sora2005/test-dataset

Japanese Visual Rich Document Information Retrieval Dataset (JaVDIR)

Overview

This dataset was created for japanese visual rich document image retrieval testing.

Dataset Content

The folder structure of the dataset is as follows: [Root directory of the dataset] ├── train │ ├── train.parquet │ ├── pdfs │ └── images ├── test │ ├── test.parquet │ ├── pdfs │ └── images ├── validation │ ├── validation.parquet │ ├── pdfs │ └── images └── ...

Each parquet file contains the following columns:

• answer: str - answer of query
  • Possible Values: 石油化学プラントです, 48.1%です
• document_filename: str - doc pdf file name
  • Possible Values: public_document00152.pdf
• query: str - query about doc pdf
  • Possible Values: コンタミ事故の最も多い原因はヒューマンエラーですか？\n解答は「はい」か「いいえ」で答えてく...
• search_topic: int - document type
  • Possible Values: {0: "Report", 1: "Pamphlet", 2: "Slide", 3: "Table"}
• page_number: int - page number of answer in doc
  • Possible Values: 2, 30, 60...
• page_contains_table: bool - flag of any table
  • Possible Values: True/False
• page_contains_figure: bool - flag of any figure
  • Possible Values: True/False
• query_type: int - query type
  • Possible Values: {0: "Yes/No questions", 1: "Factoid questions", 2: "Numerical questions", 3: "Open-ended questions"}
• document_url: str - document pdf url in repo
  • Possible Values: train/pdfs/kouhou00185.pdf
• complexity: int - document complexity based on query
  • Possible Values: {1: "text_only", 2: "text+image", 3: "text+charts", 4: "text+tables", 5: "vertical text+images(newspaper)"}
• page_image: str - page image of answer
  • Possible Values: train/images/kouhou00088_6.png
• ...

Data Format

.parquet .pdf .png

Data Size

• train/images: 8,428 images, 7.2 GB

• train/pdfs: 4,404 PDFs, 31 GB

• test/images: y images, x GB

• test/pdfs: y PDFs, x GB

• validation/images: y images, x GB

• validation/pdfs: y PDFs, x GB

Data Source

• JDocQA: shunk031/JDocQA (https://huggingface.co/datasets/shunk031/JDocQA) - Created by shunk031
• [Dataset Name 2]: [Creator's Name]/[Dataset Name 2] ([URL]) - Created by [Creator's Name]
• [Dataset Name 3]: [Creator's Name]/[Dataset Name 3] ([URL]) - Created by [Creator's Name]
• ...

Terms of Use

[Terms of use of the dataset]

Other Information

[Other special notes]

Contact

[Contact information (email address, etc.)]