A Novel E-commerce Recommendation System Based on RAG and Pretrained Large Model
Guohua Xiao, Jiongqian Wu, Shih-Pang Tseng · 2024
The exponential growth of e-commerce platforms necessitates efficient and scalable recommendation systems to enhance user experience and business performance. This paper introduces a novel recommendation system based on the Retrieval-Augmented Generation (RAG) framework, leveraging customer review data as an external, modular knowledge source. Unlike traditional systems that require global computations over user-product matrices, the proposed system employs FAISS for efficient vector-based retrieval and BART for natural language generation, reducing computational overhead and enhancing scalability. The system is implemented as a web-based application offering two functionalities: checking if a specific product is recommended based on reviews and generating product recommendations from user queries. Evaluation demonstrates the system’s ability to provide accurate, user-friendly recommendations while maintaining extensibility and adaptability to dynamic datasets. This work establishes a foundation for modular, review-driven recommendation systems in real-world e-commerce scenarios.