Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Retrieval-Augmented Text Generation for Large Language Models

Yizheng Huang, Jimmy Huang

TL;DR

Retrieval-Augmented Generation (RAG) tackles static knowledge limitations and hallucinations in large language models by grounding outputs in up-to-date external data. The paper presents a unified four-phase framework (pre-retrieval, retrieval, post-retrieval, generation) and surveys techniques across indexing, query manipulation, data modification, retrieval strategies, and generation enhancements. It analyzes evaluation paradigms, compares retrievers and generators, and discusses challenges in retrieval quality, system efficiency, and multimodal expansion, highlighting that retrieval quality often drives downstream performance. The findings emphasize iterative, adaptive, and multimodal RAG directions as promising paths for robust, knowledge-intensive AI applications with meaningful real-world impact.

Abstract

Retrieval-Augmented Generation (RAG) merges retrieval methods with deep learning advancements to address the static limitations of large language models (LLMs) by enabling the dynamic integration of up-to-date external information. This methodology, focusing primarily on the text domain, provides a cost-effective solution to the generation of plausible but possibly incorrect responses by LLMs, thereby enhancing the accuracy and reliability of their outputs through the use of real-world data. As RAG grows in complexity and incorporates multiple concepts that can influence its performance, this paper organizes the RAG paradigm into four categories: pre-retrieval, retrieval, post-retrieval, and generation, offering a detailed perspective from the retrieval viewpoint. It outlines RAG's evolution and discusses the field's progression through the analysis of significant studies. Additionally, the paper introduces evaluation methods for RAG, addressing the challenges faced and proposing future research directions. By offering an organized framework and categorization, the study aims to consolidate existing research on RAG, clarify its technological underpinnings, and highlight its potential to broaden the adaptability and applications of LLMs.

A Survey on Retrieval-Augmented Text Generation for Large Language Models

TL;DR

Retrieval-Augmented Generation (RAG) tackles static knowledge limitations and hallucinations in large language models by grounding outputs in up-to-date external data. The paper presents a unified four-phase framework (pre-retrieval, retrieval, post-retrieval, generation) and surveys techniques across indexing, query manipulation, data modification, retrieval strategies, and generation enhancements. It analyzes evaluation paradigms, compares retrievers and generators, and discusses challenges in retrieval quality, system efficiency, and multimodal expansion, highlighting that retrieval quality often drives downstream performance. The findings emphasize iterative, adaptive, and multimodal RAG directions as promising paths for robust, knowledge-intensive AI applications with meaningful real-world impact.

Abstract

Retrieval-Augmented Generation (RAG) merges retrieval methods with deep learning advancements to address the static limitations of large language models (LLMs) by enabling the dynamic integration of up-to-date external information. This methodology, focusing primarily on the text domain, provides a cost-effective solution to the generation of plausible but possibly incorrect responses by LLMs, thereby enhancing the accuracy and reliability of their outputs through the use of real-world data. As RAG grows in complexity and incorporates multiple concepts that can influence its performance, this paper organizes the RAG paradigm into four categories: pre-retrieval, retrieval, post-retrieval, and generation, offering a detailed perspective from the retrieval viewpoint. It outlines RAG's evolution and discusses the field's progression through the analysis of significant studies. Additionally, the paper introduces evaluation methods for RAG, addressing the challenges faced and proposing future research directions. By offering an organized framework and categorization, the study aims to consolidate existing research on RAG, clarify its technological underpinnings, and highlight its potential to broaden the adaptability and applications of LLMs.
Paper Structure (72 sections, 5 figures, 4 tables, 4 algorithms)

This paper contains 72 sections, 5 figures, 4 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. RAG Framework
  3. Basic RAG Workflow
  4. Indexing
  5. Retrieval
  6. Generation
  7. RAG Paradigm
  8. Pre-Retrieval
  9. Indexing
  10. Query Manipulation
  11. Data Modification
  12. Retrieval
  13. Search & Ranking
  14. Post-Retrieval
  15. Re-Ranking
  16. ...and 57 more sections

Figures (5)

  • Figure 1: An example of RAG benefits ChatGPT resolves questions that cannot be answered beyond the scope of the training data and generates correct results.
  • Figure 2: The unified RAG core concepts with basic workflow.
  • Figure 3: Taxonomy tree of RAG’s core techniques
  • Figure 4: An example of a typical RAG framework with interative retrieval strategy.
  • Figure 5: Retriever and generator experiment results sourced from eRAG salemi2024evaluating and BERGEN rau2024bergen.