Hybrid Visual-Textual Product Recommendation System for E-Commerce Platforms

Abstract

Today, e-commerce sites provide a large number of products to users. However, presenting the right products to users is important for both customer satisfaction and increasing company revenues. Recommendation systems are systems that offer personalized product suggestions by analyzing user preferences and behaviors. This study presents a novel hybrid product recommendation system that integrates collaborative filtering and content-based filtering methods, enhanced by deep learning techniques. By using both visual and textual product features through BERT and CLIP models, our system addresses cold-start problem and real-time performance constraints. The system has been successfully deployed on the Cimri e-commerce platform, providing personalized recommendations that adapt to evolving user preferences while maintaining computational efficiency.