LSTM model achieves the best performance with an accuracy of 92%, precision of 92.22%, recall of 91.73% and F1-score of 91.98%. This outcome emphasizes the effectiveness of using convolutional layers to extract local patterns and LSTM layers to capture long-term dependencies. The results also address how high constraints and hyperparameter search, as well as general-purpose represents such as BERT. Such advancements will help in creating more reliable and reliable spam detection systems to maintain consumer trust on e-commerce platforms.