This dataset contains 16,812 Tajik news articles labeled for binary classification between news (label=1) and analytics (label=0). The data are perfectly balanced: 8,406 news articles and 8,406 analytical pieces. It is intended as a benchmark for distinguishing factual news from opinion/analytical content in the Tajik language.

The articles have been cleaned, deduplicated, and filtered to ensure quality. The original categories (e.g., Хабарҳо, Ҷамъият) are also provided for reference.

Uses

Direct Use

Training and evaluating binary classifiers (news vs. analytics).
Benchmarking Tajik language models on a simple classification task.
Educational purposes in NLP for low-resource languages.

Out-of-Scope Use

The dataset is not intended for multiclass or multilabel classification; separate datasets are available for those tasks.

Dataset Structure

Data Fields

content (string): Full text of the article (title + body, concatenated with \n\n)
title (string): Article headline
label (int64): 1 for news, 0 for analytics
label_text (string): Textual description of the label ("news" or "analytics")
category (string): Original normalized category from the source (e.g., Хабарҳо, Ҷамъият)
content_length (int64): Length of content in characters
resource (string): URL of the article (if available)
date (string): Publication date (if available; otherwise empty)

Data Splits

A single split (train) with all 16,812 records. For evaluation, users should create their own train/validation/test splits (e.g., 80/10/10).

Class Distribution

Class	Label	Count	Percentage
News	1	8,406	50.0%
Analytics	0	8,406	50.0%

Top Categories (Original)

Category	Count	Percentage
Хабарҳо	8,406	50.0%
Ҷамъият	2,170	12.9%
Иқтисод	1,151	6.8%
Сиёсат	1,035	6.2%
Ҳодиса	812	4.8%
Амният	769	4.6%
Тоҷикистон	460	2.7%
Фарҳанг	453	2.7%
Варзиш	333	2.0%
Ҳукумат	241	1.4%

Dataset Creation

Curation Rationale

The dataset was created to provide a balanced binary classification task for Tajik, separating hard news from analytical/opinion pieces. This is a common real‑world application where automatic detection of news vs. commentary is valuable.

Source Data

Data Collection and Processing

Collection: Articles were collected from eight Tajik news portals (see cluster dataset card for full list) between 2015 and 2025.
Cleaning: HTML removed, whitespace normalized, deduplicated, filtered by length (50–10,000 characters).
Labeling: News articles were defined as those with the normalized category Хабарҳо (News). Analytics were all other articles (e.g., society, economy, politics, culture, etc.). The dataset was then balanced by undersampling the majority class (analytics) to match the number of news articles.
Final Format: Each article is stored with the binary label and the original category.

Who are the source data producers?

The original content was produced by the news portals listed in the cluster dataset card.

Annotations

Labels are derived from the normalized categories. News = Хабарҳо; all other categories = analytics. This is a simple, rule‑based labeling that is suitable for the intended binary task.

Personal and Sensitive Information

The dataset contains publicly available news articles; no additional personal information is included.

Bias, Risks, and Limitations

Label definition: The distinction between "news" and "analytics" is based solely on the source category. Some articles labeled as analytics may still contain factual news (e.g., political analysis), and some news articles may contain opinion elements.
Source imbalance: The majority of articles come from asiaplus.tj (62% of the original corpus), though the undersampling may have reduced this effect.
Date coverage: Not all articles have reliable dates.

Recommendations

Users should:

Evaluate on a held‑out set that reflects the original distribution of sources.
Consider that the binary labels may have some noise; cross‑validation is recommended.

Citation

If you use this dataset, please cite:

@misc{arabov2025tajikbinary,
  author = {Arabov, Mullosharaf Kurbonovich},
  title = {Tajik News Binary Classification Dataset},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/TajikNLPWorld/tajik-news-binary}}
}

Dataset Card Authors

Arabov Mullosharaf Kurbonovich
Email: cool.araby@gmail.com

Dataset Card Contact

For questions, please contact cool.araby@gmail.com.

Downloads last month: 9