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.

Lajme Shqip - Albanian News Corpus

Dataset Description

A large-scale Albanian-language news corpus containing 1,095,000 articles (~868 MB) scraped from news portals across Albania, Kosovo, and North Macedonia over a period of nearly one year. This dataset aims to address the scarcity of Albanian NLP resources by providing a substantial text corpus for training and evaluating language models.

Albanian is a low-resource language with limited publicly available datasets for NLP research. This corpus helps bridge that gap by offering diverse, real-world text data from multiple news sources and categories.

Supported Tasks

  • Language Modeling / Text Generation: Pre-train or fine-tune language models on Albanian text.
  • Text Classification: Train classifiers on news categories.
  • Summarization: Use titles as summaries for article content.
  • Named Entity Recognition: Extract entities from Albanian news text.
  • Fill-Mask / MLM: Masked language model pre-training.

Languages

Albanian (sq) - the dataset includes standard Albanian as used in journalistic writing across Albania, Kosovo, and North Macedonia.

Dataset Structure

Data Fields

Field Type Description
id int64 Unique article identifier
category string News category (e.g., sport, politikë, ekonomi)
title string Article headline
summary string Article body text / summary
portal string Source news portal
image_url string URL of the article's featured image
link string Original article URL
fetch_date string Date when the article was scraped

Data Splits

Split Examples
train 1,095,000

Source Portals

The dataset includes articles from multiple Albanian-language news portals operating in Albania, Kosovo, and North Macedonia.

Dataset Creation

Source Data

Articles were collected via web scraping from publicly available Albanian-language news websites over a period of approximately one year (2025–2026). The scraper collected article metadata (title, category, link, image) along with article text content.

Collection Process

  • Method: Automated web scraping
  • Period: ~1 year of continuous collection
  • Regions covered: Albania, Kosovo, North Macedonia
  • Language: Albanian (sq)
  • Deduplication: Some duplicate entries may exist across portals that syndicate content.

Considerations for Using the Data

Biases

  • The dataset reflects the editorial biases of the source news portals.
  • Coverage may be skewed toward certain topics or regions depending on the portals included.
  • Political and social biases inherent in news media are present in the text.

Licensing & Ethical Considerations

  • This dataset is compiled from publicly available news articles for research purposes only.
  • The original content is owned by the respective news portals.
  • Users should respect the terms of service of the original sources.
  • This dataset should not be used for commercial purposes without obtaining appropriate permissions from content owners.

Citation

If you use this dataset in your research, please cite:

@dataset{lajme_shqip_2026,
  title={Lajme Shqip - Albanian News Corpus},
  author={Arbana Kadriu},
  year={2026},
  url={https://huggingface.co/datasets/akadriu/lajme-shqip},
  note={A large-scale Albanian news corpus from portals in Albania, Kosovo, and North Macedonia}
}

Contact

For questions or issues, please open a discussion on the dataset repository.

Downloads last month
33

Models trained or fine-tuned on akadriu/lajme-shqip