Emotion Classification in Newspaper Headlines for Human Behaviour Prediction

ACL ARR 2025 February Submission184 Authors

Abstract: With the advent of Large Language Models (LLMs), substantial performance improvements have been reported across various Natural Language Processing (NLP) domains. However, further evaluation within specific NLP subdomains remains necessary. This study investigates the effectiveness of GPT-based ada-002 text embeddings in conjunction with lexical features for emotion classification in newspaper headlines. Newspapers are chosen, as prior research on emotion detection has largely concentrated on identifying the emotions expressed by the author of a text. Instead, the present study shifts the focus towards detecting emotional responses of the message recipient. To facilitate cross-linguistic comparability, an automated translation procedure is proposed and implemented on three publicly available, labeled emotion datasets to train supervised emotion classification models in both English and German. For both languages, the trained classifiers significantly outperform previous benchmark results by over 42%, demonstrating the superiority of the chosen approach. In terms of language comparison, the English classifier achieves a higher performance with an F1 score of 0.683 compared to an F1 score of 0.655 of the German classifier. To demonstrate the impact of emotional appeal on human behaviour in online marketing, the German classifier is applied to a real advertisement setting. This application reveals how emotional priming detected in the newspaper headline influences the likelihood of user interaction with the advertisements placed within the newspaper article.

Paper Type: Long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Research Area Keywords: Computational Social Science and Cultural Analytics, Sentiment Analysis, Stylistic Analysis, and Argument Mining, Machine Translation

Contribution Types: Publicly available software and/or pre-trained models

Languages Studied: english, german

Submission Number: 184