Modern AI methods for detecting propaganda in text

• Datum: 2025
• 1. Verfasser: Бойко , О.А.
• Format: Artikel
• Sprache: English
• Veröffentlicht: Інститут проблем реєстрації інформації НАН України 2025
• Schlagworte: propaganda detection; natural language processing; artificial intelligence; large language models; reasoning language models; multi-agent systems
• Online Zugang: http://drsp.ipri.kiev.ua/article/view/336147
• Назва журналу: Data Recording, Storage & Processing

Propaganda is a key weapon of modern cognitive warfare, vividly illustrated by Russia’s war against Ukraine, where information manipulation has become as strategically significant as military action. Propaganda’s subtle psychological and linguistic tactics challenge traditional detection methods, demanding more sophisticated, context-aware technologies. This paper begins by reviewing supervised AI models, noting their dependence on expertly annotated corpora. Then, it outlines recent progress in unsupervised methods, showing how large language models (LLMs) can flag manipulative tactics with minimal labeled data while supplying human-readable justifications. Finally, three directions for future research are proposed: (1) reasoning language models for stepwise analysis; (2) costef-ficient multi-agent systems; and (3) hybrid frameworks that combine the first two. Advancing these methods offers a promising pathway for safeguarding democratic societies against evolving propaganda strategies.