Polarization Detection: A Hybrid Approach with AfroXLMR-Social and DeBERTa for Low- and High-Resource Settings

A hybrid architecture combines DeBERTa for English binary detection with AfroXLMR-Social for Hausa and fine-grained polarization subtasks. The approach utilizes Low-Rank Adaptation (LoRA) and nlpaug-based data augmentation to mitigate data scarcity. This strategy targets improved performance in low-resource settings across diverse linguistic contexts.