Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RAG-KG-IL: A Multi-Agent Hybrid Framework for Reducing Hallucinations and Enhancing LLM Reasoning through RAG and Incremental Knowledge Graph Learning Integration

Hong Qing Yu, Frank McQuade

TL;DR

RAG-KG-IL addresses the hallucination and reasoning gaps in LLMs by integrating retrieval-augmented generation with structured knowledge graphs and an incremental learning protocol within a multi-agent framework. The architecture supports continuous knowledge updates, explicit KG reasoning via RDFLib, and explainable outputs through reasoning graphs, validated in healthcare-focused case studies against baselines. Key contributions include a detailed hybrid pipeline, an incremental KG growth mechanism with human-in-the-loop validation, and a modified accuracy evaluation protocol that emphasizes truth-checking and completeness. The results show substantial hallucination reductions and improved reasoning and completeness, demonstrating the practical potential of real-time knowledge integration in high-stakes domains. The work lays a foundation for scalable, domain-aware intelligent systems capable of grounded, explainable reasoning with dynamic knowledge.

Abstract

This paper presents RAG-KG-IL, a novel multi-agent hybrid framework designed to enhance the reasoning capabilities of Large Language Models (LLMs) by integrating Retrieval-Augmented Generation (RAG) and Knowledge Graphs (KGs) with an Incremental Learning (IL) approach. Despite recent advancements, LLMs still face significant challenges in reasoning with structured data, handling dynamic knowledge evolution, and mitigating hallucinations, particularly in mission-critical domains. Our proposed RAG-KG-IL framework addresses these limitations by employing a multi-agent architecture that enables continuous knowledge updates, integrates structured knowledge, and incorporates autonomous agents for enhanced explainability and reasoning. The framework utilizes RAG to ensure the generated responses are grounded in verifiable information, while KGs provide structured domain knowledge for improved consistency and depth of understanding. The Incremental Learning approach allows for dynamic updates to the knowledge base without full retraining, significantly reducing computational overhead and improving the model's adaptability. We evaluate the framework using real-world case studies involving health-related queries, comparing it to state-of-the-art models like GPT-4o and a RAG-only baseline. Experimental results demonstrate that our approach significantly reduces hallucination rates and improves answer completeness and reasoning accuracy. The results underscore the potential of combining RAG, KGs, and multi-agent systems to create intelligent, adaptable systems capable of real-time knowledge integration and reasoning in complex domains.

RAG-KG-IL: A Multi-Agent Hybrid Framework for Reducing Hallucinations and Enhancing LLM Reasoning through RAG and Incremental Knowledge Graph Learning Integration

TL;DR

RAG-KG-IL addresses the hallucination and reasoning gaps in LLMs by integrating retrieval-augmented generation with structured knowledge graphs and an incremental learning protocol within a multi-agent framework. The architecture supports continuous knowledge updates, explicit KG reasoning via RDFLib, and explainable outputs through reasoning graphs, validated in healthcare-focused case studies against baselines. Key contributions include a detailed hybrid pipeline, an incremental KG growth mechanism with human-in-the-loop validation, and a modified accuracy evaluation protocol that emphasizes truth-checking and completeness. The results show substantial hallucination reductions and improved reasoning and completeness, demonstrating the practical potential of real-time knowledge integration in high-stakes domains. The work lays a foundation for scalable, domain-aware intelligent systems capable of grounded, explainable reasoning with dynamic knowledge.

Abstract

This paper presents RAG-KG-IL, a novel multi-agent hybrid framework designed to enhance the reasoning capabilities of Large Language Models (LLMs) by integrating Retrieval-Augmented Generation (RAG) and Knowledge Graphs (KGs) with an Incremental Learning (IL) approach. Despite recent advancements, LLMs still face significant challenges in reasoning with structured data, handling dynamic knowledge evolution, and mitigating hallucinations, particularly in mission-critical domains. Our proposed RAG-KG-IL framework addresses these limitations by employing a multi-agent architecture that enables continuous knowledge updates, integrates structured knowledge, and incorporates autonomous agents for enhanced explainability and reasoning. The framework utilizes RAG to ensure the generated responses are grounded in verifiable information, while KGs provide structured domain knowledge for improved consistency and depth of understanding. The Incremental Learning approach allows for dynamic updates to the knowledge base without full retraining, significantly reducing computational overhead and improving the model's adaptability. We evaluate the framework using real-world case studies involving health-related queries, comparing it to state-of-the-art models like GPT-4o and a RAG-only baseline. Experimental results demonstrate that our approach significantly reduces hallucination rates and improves answer completeness and reasoning accuracy. The results underscore the potential of combining RAG, KGs, and multi-agent systems to create intelligent, adaptable systems capable of real-time knowledge integration and reasoning in complex domains.

Paper Structure

This paper contains 21 sections, 12 equations, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Hallucination in Generative AI
  4. RAG-LLM and Knowledge Graph-based LLM Solutions and Their Current Limitations
  5. Continual Learning Challenges in LLM and the Importance of Incremental Learning
  6. Proposed Framework
  7. Application Interface
  8. LLM-Agent (RAG-KG-IL Answer Generation Engine)
  9. LLM-Agent (Data Selection and Text Extraction Using LangChain)
  10. LLM-Agent (KG Generation and Reasoning Using RDFLib)
  11. Knowledge Graph Repository incrementally growth and Human-in-the-Loop Knowledge Addition
  12. Evaluation Methodology
  13. Knowledge Creation and Causality Reasoning Capabilities Evaluation Method
  14. Scalability of time cost Evaluation Method
  15. Initial Design of Accuracy Evaluation Method
  16. ...and 6 more sections

Figures (9)

  • Figure 1: The proposed RAG-KG-IL framework
  • Figure 2: Interface view, left: question and answer including sources, middle: subset of knowledge graph related to the question, right: causality reasoning analysis.
  • Figure 3: Knowledge Data Growing from Graph A to Graph B. The growth shows the increase in terms, triples, and unique relation types as new questions were answered.
  • Figure 4: Knowledge reasoning capability applied to Pneumonia causes after only two questions.
  • Figure 5: Knowledge reasoning capability applied to Pneumonia causes after 20 questions.
  • ...and 4 more figures