Our study uses an 89 million tweet dataset to examine Twitter discourse on the Russo-Ukrainian War, employing zero-shot learning due to diverse languages and lack of labels.
Research Innovation
Zero-Shot Learning Approach
- Multilingual Analysis: Handles diverse languages without pre-labeled data
- BART Model: Facebook’s BART zero-shot classification for robust analysis
- Scalable Framework: Methodology applicable to future crisis situations
Analysis Dimensions
Sentiment Analysis
- Public Mood Tracking: Real-time sentiment evolution throughout conflict
- Geographic Variations: Sentiment differences across regions and demographics
- Temporal Patterns: How sentiment changed with major conflict events
Topic Narratives
- Narrative Evolution: How key topics and themes developed over time
- Information Propagation: Tracking how narratives spread across the platform
- Dominant Themes: Identification of persistent vs. emerging topics
Toxicity Assessment
- Content Moderation: Analysis of harmful content patterns
- Escalation Patterns: How toxicity levels changed during crisis peaks
- Platform Response: Effectiveness of content moderation during crisis
Military Communication
- Strategic Communication: Analysis of military-related information flow
- Propaganda Detection: Identification of coordinated information campaigns
- Verification Challenges: Understanding misinformation in conflict contexts
Technical Implementation
- Large-scale Processing: Efficient handling of 89M tweet dataset
- Real-time Analysis: Continuous monitoring and analysis capabilities
- Robust Classification: Zero-shot learning for unlabeled multilingual data
Societal Impact
This research sheds light on social media’s role in modern conflicts and provides frameworks for understanding how platforms function during global crises, contributing to better crisis communication strategies and platform governance.