Graph-based Confidence Calibration for Large Language Models
Abstract: Reliable confidence estimation is essential for enhancing the trustworthiness of large language models (LLMs), especially in high-stakes scenarios. Despite its importance, accurately estimating confidence in LLM responses remains a significant challenge. In this work, we propose using an auxiliary learning model to assess response correctness based on the self-consistency of multiple outputs generated by the LLM. Our approach constructs a consistent graph to capture the agreement among different responses and employs a graph neural network (GNN) to predict the correctness likelihood of each answer based on the consistent graph. Experiments demonstrate that this method has strong calibration performance on various benchmark datasets and generalizes well to out-of-domain cases.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Tal_Schuster1
Submission Number: 4032
Loading