From knowledge retrieval to generation: a comprehensive study of RAG's architectures, evolution, and applications

Abstract

Abstract Large Language Models (LLMs) have demonstrated strong capabilities in natural language understanding, generation, and reasoning, but their reliance on static parametric knowledge still leads to knowledge staleness, hallucinations, limited domain adaptability, weak traceability, and high updating costs in knowledge-intensive scenarios. Retrieval-Augmented Generation (RAG) helps address these limitations by incorporating external knowledge sources into the generation process, making it an important approach for improving factual grounding, knowledge timeliness, and evidence traceability. However, existing RAG research is increasingly fragmented across architectures, retrieval and generation strategies, evaluation methods, application scenarios, deployment constraints, and security risks. To provide a systematic understanding of this field, this survey reviews RAG research through a structured and transparent review protocol. We organize the review around three major dimensions: the system stack and paradigm evolution of RAG, evaluation methods and application evidence, and practical deployment and security challenges. Specifically, we analyze the core components of RAG systems, synthesize their evolution from Naive RAG to Agentic RAG, examine evaluation metrics and stress-testing protocols, summarize benchmark datasets and domain applications, and discuss deployment constraints and emerging risks such as prompt injection, data poisoning, retrieval hijacking, privacy leakage, and citation laundering. This survey provides a taxonomy-oriented, evaluation-aware, and deployment-conscious reference for understanding the development, limitations, and future directions of RAG systems.