Hidden Persuasion: Detecting Manipulative Narratives on Social Media During the 2022 Russian Invasion of Ukraine

2025-05-29

AI Summary

This study addresses the detection and classification of rhetorical and stylistic manipulation techniques deployed on Telegram in wartime Ukraine to influence public opinion. We propose a two-tier classification architecture: an upper tier employs LoRA-finetuned Gemma 2 to model semantic manipulation intent, enhanced by meta-feature integration and threshold optimization for robustness; a lower tier introduces a novel multi-objective XLM-RoBERTa framework trained jointly for token-level fine-grained manipulation span identification. Our method synergizes semantic modeling, multi-source feature engineering, and span recognition to jointly optimize both intent classification and boundary precision. Evaluated on the UNLP 2025 shared task, our system ranks second in the classification subtask and third in the span detection subtask—demonstrating strong effectiveness and generalizability on low-resource, high-noise wartime social media text.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

This paper presents one of the top-performing solutions to the UNLP 2025 Shared Task on Detecting Manipulation in Social Media. The task focuses on detecting and classifying rhetorical and stylistic manipulation techniques used to influence Ukrainian Telegram users. For the classification subtask, we fine-tuned the Gemma 2 language model with LoRA adapters and applied a second-level classifier leveraging meta-features and threshold optimization. For span detection, we employed an XLM-RoBERTa model trained for multi-target, including token binary classification. Our approach achieved 2nd place in classification and 3rd place in span detection.

Problem

• Detect manipulative narratives on social media
• Classify rhetorical and stylistic manipulation techniques
• Identify influence tactics on Ukrainian Telegram users

Innovation

• Fine-tuned Gemma 2 with LoRA adapters
• Used XLM-RoBERTa for span detection
• Applied meta-features and threshold optimization