Enhanced E-commerce Recommender System Based on Deep Learning and Ensemble Approaches
Ikram Karabila, Nossayba Darraz, Anas El-Ansari, Nabil Alami, Mostafa El Mallahi · 2024
The enhancement of e-commerce conversion rates heavily relies on personalized product recommendations generated by recommendation systems (RS). Despite successful techniques, challenges like sparse data and cold-start issues hinder their effectiveness. To tackle these obstacles, leveraging additional information sources like user and item profiles becomes crucial. This study focuses on a sophisticated RS architecture that combines Neural Collaborative Filtering (NCF) with Global vectors for word representation (GloVe) and Deep Neural Network (DNN), to address sparsity and cold-start problems. By integrating GloVe-DNN, a deeper understanding of item profiles is achieved, enabling a better grasp of user preferences and item traits. This detailed understanding, alongside the Neural Collaborative Filtering model, enables the system to offer highly personalized suggestions. The proposed method involves four main stages: creating the GloVe Embedding model to generate embeddings from item profile text, utilizing these vectors to build a Deep Neural Network, implementing the NCF method, and finally, developing an ensemble learning technique that combines these approaches, followed by a thorough evaluation. Empirical results strongly confirm the effectiveness of our approach, especially the fusion of GloVe-DNN with Neural CF techniques, demonstrating significant improvements across various performance measures.