LLM Confidence Evaluation Measures in Zero-Shot CSS Classification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Farr, David, Cruickshank, Iain, Manzonelli, Nico, Clark, Nicholas, Starbird, Kate, West, Jevin
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915002432094208
author Farr, David
Cruickshank, Iain
Manzonelli, Nico
Clark, Nicholas
Starbird, Kate
West, Jevin
author_facet Farr, David
Cruickshank, Iain
Manzonelli, Nico
Clark, Nicholas
Starbird, Kate
West, Jevin
contents Assessing classification confidence is critical for leveraging large language models (LLMs) in automated labeling tasks, especially in the sensitive domains presented by Computational Social Science (CSS) tasks. In this paper, we make three key contributions: (1) we propose an uncertainty quantification (UQ) performance measure tailored for data annotation tasks, (2) we compare, for the first time, five different UQ strategies across three distinct LLMs and CSS data annotation tasks, (3) we introduce a novel UQ aggregation strategy that effectively identifies low-confidence LLM annotations and disproportionately uncovers data incorrectly labeled by the LLMs. Our results demonstrate that our proposed UQ aggregation strategy improves upon existing methods andcan be used to significantly improve human-in-the-loop data annotation processes.
format Preprint
id arxiv_https___arxiv_org_abs_2410_13047
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LLM Confidence Evaluation Measures in Zero-Shot CSS Classification
Farr, David
Cruickshank, Iain
Manzonelli, Nico
Clark, Nicholas
Starbird, Kate
West, Jevin
Human-Computer Interaction
Computation and Language
Information Retrieval
Assessing classification confidence is critical for leveraging large language models (LLMs) in automated labeling tasks, especially in the sensitive domains presented by Computational Social Science (CSS) tasks. In this paper, we make three key contributions: (1) we propose an uncertainty quantification (UQ) performance measure tailored for data annotation tasks, (2) we compare, for the first time, five different UQ strategies across three distinct LLMs and CSS data annotation tasks, (3) we introduce a novel UQ aggregation strategy that effectively identifies low-confidence LLM annotations and disproportionately uncovers data incorrectly labeled by the LLMs. Our results demonstrate that our proposed UQ aggregation strategy improves upon existing methods andcan be used to significantly improve human-in-the-loop data annotation processes.
title LLM Confidence Evaluation Measures in Zero-Shot CSS Classification
topic Human-Computer Interaction
Computation and Language
Information Retrieval
url https://arxiv.org/abs/2410.13047