AN AI-DRIVEN FRAMEWORK FOR AUTOMATIC DETECTION OF MANIPULATIVE LINGUISTIC STRATEGIES IN TELEGRAM USER COMMENTS

This paper introduces an AI-driven analytical framework designed to automatically identify manipulative linguistic strategies embedded in Telegram user comments. The platform’s open communication structure, minimal moderation, and multilingual user base create favorable conditions for the spread of manipulative discourse, often crafted to influence emotions, beliefs, and decision-making processes. Unlike traditional text classifiers, the proposed framework models manipulation as a complex phenomenon combining rhetorical, psychological, and contextual signals. The detection system integrates a compact transformer architecture with discourse-level features such as inferential intent, persuasive constructions, illusion-of-choice tactics, and emotionally charged framing. A specially created dataset of Telegram comments from regional channels was annotated using a multi-label scheme capturing diverse manipulation types. Experimental evaluation demonstrates that the proposed model achieves reliable performance in distinguishing manipulative comments from neutral communication despite noise, code-switching, and message brevity. The system’s design allows for scalable deployment in moderation tools, research platforms, and automated risk-analysis pipelines.