Abstract
Large Language Models (LLMs) demonstrate significant capabilities but facechallenges such as hallucination, outdated knowledge, and non-transparent,untraceable reasoning processes. Augmented Generation (RAG) has emerged as apromising solution to these issues by incorporating real-time data fromexternal databases into LLM responses. This enhances the accuracy andcredibility of the models, particularly for knowledge-intensive tasks, andallows for continuous knowledge updates and integration of domain-specificinformation. RAG synergistically merges LLMs' intrinsic knowledge with thevast, dynamic repositories of external databases. This survey paper provides anin-depth analysis of the evolution of RAG, focusing on three key paradigms:Naive RAG, Advanced RAG, and Modular RAG. It methodically examines the threefundamental components of RAG systems: the retriever, the generator, and theaugmentation methods, underscoring the cutting-edge technologies within eachcomponenet. Additionally, the paper introduces novel metrics and capabilitiesfor evaluating RAG models, as well as the most recent evaluation framework.Finally, the paper outlines future research directions from three perspectives:future challenges,modality extension,and the development of the RAG technicalstack and ecosystem