Deep Learning Based Personalized Recommendation Systems for E-Commerce Platforms

Due to the faster development of e-commerce platform, the information overload can be considered overbearing, which allows promoting the need of the personalized recommendation systems that can provide the most precise and context-oriented product recommendations. Despite the showed strong predictive performance of deep learning techniques in comparison to the traditional method of recommendation, the current survey studies tend to offer partial analysis, constrained methodology, or out of date scopes that do not represent in detail the recent advances in architectures. The current investigation is a systematic review of e-commerce personalized recommendation systems based on deep learning, formulated in IEEE style, and summarizing the works published since 2018. Having used a structured literature selection protocol in the context of the large scholarly databases, the thematic classification and comparative analysis of convolutional neural networks, recurrent and sequential networks, Transformer-based models, graph neural networks, and hybrid models were conducted. The analysis discusses these methods in various aspects such as prediction accuracy, scalability, interpretability, sensitivity to data sparsity, fairness factor, and deployment practicability. The results suggest that deep learning models are always superior to traditional methods that model more and intricate user-item interactions and fusion with multi-modal data sources. Nonetheless, Explainability, cold-start treatment, and computability efficiency and ethical transparency continue to be problematic. The main contribution of the review is that it forms the coherent analytical framework, which consolidates the latest achievements, sets the trade-offs that are yet to be resolved, and outlines the research directions in terms of explainable, data-efficient, fair, and trustworthy recommenders at the next-generation e-commerce platforms.