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

2026-01-14

AI Summary

This work addresses the overconfidence of existing knowledge graph retrieval-augmented generation (KG-RAG) models when confronted with incomplete or unreliable subgraphs, a critical limitation in high-stakes applications. To mitigate this issue, the authors propose Ca2KG, a novel framework that introduces a causal reasoning perspective into KG-RAG. Ca2KG employs counterfactual prompting to expose the uncertainty inherent in retrieved evidence and incorporates an intervention-based panel rescaling mechanism to jointly calibrate both knowledge quality and reasoning confidence. Experimental results on two complex question-answering benchmarks demonstrate that Ca2KG substantially improves model calibration while maintaining or even enhancing predictive accuracy.

Technology Category

• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Knowledge Graph Retrieval-Augmented Generation
• overconfidence
• calibration
• reliability
• factual accuracy

Innovation

• Causality-aware Calibration
• Counterfactual Prompting
• Knowledge Graph Retrieval-Augmented Generation
• Overconfidence Mitigation
• Panel-based Re-scoring