EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Empowering Sentiment Analysis in African Low-Resource Languages Through Transformer Models and Strategic Language Selection

Nilanjana Raychawdhary, Sutanu Bhattacharya, Cheryl Seals, Gerry V. Dozier

IEEE Access · 2025 · 13 · 147859-147873

DOI: 10.1109/access.2025.3599480 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2025. Verfasst von Nilanjana Raychawdhary und 3 weiteren Autor:innen. Die Arbeit wurde bisher 8-mal zitiert (Crossref) und nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This research addresses the significant challenges of sentiment analysis in low-resource African languages by utilizing advanced transformer-based models to bridge gaps in natural language processing (NLP) for underrepresented linguistic communities. Covering 12 diverse languages: Hausa, Yoruba, Igbo, Nigerian Pidgin, Amharic, Algerian Arabic, Moroccan Arabic Darija, Swahili, Kinyarwanda, Twi, Mozambican Portuguese, and Xitsonga. Our work includes pre-trained language models (PLMs) such as XLM-R, AfroXLMR, AfriBERTa, and mDeBERTaV3, fine-tuning them for the specific task of sentiment classification. In particular, using datasets from the AfriSenti SemEval 2023 Shared Task 12, this process involves tailoring multilingual transformer models to low-resource African languages, enabling them to capture complex linguistic nuances and sentiment polarities effectively. The results demonstrate significant improvements across key metrics, including accuracy, precision, recall, and weighted F1 score. Notably, our work using AfroXLMR achieves the top 1 rank, with a weighted F1 score of 75.8%, outperforming all other submissions. In addition to multilingual transformer models, we evaluated traditional machine learning models, including CNN, Naïve Bayes, and SVM, to compare their performance. The evaluation focused on zero-shot sentiment classification for Tigrinya, a low-resource African language not seen during model training. The results indicate that while traditional models produced meaningful results, multilingual transformer models achieved higher performance in the zero-shot setting. This highlights both the challenges and potential of multilingual transfer approaches for sentiment analysis in African low-resource languages. Moreover, this research highlights the transformative role of Artificial Intelligence (AI) in addressing linguistic diversity, fostering digital inclusion, and enabling reliable sentiment analysis. Furthermore, it lays the groundwork for future advancements in processing low-resource languages and emphasizes the importance of promoting cultural inclusivity through AI technologies.

Zitieren

APA

Raychawdhary, N., Bhattacharya, S., Seals, C., & Dozier, G. V. (2025). Empowering Sentiment Analysis in African Low-Resource Languages Through Transformer Models and Strategic Language Selection. IEEE Access, 13, 147859-147873. https://doi.org/10.1109/access.2025.3599480

Vancouver

Raychawdhary N, Bhattacharya S, Seals C, Dozier GV. Empowering Sentiment Analysis in African Low-Resource Languages Through Transformer Models and Strategic Language Selection. IEEE Access. 2025;13:147859-147873. doi:10.1109/access.2025.3599480

BibTeX
@article{Raychawdhary2025,
  author = {Raychawdhary, Nilanjana and Bhattacharya, Sutanu and Seals, Cheryl and Dozier, Gerry V.},
  title = {Empowering Sentiment Analysis in African Low-Resource Languages Through Transformer Models and Strategic Language Selection},
  journal = {IEEE Access},
  year = {2025},
  volume = {13},
  pages = {147859--147873},
  doi = {10.1109/access.2025.3599480}
}
Kostenlos starten →