Retrieval-Augmented Generation With Large Language Models

Retrieval-Augmented Generation (RAG) has emerged as a transformative approach for enhanc-ing large language models (LLMs) by integrating external knowledge sources to address limitations in parametric knowledge and factual accuracy. The survey offers a detailed comparative study of RAG methods in dense retrieval, sparse retrieval, and hybrid approaches and their performance on knowledge-intensive tasks. Through systematic evaluation of 85 recent studies (2020-2024), we present empirical comparisons showing that hybrid RAG systems achieve 15-35% performance im-provements over baseline LLMs on question answering tasks, while dense retrieval methods demon-strate superior semantic understanding capabilities. We find that although RAG systems promise much, their current approach is limited by computational efficiency, retrieval precision, and stan-dardization of evaluation. We propose an integrated assessment system and outline five essential future avenues, including multi-modal RAG combination and adaptive retrieval schemes.