Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When to Trust: A Causality-Aware Calibration Framework for Accurate Knowledge Graph Retrieval-Augmented Generation

Jing Ren, Bowen Li, Ziqi Xu, Xinkun Zhang, Haytham Fayek, Xiaodong Li

TL;DR

Ca2KG tackles the overconfidence problem in Knowledge Graph Retrieval-Augmented Generation by introducing a causality-aware calibration framework. It combines counterfactual prompting to expose retrieval-dependent uncertainties with a panel-based re-scoring mechanism to stabilize predictions across interventions. The approach is evaluated on MetaQA and WebQSP, showing improved calibration (lower $ECE$ and $BS$) without sacrificing accuracy, and often higher $AUC$. The work demonstrates that integrating causal reasoning into KG-RAG yields more trustworthy, interpretable, and cost-efficient web-scale reasoning.

Abstract

Knowledge Graph Retrieval-Augmented Generation (KG-RAG) extends the RAG paradigm by incorporating structured knowledge from knowledge graphs, enabling Large Language Models (LLMs) to perform more precise and explainable reasoning. While KG-RAG improves factual accuracy in complex tasks, existing KG-RAG models are often severely overconfident, producing high-confidence predictions even when retrieved sub-graphs are incomplete or unreliable, which raises concerns for deployment in high-stakes domains. To address this issue, we propose Ca2KG, a Causality-aware Calibration framework for KG-RAG. Ca2KG integrates counterfactual prompting, which exposes retrieval-dependent uncertainties in knowledge quality and reasoning reliability, with a panel-based re-scoring mechanism that stabilises predictions across interventions. Extensive experiments on two complex QA datasets demonstrate that Ca2KG consistently improves calibration while maintaining or even enhancing predictive accuracy.

When to Trust: A Causality-Aware Calibration Framework for Accurate Knowledge Graph Retrieval-Augmented Generation

TL;DR

Ca2KG tackles the overconfidence problem in Knowledge Graph Retrieval-Augmented Generation by introducing a causality-aware calibration framework. It combines counterfactual prompting to expose retrieval-dependent uncertainties with a panel-based re-scoring mechanism to stabilize predictions across interventions. The approach is evaluated on MetaQA and WebQSP, showing improved calibration (lower and ) without sacrificing accuracy, and often higher . The work demonstrates that integrating causal reasoning into KG-RAG yields more trustworthy, interpretable, and cost-efficient web-scale reasoning.

Abstract

Knowledge Graph Retrieval-Augmented Generation (KG-RAG) extends the RAG paradigm by incorporating structured knowledge from knowledge graphs, enabling Large Language Models (LLMs) to perform more precise and explainable reasoning. While KG-RAG improves factual accuracy in complex tasks, existing KG-RAG models are often severely overconfident, producing high-confidence predictions even when retrieved sub-graphs are incomplete or unreliable, which raises concerns for deployment in high-stakes domains. To address this issue, we propose Ca2KG, a Causality-aware Calibration framework for KG-RAG. Ca2KG integrates counterfactual prompting, which exposes retrieval-dependent uncertainties in knowledge quality and reasoning reliability, with a panel-based re-scoring mechanism that stabilises predictions across interventions. Extensive experiments on two complex QA datasets demonstrate that Ca2KG consistently improves calibration while maintaining or even enhancing predictive accuracy.
Paper Structure (29 sections, 27 equations, 3 figures, 2 tables)

This paper contains 29 sections, 27 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Structural Causal Model
  4. Causal Intervention
  5. Methodology
  6. Problem Statement
  7. Causal Principles
  8. Counterfactual Prompting
  9. Panel-based Re-scoring
  10. Accurate Answer Generation
  11. Experimental Setup
  12. Datasets
  13. Baselines
  14. Backbone LLMs
  15. Metrics
  16. ...and 14 more sections

Figures (3)

  • Figure 1: Calibration comparison on the MetaQA dataset. Calibration error quantifies the average discrepancy between a model's predicted confidence and its actual correctness. The naive KG-RAG framework (a) is consistently over-confident with a high calibration error, whereas our Ca2KG framework (b) achieves much better calibration with a reduced error.
  • Figure 2: The overall architecture of Ca2KG. Given a query, the initial KG-RAG pipeline produces a baseline answer. Counterfactual prompting introduces interventions on retrieved paths, simulating quality and reasoning failures to generate alternative answers. Panel-based re-scoring evaluates all candidates under each prompt, forming a $3 \times N$ probability matrix. Finally, the Causal Calibration Index (CCI) combines support and stability across interventions to select the final calibrated answer.
  • Figure 3: Efficiency analysis on MetaQA. (a) Token usage per correct prediction in the 1-hop setting. (b) Accuracy under different token caps.

Theorems & Definitions (1)

  • Definition 1: Back-Door Criterion