Kratos-AI/Japanese_Documents_Dataset_PDF

Japanese Documents Dataset (PDF)

This dataset contains a curated collection of Japanese-language documents in PDF format. The corpus includes textbooks, research papers, news articles, public-domain books, and government publications written in Japanese. It is intended to support AI research in OCR, document understanding, and multilingual text recognition.

Contact

For queries or collaborations related to this dataset, contact:

• [email protected]
• [email protected]

Supported Tasks

• Task Categories:

  • Document Classification
  • OCR and Text Recognition
  • Layout and Structure Analysis
  • Language Modeling for Japanese

• Supported Tasks:

  • Extraction of Japanese text from PDFs
  • OCR for Kanji, Hiragana, and Katakana scripts
  • Classification of documents by type (academic, literary, official, educational)
  • Benchmarking AI systems for Japanese document parsing and layout understanding

Languages

• Primary Language: Japanese
• Secondary Presence: English numerals, Romanized Japanese, and technical symbols (common in bilingual or academic PDFs)

Dataset Creation

Curation Rationale

The dataset was curated to improve AI systems’ ability to read and understand Japanese-language documents, including vertically written text and mixed-script layouts, enabling more accurate OCR and multilingual NLP research.

Source Data

• Contributors: Open-access Japanese libraries, educational institutions, and public-domain repositories.
• Collection Process: PDFs were collected from publicly available sources with clear open licenses permitting use for research and AI model training.

Other Known Limitations

• Bias: Primarily educational and official documents; limited representation of informal or handwritten material
• Layout Complexity: Some documents contain vertical text or mixed Japanese-English layouts that may challenge OCR accuracy
• Script Variation: Coverage includes modern Japanese; limited inclusion of historical or classical scripts

Intended Uses

✅ Direct Use

• Training OCR systems for Japanese text recognition
• Research in document layout understanding and vertical text OCR
• Digitization and analysis of Japanese-language academic or archival materials

❌ Out-of-Scope Use

• Identifying or profiling individuals from document content
• Commercial redistribution of copyrighted PDFs
• Use in surveillance, handwriting identification, or behavioral analysis

License

CC BY 4.0