Multi-Label Emotion Detection in Dialectal Arabic and English Using DeepSeek and SILMA AI
Khloud Khaled Elsayed, Ensaf Hussein Mohamed, Walaa Medhat · 2025
Multi-label emotion detection is a crucial task in natural language processing (NLP), particularly for recognizing multiple emotions within a single text. This challenge is further amplified in under-resourced languages like dialectal Arabic due to limited linguistic resources and morphological complexity. Existing research has largely overlooked these dialects, leaving significant gaps in performance evaluation. In this study, we address these challenges by assessing and enhancing the capabilities of DeepSeek-V3 and SILMA AI on Algerian Arabic, Moroccan Arabic, and English datasets—languages not previously explored in this context. We employ diverse prompting strategies and fine-tuning techniques to optimize model performance. Our findings demonstrate the effectiveness of tailored prompts and task-specific fine-tuning, leading to notable improvements in emotion detection. These insights contribute to advancing NLP applications in under-resourced languages, bridging linguistic gaps, and refining emotion analysis in complex textual environments.