This dataset is an enhanced version of the original KHF News Dataset. It contains 5,539 news articles from the official website of the Committee for Emergency Situations and Civil Defense of Tajikistan (www.khf.tj), labelled with crisis-related categories using a LLaMA-3 8B model in a zero-shot setting. The articles cover the period from 2016 to 2025.

Curated by: Arabov Mullosharaf Kurbonovich (TajikNLPWorld)
Language(s) (NLP): Tajik (tg)
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 from khf.tj) remain the property of their respective authors and publishers. They are not owned by the dataset curator 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/TajikNLPWorld/khf_news_labeled

Uses

Direct Use

The dataset can be used for:

Training and evaluating text classification models for crisis/disaster-related news in Tajik
Monitoring emergency situations and disaster response
Developing NLP applications for low-resource languages in the domain of civil defense
Studying category distribution and trends in emergency reporting

Out-of-Scope Use

The dataset is not intended for:

Real-time emergency response without human verification
Any use that violates the original copyright of the news texts (see Licensing & Legal Notice)
Tasks requiring fine-grained temporal analysis beyond the available date field

Dataset Structure

Data Fields

Field	Type	Description
`url`	string	Direct URL to the article
`title`	string	Headline of the news item
`content`	string	Full text of the article
`excerpt`	string	First 200 characters (preview)
`date`	string	Publication date in ISO format
`category`	string	One of 7 crisis categories (see below)
`author`	string	Author name
`site`	string	Domain (`khf.tj`)
`hash`	string	SHA256 hash of the content (for deduplication)

Data Splits

The dataset contains a single split (train) with all 5,539 records.

Categories (7 classes)

Category (English)	Category (Tajik/Russian)	Count	Percentage
General	умумӣ / общие новости	3,110	56.1%
Rescue	наҷот / спасательные работы	814	14.7%
Training	машқ / учения, профилактика	705	12.7%
Flood	обхезӣ / наводнение, сель	602	10.9%
Earthquake	заминларза / землетрясение	213	3.8%
Fire	сӯхтор / пожар	59	1.1%
Landslide	ярч / оползень, обвал	36	0.6%

Dataset Creation

Curation Rationale

The dataset was created to provide a labelled resource for crisis/disaster classification in Tajik, addressing the lack of annotated data for low-resource languages in the emergency management domain. The original KHF News Dataset was unlabelled; this version adds meaningful categories to enable supervised learning.

Source Data

Data Collection and Processing

The articles were collected from the official website of the Committee for Emergency Situations and Civil Defense of Tajikistan (www.khf.tj). The labelling process used a LLaMA-3 8B model (quantized to 4-bit) via Ollama in a zero-shot prompting setup. The model was instructed to output a single category in JSON format. Only 73 out of 5,612 articles (1.3%) were marked as unknown_category and excluded from this release.

Who are the source data producers?

The original news texts were produced by the Committee for Emergency Situations and Civil Defense of Tajikistan and its staff. The automatic labels were generated by the curator using a large language model.

Annotations

Categories were assigned automatically by the LLaMA-3 8B model. No manual validation was performed on the labels; the model's zero-shot predictions were accepted as-is.

Personal and Sensitive Information

The dataset consists of publicly available news articles and does not intentionally include personal or sensitive information beyond what appears in the original texts.

Bias, Risks, and Limitations

Label noise: Automatic labelling may contain errors; some articles could be misclassified.
Class imbalance: The general category dominates (56.1%), while fire and landslide are very rare.
Temporal coverage: Articles span 2016–2025, but distribution across years may be uneven.
Copyright constraints: The underlying texts may be protected; users must respect original rights (see Licensing & Legal Notice).

Recommendations

Use class weights or oversampling to handle imbalance when training.
Validate labels on a small manual subset if high accuracy is required.
Be cautious when using the dataset for real-time decision-making.

Citation

BibTeX (dataset):

@dataset{khf_news_labeled_2026,
  author = {Arabov, Mullosharaf Kurbonovich},
  title = {KHF News Labeled Dataset – Tajikistan Emergency Committee},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/TajikNLPWorld/khf_news_labeled}
}

APA (dataset):

Arabov, M. K. (2026). KHF News Labeled Dataset – Tajikistan Emergency Committee [Data set]. Hugging Face. https://huggingface.co/datasets/TajikNLPWorld/khf_news_labeled

Glossary

Zero-shot prompting – a method where a model performs a task without explicit training examples.
Low-resource language – a language with limited digital resources for NLP.

More Information

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

Dataset Card Authors

Arabov Mullosharaf Kurbonovich (TajikNLPWorld)

Dataset Card Contact

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

Downloads last month: 28

Total file size:

11.5 MB