README.md

metadata

dataset_info:
  - config_name: default
    features:
      - name: id
        dtype: int64
      - name: event
        dtype: string
      - name: news_headline
        dtype: string
      - name: news_body
        dtype: string
      - name: source_link
        dtype: string
      - name: date
        dtype: string
      - name: news_corpora_name
        dtype: string
      - name: label
        dtype: string
    splits:
      - name: test
        num_bytes: 2337300
        num_examples: 200
    download_size: 818627
    dataset_size: 2337300
  - config_name: test
    features:
      - name: id
        dtype: int64
      - name: event
        dtype: string
      - name: news_headline
        dtype: string
      - name: news_body
        dtype: string
      - name: source_link
        dtype: string
      - name: date
        dtype: string
      - name: news_corpora_name
        dtype: string
      - name: label
        dtype: string
    splits:
      - name: train
        num_bytes: 2337300
        num_examples: 200
    download_size: 818627
    dataset_size: 2337300
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
  - config_name: test
    data_files:
      - split: train
        path: test/train-*
license: mit
task_categories:
  - text-classification
language:
  - bn
tags:
  - politics
  - political-bias
  - stance-detection
  - political-bias
  - news-analysis
  - bengali
  - bangla
pretty_name: BanglaBias
size_categories:
  - n<1K

Bangla Article Stance Detection Benchmark Dataset

Dataset Summary

The Bangla Article Stance Detection Benchmark Dataset is a comprehensive resource for detecting political bias and stance in Bengali (Bangla) news articles. This dataset features human-annotated labels for stance detection, enabling researchers to train and evaluate models for identifying political bias in Bengali journalism.

This dataset is part of the research paper "Read Between the Lines: A Benchmark for Uncovering Political Bias in Bangla News Articles" (Accepted to BLP at AACL 2025).

📄 Paper: arXiv:2510.03898
🌐 Project Website: https://nusrat-lia.github.io/BanglaBias/

Supported Tasks and Leaderboards

• Text Classification: Stance detection and political bias classification
• Sentiment Analysis: Analyzing the stance of news articles toward government actions
• Media Bias Analysis: Understanding bias patterns in Bengali journalism

Languages

The dataset is in Bengali (Bangla) (bn).

Dataset Structure

Data Instances

A typical data instance contains:

{
    'id': 1,
    'event': 'তাজরীন ফ্যাশন অগ্নিকাণ্ড',
    'news_headline': 'তাজরীন অগ্নিকাণ্ড: 'কেয়ামতের আগে আর মেয়েটাকে দেখতে পাবো না'',
    'news_body': 'বাংলাদেশে তাজরীন গার্মেন্টসে ভয়াবহ অগ্নিকাণ্ডের সাত বছর পরও...',
    'source_link': 'https://www.bbc.com/bengali/news-42103660',
    'date': '2017-11-24',
    'news_corpora_name': 'BBC Bangla',
    'label': 'Govt critique'
}

Data Fields

• id (int): Unique identifier for each article
• event (string): The event or topic the article discusses (in Bengali)
• news_headline (string): The headline of the news article in Bengali
• news_body (string): The full text content of the news article in Bengali
• source_link (string): URL to the original article source
• date (string): Publication date of the article
• news_corpora_name (string): Source publication name (e.g., BBC Bangla, The Daily Star, Prothom Alo, etc.)
• label (string): Human-annotated stance label with three possible values:
  • Govt critique: Articles critical of the government
  • Govt leaning: Articles supportive or favorable toward the government
  • Neutral: Articles maintaining a neutral stance

Usage Example

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("dipta007/BanglaBias")

# Access the data
for example in dataset['train']:
    print(f"ID: {example['id']}")
    print(f"Event: {example['event']}")
    print(f"Headline: {example['news_headline']}")
    print(f"Source: {example['news_corpora_name']}")
    print(f"Date: {example['date']}")
    print(f"Label: {example['label']}")
    print(f"Article preview: {example['news_body'][:100]}...")
    print(f"Link: {example['source_link']}")
    print("---")

Additional Information

Dataset Curators

This dataset was curated by researchers from the paper:

• Nusrat Jahan Lia
• Shubhashis Roy Dipta
• Abdullah Khan Zehady
• Naymul Islam
• Madhusodan Chakraborty
• Abdullah Al Wasif

Citation Information

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

@article{lia2025read,
  title={Read Between the Lines: A Benchmark for Uncovering Political Bias in Bangla News Articles},
  author={Lia, Nusrat Jahan and Dipta, Shubhashis Roy and Zehady, Abdullah Khan and Islam, Naymul and Chakraborty, Madhusodan and Wasif, Abdullah Al},
  journal={arXiv preprint arXiv:2510.03898},
  year={2025}
}

Contact

For questions or concerns about this dataset, please refer to the paper or contact the authors.