Confidence Estimation in Automatic Short Answer Grading with LLMs

TL;DR

This paper explores confidence estimation methods for LLM-based automatic short answer grading, proposing a hybrid approach that combines model signals with dataset-derived uncertainty to improve reliability.

Contribution

It introduces a hybrid confidence framework that integrates model-based signals with dataset-derived uncertainty, enhancing confidence estimation in LLM-based grading.

Findings

Hybrid confidence measure improves reliability over single-source methods.

Clustering semantically embedded responses quantifies response heterogeneity.

Proposed approach enhances trustworthiness in AI-assisted educational assessment.

Abstract

Automatic Short Answer Grading (ASAG) with generative large language models (LLMs) has recently demonstrated strong performance without task-specific fine-tuning, while also enabling the generation of synthetic feedback for educational assessment. Despite these advances, LLM-based grading remains imperfect, making reliable confidence estimates essential for safe and effective human-AI collaboration in educational decision-making. In this work, we investigate confidence estimation for ASAG with LLMs by jointly considering model-based confidence signals and dataset-derived uncertainty. We systematically compare three model-based confidence estimation strategies, namely verbalizing, latent, and consistency-based confidence estimation, and show that model-based confidence alone is insufficient to reliably capture uncertainty in ASAG. To address this limitation, we propose a hybrid confidence framework that integrates model-based confidence signals with an explicit estimate of dataset-derived aleatoric uncertainty. Aleatoric uncertainty is operationalized by clustering semantically embedded student responses and quantifying within-cluster heterogeneity. Our results demonstrate that the proposed hybrid confidence measure yields more reliable confidence estimates and improves selective grading performance compared to single-source approaches. Overall, this work advances confidence-aware LLM-based grading for human-in-the-loop assessment, supporting more trustworthy AI-assisted educational assessment systems.