README.md

metadata

dataset_info:
  features:
    - name: url
      dtype: string
    - name: Name
      dtype: string
    - name: media
      list:
        - name: url
          dtype: string
        - name: url_ori
          dtype: string
        - name: caption
          dtype: string
        - name: source
          dtype: string
        - name: type
          dtype: string
        - name: header_text
          dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 172283593
      num_examples: 253675
  download_size: 59583682
  dataset_size: 172283593
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

WikipediaMM-Indo Dataset

WikipediaMM-Indo is a comprehensive multimedia dataset collected from Indonesian regional Wikipedia dumps, including Indonesian, Javanese, and Minangkabau languages. The dataset contains images, videos, and audio files along with their associated captions.

Dataset Description

This dataset is created by scraping multimedia content from the latest HTML dumps provided by Wikimedia Enterprise. It includes content from three different Wikipedia language versions:

• Indonesian (id.wikipedia.org)
• Javanese (jv.wikipedia.org)
• Minangkabau (min.wikipedia.org)

Data Collection

• Source: Wikimedia Enterprise HTML dumps (November 2024)
• Extraction Method: Automated scraping of media elements and their captions
• Media Types: Images, Videos, and Audio files
• Caption Collection: Associated text descriptions when available

Statistics

Indonesian Wikipedia

• Total unique media links: 354,516
• Breakdown by media type:
  • Images: 352,150
  • Audio files: 1,057
  • Video files: 1,309

Javanese Wikipedia

• Statistics to be determined

Minangkabau Wikipedia

• Statistics to be determined

Data Format

Each entry in the dataset contains:

• Media URL
• Media type (image/video/audio)
• Associated caption (if available)
• Source language
• Additional metadata

Limitations and Considerations

1. Content Coverage

   • Some images from infoboxes might not be captured in the dataset
   • Complex page layouts may affect content extraction

2. Quality Considerations

   • Dataset may include icons and UI elements with non-meaningful captions
   • Additional filtering may be required depending on use case
   • Users should implement appropriate content filtering based on their needs

Usage

This dataset is suitable for:

• Multilingual multimedia research
• Cross-lingual media understanding
• Indonesian regional language processing
• Cultural and linguistic studies
• Machine learning model training

License

The dataset follows Wikimedia's licensing terms. Individual media files retain their original Wikipedia licenses.

Citation

@misc{wikipediamm_indo2024,
  title={WikipediaMM-Indo: A Multimedia Dataset from Indonesian Regional Wikipedia},
  year={2024},
  publisher={Hugging Face}
}

Future Updates

• Complete statistics for Javanese and Minangkabau content
• Regular updates with new Wikipedia dumps
• Additional metadata and filtering options

Contact

For questions, feedback, or contributions, please open an issue in this repository.