metuKKhud/bashqort-task

Bashqort Topic Classification

Description

A topic classification dataset for Bashkir news headlines. This dataset was created because no publicly available topic classification benchmark exists for Bashkir. It is intended for task‑specific fine‑tuning and evaluation of LLMs adapted to Bashkir.

Data Creation

• Source: Random sampling of headlines from bash.news
• Annotation: Manual labeling by Ilyas Khatipov (native speaker of Bashkir)
• Cleaning: Removed categories with fewer than 17 samples to improve class balance

Class Distribution (after cleaning)

Topic	Count	Proportion
Culture (Мәҙәниәт)	58	21.7%
Healthcare (Һаулыҡ һаҡлау)	35	13.1%
Education (Мәғариф)	32	12.0%
Social sphere (Социаль өлкә)	25	9.4%
Politics (Сәйәсәт)	25	9.4%
Sports (Спорт)	21	7.9%
Military service (Хәрби хеҙмәт)	18	6.7%
Economy (Иҡтисад)	18	6.7%
Incidents (Ваҡиғалар)	18	6.7%
Security (Хәүефһеҙлек)	17	6.4%

Total: 267 headlines (10 classes)

Format

The dataset is provided in CSV/JSON format with the following columns:

• title: Bashkir news headline (string)
• topic: topic label (string, one of the 10 classes)

{
  "title": "Өфөлә мәктәптәрҙә яңы уҡыу йылы башланды",
  "topic": "Education"
}

Splits

No fixed train/test split. Users are encouraged to create their own splits (e.g., 80/20) for reproducibility.

Intended Use

Task‑specific fine‑tuning of LLMs for topic classification

Zero‑shot and few‑shot evaluation of Bashkir language understanding

Benchmark for future Bashkir NLP work

Licensing

MIT License

Citation

@misc{khudiakova2025bashqorttask,
  author       = {Khudiakova, Kseniia and Khatipov, Ilyas},
  title        = {Bashqort Topic Classification: News Headlines with 10 Topics},
  year         = {2025},
  howpublished = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/metuKKhud/bashqort-task}
}

Acknowledgements

Thanks to Ilyas Khatipov for native speaker validation and annotation.