This dataset contains 22,318 Bashkir-language news and analytical articles annotated with 14 thematic labels for multi-label text classification tasks. Each article can belong to several categories simultaneously. The average number of labels per article is 3.6. The dataset is designed to support NLP research and applications for the Bashkir language, a low-resource Turkic language.

Curated by: Arabov Mullosharaf Kurbonovich, Khaybullina Svetlana Sergeevna (BashkirNLPWorld)
Language(s) (NLP): Bashkir (ba), Cyrillic script
License: other – see Licensing & Legal Notice below.

Licensing & Legal Notice

This dataset follows the practice established by large web‑crawled corpora such as HPLT and OSCAR:

Original source texts (news articles, analytical pieces) remain the property of their respective authors and publishers. They are not owned by the BashkirNLPWorld team and are not covered by the CC BY‑SA 4.0 license.
The structured compilation, metadata, and any original annotations created during dataset preparation are released under the CC BY‑SA 4.0 license.
Users are solely responsible for ensuring their use of the underlying texts complies with applicable copyright law. For commercial use of verbatim excerpts, permission from the original copyright holders may be required.
A notice‑and‑takedown procedure is in place: rights holders can request removal of specific content by contacting the dataset maintainers (see Dataset Card Contact). We commit to responding within 14 business days and removing disputed content in the next release.

Dataset Sources

Repository: https://huggingface.co/datasets/BashkirNLPWorld/bashkir-news-multilabel
Related datasets:

Uses

Direct Use

This dataset is suitable for:

Multi-label text classification (14 possible labels)
Training multi-label classifiers (transformers, SVM, ensemble methods)
Multi-task learning
Fine-tuning large language models for Bashkir text categorization

Out-of-Scope Use

The dataset should not be used for single-label classification tasks (use the multiclass version instead).
It is not intended for generating offensive content or for any unethical applications.
Redistribution of verbatim texts without permission from original copyright holders is prohibited.

Dataset Structure

Data Fields

Field	Type	Description
`content`	string	Full article text
`title`	string	Article title
`labels`	list[string]	List of thematic labels for this article (e.g., `["Йәмғиәт", "Дин"]`)
`label_vector`	list[int32]	Binary vector of length 14 indicating which labels are present
`num_labels`	int64	Number of labels assigned (1–14)
`original_category`	string	Original normalized category from source data
`content_length`	int64	Length of the text in characters
`resource`	string	Original URL or resource identifier (if available)
`date`	string	Publication date (when available)

Data Splits

The dataset contains a single split (train) with all 22,318 examples. Users are encouraged to create their own train/validation/test splits.

Label Definitions

Index	Label (Bashkir)	Label (English)	Count	Percentage
0	Йәмғиәт	Society	12,251	55.1%
1	Мәҙәниәт	Culture	~?	~?
2	Дин	Religion	~?	~?
3	Сәйәсәт	Politics	~?	~?
4	Иҡтисад	Economy	~?	~?
5	Мәғариф	Education	~?	~?
6	Сәләмәтлек	Health	~?	~?
7	Спорт	Sports	~?	~?
8	Шоу-бизнес	Show Business	~?	~?
9	Татарстан	Tatarstan	~?	~?
10	Башҡортостан	Bashkortostan	~?	~?
11	Донъя	World	~?	~?
12	Әҙәбиәт	Literature	~?	~?
13	Хикәйә	Fiction/Stories	1,880	8.4%

Note: Detailed per-label statistics were not fully provided in the original card; only the most/least frequent are shown above.

Label Statistics

Most frequent label: Йәмғиәт (Society) – 12,251 occurrences (55.1%)
Least frequent label: Хикәйә (Fiction/Stories) – 1,880 occurrences (8.4%)
Average labels per article: 3.6
Most common label pairs:
- Дин + Йәмғиәт (22.6%)
- Донъя + Йәмғиәт (20.8%)
- Башҡортостан + Йәмғиәт (19.2%)

Dataset Creation

Curation Rationale

The goal was to create a multi-label dataset for Bashkir that reflects the natural overlap of topics in news articles. Unlike single-label datasets, this captures the complexity where an article about education might also touch on society or politics.

Source Data

Data Collection and Processing

Articles were collected from 14 Bashkir online sources. Processing steps:

Extracted JSONL files from raw HTML.
Removed texts shorter than 50 characters or longer than 10,000 characters.
Removed exact duplicates.
Applied keyword-based multi-label annotation.
Normalized category names in the original_category field.

Who are the source data producers?

The articles were originally written by journalists, authors, and contributors of the respective online publications. The BashkirNLPWorld team does not claim ownership of the content. The original texts remain the property of their authors and publishers.

Annotations

Annotation process

Labels were generated using a keyword-based approach:

A dictionary of Bashkir keywords was created for each label.
Each article's title and content were scanned for these keywords.
If no keywords were found, the article's original category was used as a fallback (if it belonged to one of the 14 labels).

Who are the annotators?

The annotation was performed automatically by the BashkirNLPWorld team. No manual human annotation was carried out.

Personal and Sensitive Information

The texts are public news articles and do not intentionally contain personally identifiable information beyond what is already published.

Bias, Risks, and Limitations

Label bias: The keyword-based approach may introduce bias towards certain terms or topics.
Multi-label sparsity: Some labels (e.g., Хикәйә) are rare and may not provide enough examples for robust classification.
Source bias: The dataset is dominated by certain sources (e.g., azatliqorg accounts for 28% of data).
Genre bias: All texts are from news sources; may not represent other domains.
Date incompleteness: Many articles lack publication dates.
Copyright constraints: The underlying texts are protected; users must respect original rights (see Licensing & Legal Notice).

Recommendations

Be aware of label distribution and consider techniques for handling imbalanced multi-label data.
For better label quality, consider filtering by confidence or using ensemble methods.
For tasks requiring high precision, consider using only articles where labels came from explicit keyword matches.

Citation

BibTeX (dataset):

@dataset{arabov2026bashkirmultilabel,
  author       = {Arabov, Mullosharaf Kurbonovich and Khaybullina, Svetlana Sergeevna},
  title        = {Bashkir News Multilabel Classification Dataset},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/BashkirNLPWorld/bashkir-news-multilabel}
}

APA (dataset): Arabov, M. K., & Khaybullina, S. S. (2026). Bashkir News Multilabel Classification Dataset [Data set]. Hugging Face. https://huggingface.co/datasets/BashkirNLPWorld/bashkir-news-multilabel

Glossary

Multi-label classification – a task where each input can be assigned multiple labels simultaneously.
Low-resource language – a language with limited digital resources and tools.
Keyword-based annotation – assigning labels based on the presence of predefined keywords in the text.

More Information

For questions, contributions, or feedback, please open an issue on the Hugging Face repository.

Dataset Card Authors

Arabov Mullosharaf Kurbonovich
Khaybullina Svetlana Sergeevna
BashkirNLPWorld

Dataset Card Contact

For questions, takedown requests, or collaboration, please open an issue on the Hugging Face repository:
https://huggingface.co/datasets/BashkirNLPWorld/bashkir-news-multilabel/discussions
📧 Direct email: cool.araby@gmail.com

Downloads last month: 14

Total file size:

44.4 MB