<![CDATA[The rapid growth of e-commerce presents challenges in delivering relevant product recommendations to users. This study develops a deep learning–based recommendation system by comparing the performance of Neural Collaborative Filtering (NCF) and Autoencoder models with the classical User-Based Collaborative Filtering approach using the RetailRocket dataset, which contains 2,756,101 user–product interactions. The research focuses on the application of negative sampling techniques to address the extremely high level of data sparsity. The experimental results show that NCF achieves the best performance, outperforming both the Autoencoder and the classical method in terms of Precision@10, Recall@10, and F1@10 metrics. The main contribution of this study lies in the application of NCF to a large-scale and highly sparse e-commerce dataset, demonstrating its superiority in handling extreme sparsity and producing more relevant and accurate recommendations. In addition, the study confirms the effectiveness of negative sampling techniques in improving recommendation prediction quality. These findings have theoretical implications by reinforcing the role of neural architectures in modern recommendation systems and practical implications for deploying more efficient and accurate models in real-world e-commerce platforms, potentially enhancing user experience and customer satisfaction.]]>
Copyrights © 2026
