A Comprehensive Survey of Deep Learning Models for Product Ranking

Abstract

Online customer reviews have a significant impact on consumer purchasing behavior in online marketplaces. Traditional product ranking approaches based on simple average ratings are not semantically rich or nuanced enough, nor do they capture opinion diversity, to be effectively applied to large scale review corpora. The recent development of deep learning and natural language processing (NLP) has revolutionized the way unstructured review data can be used to gain actionable insights. In this paper, we systematically review deep learning architectures used to solve review-based product ranking tasks, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long-short term memory networks (LSTMs) and gated recurrent units (GRUs), as well as bidirectional models and transformer-based models like BERT and DeBERTa. The mathematical formulation, representational capacity and empirical performance of each architecture are analyzed in the context of sentiment analysis and product ranking tasks. Hybrid models which incorporate sentiment signals with user credibility scores, cross-platform information fusion, and product attribute features are critically analyses. Key research gaps are systematically identified and quantified, such as sensitivity to adversarial reviews, limited model interpretability, scalability limitations, and ranking fairness, in addition to providing concrete research directions. This survey offers a structured roadmap to designing valid, interpretable and scalable next-generation product ranking systems.