Explainable AI (XAI) for Sentiment Analysis in low-resource languages Düşük Kaynakli Dillerde Duygu Analizi için Açiklanabilir Yapay Zeka (XAI)
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636747
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Bias Detection, Cross-Lingual NLP, Cultural Validation, Explainable AI, Hausa, Low-Resource Languages, Sentiment Analysis, Swahili
- Ankara Üniversitesi Adresli: Evet
Özet
Natural Language Processing has achieved progress for high-resource languages, but low-resource African languages remain underserved. This paper presents an XAI framework for sentiment analysis in Swahili and Hausa. We fine-tune the afro-xlmr-small model and integrate Integrated Gradients, LIME, and SHAP to analyze model decisions. We introduce fidelity metrics to assess explanation quality and investigate biases related to loanwords and entities, alongside cultural validation. Results show that dataset quality and class balance are the primary determinants of performance in these settings.