The Tatar News Multilabel Classification Dataset contains 55,709 Tatar language news articles annotated with 13 distinct topic labels in a multi-label setting (each article can have multiple labels). Each entry includes the full article content, title, label indices, multi-hot label vector, number of labels, original single category, source URL, publication date, and content length. The dataset is designed for training and evaluating multi-label text classification models for the Tatar language, a low-resource Turkic language. It is part of the Tat2Vec project by TatarNLPWorld.

Curated by: TatarNLPWorld Community
Language(s) (NLP): Tatar (tt)
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, titles, etc.) remain the property of their respective authors and publishers. They are not owned by the TatarNLPWorld team and are not covered by the MIT license or any other open license applied to the annotations.
The structured compilation, metadata, and any original annotations created by the TatarNLPWorld community are released under the MIT 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 articles 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/TatarNLPWorld/tatar-news-analysis-multilabel
Paper: Arabov, M. K., Gilmullin, R. A., & Burnashev, R. A. (2026). Combining Classical and Transformer-based Approaches for Text Classification and Topic Modeling of the Tatar Language. In 2026 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT) (pp. 1–4). IEEE. https://doi.org/10.1109/USBEREIT70063.2026.11580632
Demo: [Coming Soon]

Uses

Direct Use

This dataset is intended for:

Multi-label text classification (13 labels, multiple labels per document)
Topic modeling and clustering
Feature extraction for embedding Tatar texts
Sentence similarity and semantic search
Text generation and language modeling
Summarization of Tatar news articles
Zero-shot classification and cross-lingual experiments

Out-of-Scope Use

This dataset should not be used for:

Redistribution of verbatim news articles without permission from original copyright holders
Any use that could misrepresent the original authors' intent (e.g., altering labels without validation)
Applications requiring real-time or up-to-date news content – the dataset is a static snapshot
Automated decision-making in sensitive domains without human oversight

Dataset Structure

Data Fields

Field	Type	Description
`content`	string	Full article content in Tatar language
`title`	string	Article title
`labels`	list[int64]	List of label indices (0–12)
`labels_text`	list[string]	List of label names (Tatar)
`label_vector`	list[int32]	Multi-hot vector of length 13 (1 indicates label present)
`num_labels`	int32	Number of labels assigned
`original_category`	string	Original single category from source metadata
`content_length`	int32	Length of content in characters
`resource`	string	Source URL or identifier
`date`	string	Publication date (ISO format)

Data Splits

The dataset is split using stratified sampling to preserve the label distribution:

Split	Size	Percentage
`train`	50,138	90%
`validation`	5,571	10%

Dataset Creation

Curation Rationale

Tatar language multi-label classification resources are extremely scarce. This dataset was created to enable research in multi-label text classification for Tatar, a low-resource language, and to support the development of NLP tools for categorizing news articles with multiple overlapping topics.

Source Data

Data Collection and Processing

The texts were collected from various Tatar language news sources from 2008 to 2026. The processing pipeline included:

Web crawling of publicly available Tatar news websites
Extraction of article content, titles, URLs, and publication dates
Assignment of multi-label annotations using a combination of:
- Keyword matching
- Source metadata (categories, tags)
- Manual review for a subset
Cleaning and normalization (HTML removal, whitespace)
Creation of label_vector and num_labels fields

Who are the source data producers?

The original texts were produced by journalists, editors, and media organizations from Tatar‑language outlets. The TatarNLPWorld community performed the collection, structuring, and multi-label annotation, but the underlying content remains the property of the original authors.

Annotations

Annotation process

Multi-label annotations were obtained semi-automatically:

Initial automatic labeling using keyword rules and source metadata (RSS categories, tags)
Manual verification and correction by Tatar-speaking annotators
Final quality control to ensure label consistency

Who are the annotators?

Annotators were Tatar language speakers with experience in linguistics or NLP, recruited from the TatarNLPWorld community.

Personal and Sensitive Information

The dataset consists solely of publicly available news articles and does not intentionally include personal or sensitive information. No anonymization was performed because the data is already public.

Bias, Risks, and Limitations

Domain bias: The dataset may overrepresent certain sources or time periods.
Label noise: Automatic annotation may contain errors; manual verification was limited.
Temporal coverage: Articles span 2008–2026, but distribution across years may be uneven.
Copyright constraints: The underlying texts are protected; users must respect original rights (see Licensing & Legal Notice).

Recommendations

Use appropriate multi-label evaluation metrics (e.g., F1 micro/macro, precision@k).
Validate performance on your specific target domain.
For commercial applications, verify the copyright status of articles before reuse.

Citation

BibTeX (dataset):

@dataset{tatar_news_multilabel_2026,
    title = {Tatar News Multilabel Classification Dataset},
    author = {TatarNLPWorld Community},
    year = {2026},
    publisher = {Hugging Face},
    url = {https://huggingface.co/datasets/TatarNLPWorld/tatar-news-analysis-multilabel}
}

APA (dataset): TatarNLPWorld Community. (2026). Tatar News Multilabel Classification Dataset [Data set]. Hugging Face. https://huggingface.co/datasets/TatarNLPWorld/tatar-news-analysis-multilabel

Related Publication

The following paper used this dataset or a similar version:

BibTeX:

@inproceedings{arabov2026combining,
  author = {Arabov, M. K. and Gilmullin, R. A. and Burnashev, R. A.},
  title = {Combining Classical and Transformer-based Approaches for Text Classification and Topic Modeling of the Tatar Language},
  booktitle = {2026 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT)},
  year = {2026},
  pages = {1--4},
  doi = {10.1109/USBEREIT70063.2026.11580632}
}

APA: Arabov, M. K., Gilmullin, R. A., & Burnashev, R. A. (2026). Combining Classical and Transformer-based Approaches for Text Classification and Topic Modeling of the Tatar Language. In 2026 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT) (pp. 1–4). IEEE. https://doi.org/10.1109/USBEREIT70063.2026.11580632

Glossary

Multi-label classification – a task where each input can be assigned multiple labels simultaneously.
Multi-hot vector – a binary vector where each dimension corresponds to a label; 1 indicates presence, 0 absence.
Stratified split – splitting method that preserves label distribution in each subset.

More Information

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

Dataset Card Authors

TatarNLPWorld Community

Dataset Card Contact

Hugging Face: https://huggingface.co/TatarNLPWorld
GitHub: https://github.com/TatarNLPWorld
Email: marabov@kpfu.ru

Downloads last month: 9

Total file size:

113 MB