Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Retrieval to Generation: Unifying External and Parametric Knowledge for Medical Question Answering

Lei Li, Xiao Zhou, Yingying Zhang, Xian Wu

TL;DR

This work addresses the reliability challenges in medical QA by unifying external (retrieved) and parametric (generated) knowledge through a retrieval–generation–then–read framework called MedRGAG. It introduces two core modules: Knowledge-Guided Context Completion (KGCC) to generate complementary background knowledge and Knowledge-Aware Document Selection (KADS) to adaptively fuse retrieved and generated evidence. Through source-balanced retrieval and multi-step reasoning, MedRGAG achieves significant improvements over both RAG and GAG baselines across five medical QA benchmarks, demonstrating that integrated evidence supports more accurate and trustworthy answers. The approach reduces hallucinations by grounding generated content with retrieved evidence and selects concise, diverse evidence tailored to each question, with strong ablations and case studies confirming the contributions of KGCC and KADS.

Abstract

Medical question answering (QA) requires extensive access to domain-specific knowledge. A promising direction is to enhance large language models (LLMs) with external knowledge retrieved from medical corpora or parametric knowledge stored in model parameters. Existing approaches typically fall into two categories: Retrieval-Augmented Generation (RAG), which grounds model reasoning on externally retrieved evidence, and Generation-Augmented Generation (GAG), which depends solely on the models internal knowledge to generate contextual documents. However, RAG often suffers from noisy or incomplete retrieval, while GAG is vulnerable to hallucinated or inaccurate information due to unconstrained generation. Both issues can mislead reasoning and undermine answer reliability. To address these challenges, we propose MedRGAG, a unified retrieval-generation augmented framework that seamlessly integrates external and parametric knowledge for medical QA. MedRGAG comprises two key modules: Knowledge-Guided Context Completion (KGCC), which directs the generator to produce background documents that complement the missing knowledge revealed by retrieval; and Knowledge-Aware Document Selection (KADS), which adaptively selects an optimal combination of retrieved and generated documents to form concise yet comprehensive evidence for answer generation. Extensive experiments on five medical QA benchmarks demonstrate that MedRGAG achieves a 12.5% improvement over MedRAG and a 4.5% gain over MedGENIE, highlighting the effectiveness of unifying retrieval and generation for knowledge-intensive reasoning. Our code and data are publicly available at https://anonymous.4open.science/r/MedRGAG

From Retrieval to Generation: Unifying External and Parametric Knowledge for Medical Question Answering

TL;DR

This work addresses the reliability challenges in medical QA by unifying external (retrieved) and parametric (generated) knowledge through a retrieval–generation–then–read framework called MedRGAG. It introduces two core modules: Knowledge-Guided Context Completion (KGCC) to generate complementary background knowledge and Knowledge-Aware Document Selection (KADS) to adaptively fuse retrieved and generated evidence. Through source-balanced retrieval and multi-step reasoning, MedRGAG achieves significant improvements over both RAG and GAG baselines across five medical QA benchmarks, demonstrating that integrated evidence supports more accurate and trustworthy answers. The approach reduces hallucinations by grounding generated content with retrieved evidence and selects concise, diverse evidence tailored to each question, with strong ablations and case studies confirming the contributions of KGCC and KADS.

Abstract

Medical question answering (QA) requires extensive access to domain-specific knowledge. A promising direction is to enhance large language models (LLMs) with external knowledge retrieved from medical corpora or parametric knowledge stored in model parameters. Existing approaches typically fall into two categories: Retrieval-Augmented Generation (RAG), which grounds model reasoning on externally retrieved evidence, and Generation-Augmented Generation (GAG), which depends solely on the models internal knowledge to generate contextual documents. However, RAG often suffers from noisy or incomplete retrieval, while GAG is vulnerable to hallucinated or inaccurate information due to unconstrained generation. Both issues can mislead reasoning and undermine answer reliability. To address these challenges, we propose MedRGAG, a unified retrieval-generation augmented framework that seamlessly integrates external and parametric knowledge for medical QA. MedRGAG comprises two key modules: Knowledge-Guided Context Completion (KGCC), which directs the generator to produce background documents that complement the missing knowledge revealed by retrieval; and Knowledge-Aware Document Selection (KADS), which adaptively selects an optimal combination of retrieved and generated documents to form concise yet comprehensive evidence for answer generation. Extensive experiments on five medical QA benchmarks demonstrate that MedRGAG achieves a 12.5% improvement over MedRAG and a 4.5% gain over MedGENIE, highlighting the effectiveness of unifying retrieval and generation for knowledge-intensive reasoning. Our code and data are publicly available at https://anonymous.4open.science/r/MedRGAG
Paper Structure (27 sections, 10 equations, 4 figures, 5 tables, 1 algorithm)

This paper contains 27 sections, 10 equations, 4 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Medical Question Answering
  4. Medical Retrieval-Augmented Generation
  5. Medical Generation-Augmented Generation
  6. Methodology
  7. Problem Formulation
  8. MedRGAG
  9. Source-Balanced Evidence Retrieval
  10. Knowledge-Guided Context Completion
  11. Step 1: Summarization of Retrieved Knowledge.
  12. Step 2: Exploration of Missing Knowledge.
  13. Step 3: Generation of Background Documents.
  14. Knowledge-Aware Document Selection
  15. Experiments
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Comparison of MedRGAG with standard RAG and GAG. Figure (a) illustrates RAG, where the retriever extracts external knowledge to enhance the reader’s answer generation. Figure (b) depicts GAG, where the generator activates its internal parametric knowledge to assist the reader in producing answers. Figure (c) presents MedRGAG, which enables the retriever and generator to jointly unify external and parametric knowledge, providing comprehensive evidence to enhance the reader’s answer reliability.
  • Figure 2: Framework of MedRGAG, which comprises three stages: source-balanced evidence retrieval, knowledge-guided context completion, and knowledge-aware document selection.
  • Figure 3: Results of different model scales on the Qwen2.5-7B-Instruct reader. Figures (a) and (b) show different LLM scales used as generators and summarizer-explorer-integrator.
  • Figure 4: Results of different document selection strategies on the Qwen2.5-7B-Instruct reader. Figure (a) shows different document selection strategies on accuracy. Figure (b) shows the proportion of retrieved documents within the final top-5 across different selection strategies.