Paired Completion: Flexible Quantification of Issue-framing at Scale with LLMs

TL;DR

This paper introduces 'paired completion', a novel method leveraging large language models' next-token probabilities to detect contrasting issue frames in text efficiently and accurately, especially in low-resource scenarios.

Contribution

The paper presents a new approach using LLMs' log probabilities for issue framing detection, providing a scalable, low-bias alternative to existing methods.

Findings

Effective in synthetic and real datasets

Cost-efficient and scalable for large collections

Performs well in low-resource settings

Abstract

Detecting issue framing in text - how different perspectives approach the same topic - is valuable for social science and policy analysis, yet challenging for automated methods due to subtle linguistic differences. We introduce `paired completion', a novel approach using LLM next-token log probabilities to detect contrasting frames using minimal examples. Through extensive evaluation across synthetic datasets and a human-labeled corpus, we demonstrate that paired completion is a cost-efficient, low-bias alternative to both prompt-based and embedding-based methods, offering a scalable solution for analyzing issue framing in large text collections, especially suited to low-resource settings.