Analysing Personal Attacks in U.S. Presidential Debates

TL;DR

This paper presents a framework using transformer models to detect personal attacks in U.S. presidential debates, combining manual annotation and analysis across multiple election cycles to enhance understanding of political discourse.

Contribution

It introduces a novel approach that fine-tunes transformer models for detecting personal attacks in formal political speech, supported by extensive annotated debate data.

Findings

Transformer models outperform traditional methods in attack detection

Fine-tuning improves model accuracy on debate transcripts

Analysis reveals patterns of personal attacks across election cycles

Abstract

Personal attacks have become a notable feature of U.S. presidential debates and play an important role in shaping public perception during elections. Detecting such attacks can improve transparency in political discourse and provide insights for journalists, analysts and the public. Advances in deep learning and transformer-based models, particularly BERT and large language models (LLMs) have created new opportunities for automated detection of harmful language. Motivated by these developments, we present a framework for analysing personal attacks in U.S. presidential debates. Our work involves manual annotation of debate transcripts across the 2016, 2020 and 2024 election cycles, followed by statistical and language-model based analysis. We investigate the potential of fine-tuned transformer models alongside general-purpose LLMs to detect personal attacks in formal political speech. This study demonstrates how task-specific adaptation of modern language models can contribute to a deeper understanding of political communication.