Online product decision support using review analysis and Transformer-GRU hybrid model through e-commerce platform

The rapid expansion of e-commerce platforms has generated an enormous volume of user-written product reviews, creating a growing need for intelligent decision-support mechanisms that can interpret complex and unstructured feedback. Traditional recommendation approaches such as Support Vector Machines, Naïve Bayes classifiers, and Recurrent Neural Networks have been widely used for sentiment analysis and rating prediction, yet they continue to struggle with cold-start issues, limited interpretability, and an inability to effectively capture long-range dependencies in lengthy reviews. To address these limitations, this study introduces a hybrid Transformer–GRU (Gated Recurrent Unit) model designed to enhance the accuracy and robustness of product recommendations. The Transformer module leverages multi-head self-attention to extract deep contextual information from reviews, while the GRU component efficiently models sequential dependencies and evolving user behavior over time. Additionally, the integration of BERT embeddings, TF–IDF scores, and Word2Vec representations enriches semantic understanding and improves feature expressiveness. Experiments conducted on the Kaggle Women’s Clothing E-Commerce Reviews dataset demonstrate that the proposed model achieves superior performance, yielding a low MAE of 0.02 and RMSE of 0.07. These results highlight the effectiveness of the Transformer–GRU hybrid architecture in delivering accurate and personalized e-commerce recommendations.