Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FiDeLiS: Faithful Reasoning in Large Language Model for Knowledge Graph Question Answering

Yuan Sui, Yufei He, Nian Liu, Xiaoxin He, Kun Wang, Bryan Hooi

TL;DR

FiDeLiS tackles LLM hallucinations in knowledge graph–based QA by grounding responses in verifiable KG reasoning steps. It combines Path-RAG, a retrieval-augmented module that pre-selects high-quality candidate steps, with DVBS, a deductive, plan-guided beam search that validates each step. The training-free framework achieves superior factuality and interpretability across multiple KGQA benchmarks, while reducing computation through constrained search. Its robustness across domains and potential for faster large-language models underscore its practical impact for scalable, trustworthy KG reasoning.

Abstract

Large Language Models (LLMs) are often challenged by generating erroneous or hallucinated responses, especially in complex reasoning tasks. Leveraging Knowledge Graphs (KGs) as external knowledge sources has emerged as a viable solution. However, existing KG-enhanced methods, either retrieval-based or agent-based, encounter difficulties in accurately retrieving knowledge and efficiently traversing KGs at scale. In this paper, we propose a unified framework, FiDeLiS, designed to improve the factuality of LLM responses by anchoring answers to verifiable reasoning steps retrieved from KGs. To achieve this, we leverage step-wise beam search with a deductive scoring function, allowing the LLM to validate reasoning process step by step, and halt the search once the question is deducible. In addition, we propose a Path-RAG module to pre-select a smaller candidate set for each beam search step, reducing computational costs by narrowing the search space. Extensive experiments show that our method, as a training-free framework, not only improve the performance but also enhance the factuality and interpretability across different benchmarks. Code is released at https://github.com/Y-Sui/FiDeLiS.

FiDeLiS: Faithful Reasoning in Large Language Model for Knowledge Graph Question Answering

TL;DR

FiDeLiS tackles LLM hallucinations in knowledge graph–based QA by grounding responses in verifiable KG reasoning steps. It combines Path-RAG, a retrieval-augmented module that pre-selects high-quality candidate steps, with DVBS, a deductive, plan-guided beam search that validates each step. The training-free framework achieves superior factuality and interpretability across multiple KGQA benchmarks, while reducing computation through constrained search. Its robustness across domains and potential for faster large-language models underscore its practical impact for scalable, trustworthy KG reasoning.

Abstract

Large Language Models (LLMs) are often challenged by generating erroneous or hallucinated responses, especially in complex reasoning tasks. Leveraging Knowledge Graphs (KGs) as external knowledge sources has emerged as a viable solution. However, existing KG-enhanced methods, either retrieval-based or agent-based, encounter difficulties in accurately retrieving knowledge and efficiently traversing KGs at scale. In this paper, we propose a unified framework, FiDeLiS, designed to improve the factuality of LLM responses by anchoring answers to verifiable reasoning steps retrieved from KGs. To achieve this, we leverage step-wise beam search with a deductive scoring function, allowing the LLM to validate reasoning process step by step, and halt the search once the question is deducible. In addition, we propose a Path-RAG module to pre-select a smaller candidate set for each beam search step, reducing computational costs by narrowing the search space. Extensive experiments show that our method, as a training-free framework, not only improve the performance but also enhance the factuality and interpretability across different benchmarks. Code is released at https://github.com/Y-Sui/FiDeLiS.
Paper Structure (40 sections, 3 equations, 6 figures, 12 tables)

This paper contains 40 sections, 3 equations, 6 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Method
  4. Path-RAG: Reasoning Path Retrieval-Augmented Generation
  5. Initialization.
  6. Keyword-Driven Retrieval.
  7. Reasoning Step Candidates Construction.
  8. Deductive-Verification Beam Search
  9. Plan Generation.
  10. Beam Search.
  11. Deductive Verification.
  12. Experiments
  13. Main Results
  14. Ablation Study
  15. Robustness Analysis
  16. ...and 25 more sections

Figures (6)

  • Figure 1: Challenges for existing KG-enhanced methods: How to balance faithfulness and efficiency?
  • Figure 2: An illustration of FiDeLiS. Top: The workflow of Path-RAG. An LLM first extracts key terms and generates dense embeddings that feed into the Path-RAG module. Then, Path-RAG rapidly retrieves relevant entities and relations from a pre-embedded KG and constructs candidate reasoning steps by combining semantic similarity with graph connectivity. Bottom: The workflow of DVBS. Next, the DVBS module uses LLM-generated planning to guide a beam search that builds reasoning paths step-by-step over candidates constructed by Path-RAG, with deductive verification ensuring each step logically follows the previous steps and support the user question.
  • Figure 3: Analysis of reasoning errors in RoG luo2024reasoninggraphsfaithful over WebQSP.
  • Figure 4: Distribution of CR-LT-KGQA dataset.
  • Figure 5: Parameter tuning for $\alpha$ for scoring function over WebQSP
  • ...and 1 more figures