docs/DATASET_CARD.md

# Dataset Card – Khmer Document Corpus

This document provides additional technical detail for the HuggingFace Dataset Card.

## Dataset Summary

The Khmer Document Corpus is an open-source collection of PDF documents targeting Document AI research for Khmer and multilingual Khmer-English documents.

## Supported Tasks

- Document Classification
- OCR / Text Recognition
- Document Layout Analysis
- Table Recognition
- Font Recognition
- Information Extraction
- Vision-Language Modelling

## Languages

| Code | Language |
|------|----------|
| `km` | Khmer |
| `en` | English |
| `mixed` | Mixed Khmer-English |

## Dataset Structure

```
khmer-document-corpus/
├── metadata/          # JSON metadata per document
├── pdf/               # Raw PDF files (tracked via Git LFS)
├── preview/           # Page preview PNG images
├── datasets/          # HuggingFace loading script
└── scripts/           # Utility scripts
```

## Source Data

Documents are collected from publicly available government publications, universities, open-access research, and legal public forms.

## Annotations

Currently, only raw PDFs and metadata stubs are included (v0.1). Future versions will include OCR, layout, table, and font annotations.

## Licensing

The dataset organization and metadata are licensed under CC BY 4.0.  
Individual documents retain their original licenses; see each document's `license` metadata field.

## Citation

```bibtex
@dataset{khmer_document_corpus_2026,
  title={Khmer Document Corpus},
  author={Darachhat},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/Darachhat/khmer-document-corpus}
}
```