Against Political Polarization: A Unified Framework for Tracing Evolving Political Ideologies on Social Media
| Source: arXiv AI
Tags: political polarization, social media, graph neural networks, NLP, ideology detection, temporal modeling
TSN4PI, accepted at ACM Transactions on Intelligent Systems and Technology, combines LLM-based ideology detection with temporal graph neural networks to track how political positions shift over time on X and Truth Social, releasing two large-scale datasets for noncommercial research.
Details
Tracking political ideology on social media faces compounding challenges: data scarcity, noisy cross-domain content, expensive annotation, and the difficulty of modeling temporal shifts. TSN4PI proposes a unified framework addressing all four. Its first module, the Political Ideology Detection Network (PIDN), uses LLMs with style transfer and unsupervised domain adaptation to detect political leaning and filter irrelevant content without requiring heavy annotation. The second module, the Political Ideology Prediction Network (PIPN), models how those detected ideologies shift over time using temporal graph neural networks. The paper was accepted at ACM Transactions on Intelligent Systems and Technology and validated on X and Truth Social. Two large-scale datasets are released for noncommercial research. The framework captures ideology presence, intensity, and temporal evolution rather than static left-right classification. For researchers studying online radicalization, filter bubbles, or platform-specific polarization dynamics, TSN4PI offers a scalable automated methodology. The dataset release enables downstream reproducibility and gives future researchers a shared benchmark.