You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Cleaned Georgian News Articles Dataset (Large-Scale CPT Corpus)

A large-scale, high-quality Georgian news corpus specifically optimized for Continued Pre-training (CPT), domain adaptation, and vocabulary expansion of Georgian Large Language Models (LLMs). This dataset contains over 1.3 million articles and has been processed using a strict cleaning, de-duplication, and anonymization pipeline.

Dataset Summary

To build a high-performance Georgian LLM, models must be exposed to large amounts of natural language prose beyond standard encyclopedic entries (Wikipedia). This dataset provides a massive collection of Georgian articles covering general news, politics, business, culture, and technology.

To protect the privacy of publishers and prevent automated detection systems from flags on bulk source scraping, the dataset has been strictly processed: No Metadata, Anonymized Text, Shuffled Rows, Exact text de-duplication, Length Filter.


Dataset Structure

The dataset contains a single train split in a compressed Parquet format:

Column Type Description
text string Cleaned, anonymized Georgian article prose (no URLs, no metadata)

How to Use

You can load this dataset directly in Python using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset from Hugging Face
dataset = load_dataset("iraklixyz/georgian-news-cleaned")

# Inspect a sample
sample = dataset["train"][0]
print(f"Text Preview:\n{sample['text'][:500]}")

Creator Attribution & Collaborations

This dataset was created and preprocessed by Irakli Maisuradze (@iraklixyz).

If you have suggestions or want to contribute to the scraping and cleaning tools, feel free to open an issue or pull request on the GitHub repository!


License

This dataset is distributed under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

Downloads last month
31