Visegradmedia Project: Identifying News Slant in Crisis Communication Using Artificial Intelligence

📘 Project Overview

“Identifying News Slant in Crisis Communication Using Artificial Intelligence” is a research project applying deep learning and NLP methods to uncover emotional framing and slanted narratives in online news from Czechia, Hungary, Poland, and Slovakia.

The project uses a combination of:

• Text preprocessing and topic modeling (LDA)
• Emotion classification based on multilingual fine-tuned transformer models
• Manual annotation and synthetic data generation
• Cross-linguistic model evaluation

🔍 Emotion Models

These models are fine-tuned XLM-RoBERTa classifiers trained on project-specific emotion labels:

• 🇭🇺 Hungarian Emotion Classifier
• 🇵🇱 Polish Emotion Classifier
• 🇨🇿 Czech Emotion Classifier
• 🇸🇰 Slovak Emotion Classifier

All models are available for non-commercial academic use.

📂 Dataset Access

The annotated news corpus used in this project is hosted on OSF:

🔗 Visegradmedia – Central European News Corpus (OSF)
Version: 1.0 | Year: 2025

📌 Terms of use:

• Permitted: academic research, educational use
• Prohibited: redistribution, commercial use, training commercial models

To request access, contact: [email protected]

🧾 Citation

If you use any of the models, please cite:

Üveges, I. & Ring, O. (2025). Evaluating the Impact of Synthetic Data on Emotion Classification: A Linguistic and Structural Analysis. Information, 16(4), 330.
DOI: 10.3390/info16040330

🌍 Funding Acknowledgement

This project is co-financed by the Governments of Czechia, Hungary, Poland, and Slovakia through Visegrad Grants from the International Visegrad Fund.