Seven categories of Yoruba news headlines sourced from BBC Yoruba, covering topics such as Africa, Nigeria, and world news. The dataset provides labeled text data specifically for low-resource African language processing as featured in EMNLP 2020 research on multilingual transformer models.
Use Cases
- Train a text classifier to predict news categories using the headline text and the 7 provided topic labels.
- Evaluate the performance of multilingual transformer models on Yoruba by testing accuracy across the 'nigeria' and 'world' classes.
- Conduct cross-lingual transfer learning research by comparing these Yoruba headlines against English BBC news datasets.
Strengths
- Features 7 distinct classification labels: africa, entertainment, health, nigeria, politics, sport, and world.
- Derived from authentic BBC Yoruba news article headlines.
- Validated through the EMNLP 2020 paper on distant supervision for African languages.