UPR at SemEval-2026 Task 9: Multi-Label Classification of Polarization Across Social Dimensions and Manifestation Identification in Urdu
Year: 2026
Type: conference-paper
Abstract: Polarization detection in low-resource languages such as Urdu remains underexplored despite its importance for public discourse analysis.We address two complementary subtasks of polarization analysis in Urdu social media text.For social-dimension classification, we formulate the task as a multi-label problem across five dimensions: political, religious, racial/ethnic, gender/sexual, and other.For manifestation identification, we target six polarization expressions: stereotype, vilification, dehumanization, extreme language, lack of empathy, and invalidation.We fine-tune XLM-RoBERTa with language-specific preprocessing, duplicate filtering, and imbalanceaware augmentation and resampling for both subtasks.The proposed framework achieves a Macro-F1 of 0.75 on Subtask-2 and 0.72 on Subtask-3 on the validation set, which demonstrates the effectiveness of multilingual transformer models for multi-dimensional polarization analysis in low-resource Urdu text.
Institutions: University of Poonch Rawalakot, Sejong University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166877331