databek/uzbek_news_site

📰 Uzbek News Dataset

The Uzbek News Dataset is a large-scale text dataset automatically collected from multiple Uzbek online news websites.
Each record represents one crawled article that has been cleaned, normalized, and converted into Markdown format for easier use in Natural Language Processing (NLP) tasks.

📦 Dataset Summary

• Total Records: 42,800 articles
• Languages: Uzbek (uz, ru, en)
• Data Format: Parquet
• Sources: daryo.uz and other local Uzbek news portals
• Size: ~8.6 MB (compressed Parquet files)
• Split: train (full dataset)

The dataset is ideal for tasks such as:

• Text classification (e.g., topic or sentiment detection)
• News summarization
• Language modeling and text generation
• Named Entity Recognition (NER)
• Information retrieval and clustering

🧱 Dataset Structure

Column	Type	Description
url	string	Original URL of the article
text	string	Clean plain text version of the article
markdown	string	Markdown-formatted version of the article
lang	string	Language code
source	string	Source website (e.g., daryo.uz)
created_at	string	Date when the article was crawled

🧩 Example

from datasets import load_dataset

dataset = load_dataset("databek/uzbek_news_site")
print(dataset["train"][0])

📚 Citation

@dataset{uzbek_news_site,
  author    = {Databek Community},
  title     = {Uzbek News Dataset},
  year      = {2025},
  url       = {https://huggingface.co/datasets/databek/uzbek_news_site},
  note      = {Compiled and curated by the Databek AI & Data Community}
}