Aligning LLM Uncertainty with Human Disagreement in Subjectivity Analysis

TL;DR

This paper introduces a framework for large language models to express uncertainty in subjectivity analysis, aligning model confidence with human disagreement to improve reliability.

Contribution

The proposed DPUA framework jointly models predictions, rationales, and uncertainty, enhancing sensitivity to disagreement and aligning confidence with human judgment.

Findings

DPUA maintains task performance while better reflecting human disagreement.

It reduces overconfidence on boundary samples.

Improves out-of-distribution generalization.

Abstract

Large language models for subjectivity analysis are typically trained with aggregated labels, which compress variations in human judgment into a single supervision signal. This paradigm overlooks the intrinsic uncertainty of low-agreement samples and often induces overconfident predictions, undermining reliability and generalization in complex subjective settings. In this work, we advocate uncertainty-aware subjectivity analysis, where models are expected to make predictions while expressing uncertainty that reflects human disagreement. To operationalize this perspective, we propose a two-phase Disagreement Perception and Uncertainty Alignment (DPUA) framework. Specifically, DPUA jointly models label prediction, rationale generation, and uncertainty expression under an uncertainty-aware setting. In the disagreement perception phase, adaptive decoupled learning enhances the model's sensitivity to disagreement-related cues while preserving task performance. In the uncertainty alignment phase, GRPO-based reward optimization further improves uncertainty-aware reasoning and aligns the model's confidence expression with the human disagreement distribution. Experiments on three subjectivity analysis tasks show that DPUA preserves task performance while better aligning model uncertainty with human disagreement, mitigating overconfidence on boundary samples, and improving out-of-distribution generalization.