Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GE-Chat: A Graph Enhanced RAG Framework for Evidential Response Generation of LLMs

Longchao Da, Parth Mitesh Shah, Kuan-Ru Liou, Jiaxing Zhang, Hua Wei

TL;DR

The paper tackles LLM hallucination by introducing GE-Chat, a graph-enhanced RAG framework that grounds responses in evidence extracted from a knowledge graph built from user documents. The approach combines Graph-RAG construction, chain-of-thought reasoning elicitation, and efficient sub-graph searching with an entailment-based sentence optimization to select concise, source-backed evidence. A formal objective balances entailment probability and sentence length, enabling real-time, sentence-level evidence retrieval anchored to original chunks. Experimental results on a 1000-case, multi-domain dataset show GE-Chat improves evidential grounding across multiple LLMs, with GPT-4o delivering the strongest direct evidence, suggesting broad applicability for trustworthy AI-assisted decision making.

Abstract

Large Language Models are now key assistants in human decision-making processes. However, a common note always seems to follow: "LLMs can make mistakes. Be careful with important info." This points to the reality that not all outputs from LLMs are dependable, and users must evaluate them manually. The challenge deepens as hallucinated responses, often presented with seemingly plausible explanations, create complications and raise trust issues among users. To tackle such issue, this paper proposes GE-Chat, a knowledge Graph enhanced retrieval-augmented generation framework to provide Evidence-based response generation. Specifically, when the user uploads a material document, a knowledge graph will be created, which helps construct a retrieval-augmented agent, enhancing the agent's responses with additional knowledge beyond its training corpus. Then we leverage Chain-of-Thought (CoT) logic generation, n-hop sub-graph searching, and entailment-based sentence generation to realize accurate evidence retrieval. We demonstrate that our method improves the existing models' performance in terms of identifying the exact evidence in a free-form context, providing a reliable way to examine the resources of LLM's conclusion and help with the judgment of the trustworthiness.

GE-Chat: A Graph Enhanced RAG Framework for Evidential Response Generation of LLMs

TL;DR

The paper tackles LLM hallucination by introducing GE-Chat, a graph-enhanced RAG framework that grounds responses in evidence extracted from a knowledge graph built from user documents. The approach combines Graph-RAG construction, chain-of-thought reasoning elicitation, and efficient sub-graph searching with an entailment-based sentence optimization to select concise, source-backed evidence. A formal objective balances entailment probability and sentence length, enabling real-time, sentence-level evidence retrieval anchored to original chunks. Experimental results on a 1000-case, multi-domain dataset show GE-Chat improves evidential grounding across multiple LLMs, with GPT-4o delivering the strongest direct evidence, suggesting broad applicability for trustworthy AI-assisted decision making.

Abstract

Large Language Models are now key assistants in human decision-making processes. However, a common note always seems to follow: "LLMs can make mistakes. Be careful with important info." This points to the reality that not all outputs from LLMs are dependable, and users must evaluate them manually. The challenge deepens as hallucinated responses, often presented with seemingly plausible explanations, create complications and raise trust issues among users. To tackle such issue, this paper proposes GE-Chat, a knowledge Graph enhanced retrieval-augmented generation framework to provide Evidence-based response generation. Specifically, when the user uploads a material document, a knowledge graph will be created, which helps construct a retrieval-augmented agent, enhancing the agent's responses with additional knowledge beyond its training corpus. Then we leverage Chain-of-Thought (CoT) logic generation, n-hop sub-graph searching, and entailment-based sentence generation to realize accurate evidence retrieval. We demonstrate that our method improves the existing models' performance in terms of identifying the exact evidence in a free-form context, providing a reliable way to examine the resources of LLM's conclusion and help with the judgment of the trustworthiness.
Paper Structure (11 sections, 4 equations, 4 figures)

This paper contains 11 sections, 4 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Approach
  3. Graph-RAG construction
  4. CoT Reasoning Elicitation
  5. Efficient Sub-Graph Searching
  6. Evidence Content Optimization
  7. Problem Setting
  8. Experiment
  9. Experiment Setup
  10. Experiment Result
  11. Conclusion

Figures (4)

  • Figure 1: The overview of the GE-Chat framework. As shown in this pipeline, when user uploads the document, it is used for the 1. Graph-RAG Construction, which contains two main steps using LLMs, ➀ Extract the entities A, B, etc., from the document chunks, then ➁ Probe the contextual relations between these entities. Then a derived Knowledge Graph is formed and used for question answering. In order to realize evidence generation on Graph-RAG, 2. CoT Reasoning Elicitation is proposed to elicit the reasoning chain for answers. Then we have 3. Sub-Graph Searching based on Entity Matching, and N-hop Relations Probing, this sub-graph contains entities and relations used to retrieve the source chunks, guaranteeing originality of content. We then apply an 4. Optimize objective to balance meaningfulness and conciseness to get the high-quality evidence.
  • Figure 2: A Simple Abstraction for Sub-graph Searching.
  • Figure 3: The demonstration of the deployed GE-Chat framework.The user can upload PDF or relevant files, and the highlighted evidence comes along with the answers the LLMs made. For more examples please check the live demo video.
  • Figure 4: The comparison between the original LLMs evidence generation and LLMs with GE-Chat framework.