This paper details a submission to SemEval-2026 Task 9, focusing on multilingual polarization detection across English and Swahili. The researchers employed transformer-based models, specifically RoBERTa-base and AfroXLMR-base, incorporating class-weighted loss functions and threshold tuning to manage imbalanced datasets. Their approach achieved competitive F1 macro scores on binary polarization detection, polarization type classification, and manifestation identification, though error analysis indicated challenges with detecting dehumanization and lack of empathy. AI
IMPACT This research contributes to the development of models capable of understanding and classifying online polarization across different languages and cultural contexts.
RANK_REASON The cluster contains an academic paper detailing a submission to a specific task within a research competition.
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
