English

Double-Calibration: Towards Reliable LLMs via Calibrating Knowledge and Reasoning Confidence

Computation and Language 2026-05-19 v2 Artificial Intelligence

Abstract

Reliable reasoning in Large Language Models (LLMs) is challenged by their propensity for hallucination. While augmenting LLMs with Knowledge Graphs (KGs) improves factual accuracy, existing KG-augmented methods fail to quantify epistemic uncertainty in both the retrieved evidence and LLMs' reasoning. To bridge this gap, we introduce DoublyCal, a framework built on a novel double-calibration principle. DoublyCal employs a lightweight proxy model to first generate KG evidence alongside a calibrated evidence confidence. This calibrated supporting evidence then guides a black-box LLM, yielding final predictions that are not only more accurate but also well-calibrated, with confidence scores traceable to the uncertainty of the supporting evidence. Experiments on knowledge-intensive benchmarks show that DoublyCal significantly improves both the accuracy and confidence calibration of black-box LLMs while maintaining low token cost.

Keywords

Cite

@article{arxiv.2601.11956,
  title  = {Double-Calibration: Towards Reliable LLMs via Calibrating Knowledge and Reasoning Confidence},
  author = {Yuyin Lu and Ziran Liang and Yanghui Rao and Wenqi Fan and Fu Lee Wang and Qing Li},
  journal= {arXiv preprint arXiv:2601.11956},
  year   = {2026}
}

Comments

This work is to appear in the Proceedings of the 35th International Joint Conference on Artificial Intelligence (IJCAI 2026)

R2 v1 2026-07-01T09:08:45.077Z