Barta / README.md
dipta007's picture
Upload README.md with huggingface_hub
0f711d0 verified
metadata
license: cc-by-4.0
task_categories:
  - text-generation
  - summarization
language:
  - bn
tags:
  - bengali
  - news
  - headline-generation
  - pretraining
  - nlp
size_categories:
  - 1M<n<10M

Barta: Bengali News Corpus

Paper Coming Soon

Paper GitHub

Overview

Barta (বার্তা, "news" in Bengali) is a large-scale Bengali news corpus containing 1.7M articles from multiple Bangladeshi news sources, spanning diverse categories including politics, sports, entertainment, technology, and more.

Key features:

  • 1.7M articles from major Bangladeshi news outlets
  • Diverse categories covering national, international, sports, entertainment, technology, lifestyle, etc.
  • Structured fields including headline, article body, source, and category
  • Deduplicated and cleaned via content hashing

Dataset Statistics

Split Rows
train 1,708,308

Loading the Dataset

from datasets import load_dataset

# Load full dataset
dataset = load_dataset("dipta007/Barta", split="train")

# Stream without downloading
dataset = load_dataset("dipta007/Barta", split="train", streaming=True)

Data Fields

Field Type Description
id string Unique identifier (e.g., bd-pratidin-1196327)
article string Full article body in Bengali
headline string Article headline in Bengali
source string News outlet name (e.g., bd-pratidin, prothom-alo)
category string News category in Bengali (e.g., জাতীয়, খেলা, বিনোদন)

Example

{
  "id": "bd-pratidin-1196327",
  "source": "bd-pratidin",
  "category": "দেশগ্রাম",
  "headline": "বগুড়ায় মাদককারবারীকে গ্রেফতার করল পুলিশ",
  "article": "বগুড়ায় ৩০ হাজার পিস ইয়াবা ট্যাবলেটসহ মো. শামিম (৩০) নামের এক মাদককারবারীকে গ্রেফতার করেছে হাইওয়ে পুলিশ..."
}

Intended Use

  • Pretraining Bengali language models
  • Headline generation training and evaluation
  • Text summarization for Bengali
  • News categorization and topic modeling
  • Bengali NLP research and benchmarking

Source

Crawled from major Bangladeshi news websites. Articles are deduplicated by content hash to remove exact duplicates.

Citation

@article{Barta2025,
  title={Barta: A Large-Scale Bengali News Corpus for Headline Generation},
  author={Shubhashis Roy Dipta},
  year={2025}
}