Reproduce “Aligning LLM Uncertainty with Human Disagreement in Subjectivity Analysis” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Junyu Lu, Deyi Ji, Xuanyi Liu, Lanyun Zhu, Bo Xu, Liang Yang, Xian-Sheng Hua, Hongfei Lin

DPUA trains models to predict subjective labels, explain evidence and disagreement, and report confidence aligned with annotator distributions. It weights label and rationale learning by agreement, then rewards rationale quality and distribution alignment. Across three binary tasks, results improve uncertainty alignment while retaining competitive prediction; this measures annotation fit, not objective truth or why people disagree.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.