metadata

language:
  - ko
license: cc
dataset_info:
  features:
    - name: title
      dtype: string
    - name: url
      dtype: string
    - name: context
      dtype: string
    - name: question
      dtype: string
    - name: answer
      struct:
        - name: answer_start
          dtype: int64
        - name: html_answer_start
          dtype: int64
        - name: html_answer_text
          dtype: string
        - name: text
          dtype: string
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 4252605635
      num_examples: 83486
    - name: validation
      num_bytes: 529418488
      num_examples: 10165
  download_size: 519777736
  dataset_size: 4782024123
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
task_categories:
  - question-answering
size_categories:
  - 10K<n<100K

Dataset Card for KorQuAD V2.1

Dataset Description

Homepage https://korquad.github.io/
Repository https://github.com/korquad/korquad.github.io
Paper KorQuAD 2.0: Korean QA Dataset for Web Document Machine Comprehension

Dataset Summary

KorQuAD 2.0 is a Korean question-answering dataset consisting of over 100,000 pairs. It differs from the previous standard dataset, KorQuAD 1.0, in three major ways. First, instead of being limited to one or two paragraphs, the given context consists of an entire Wikipedia page. Second, since the context may include tables and lists, it requires understanding of documents structured with HTML tags. Lastly, the answers are not limited to single words or phrases, but can span long sections such as entire paragraphs, tables, or lists.

Supported Tasks and Leaderboards

Supported Tasks: Question Answering

See the homepage for the leaderboard.

Languages

Korean

Dataset Structure

Data Instances

{
  'answer': {
    "text": "<table>\n<tbody><tr>\n<th colspan=\"3\"> [omitted]",
    "html_answer_start": 114760,
    "html_answer_text": "<table class=\"wikitable\" style=\"text-align:center\">\n<tbody>[omitted]",
    "answer_start": 33018
  },
 'context': '<!DOCTYPE html>\n<html>\n<head>\n<meta>\n<title>파비오 칸나바로 - 위키백과, 우리 모두의 백과사전</title>\n\n\n<link>\n.....[omitted]',
 'id': '28097',
 'question': '칸나바로가 그의 조국을 위해 뛴 연도별 경기수와 골수는 어떻게 될까?',
 'title': '파비오_칸나바로',
 'url': 'https://ko.wikipedia.org/wiki/파비오_칸나바로'
}

Data Fields

{
 'id': Value(dtype='string', id=None),
 'title': Value(dtype='string', id=None),
 'context': Value(dtype='string', id=None),
 'question': Value(dtype='string', id=None),
 'answer': {
   'text': Value(dtype='string', id=None),
   'answer_start': Value(dtype='int64', id=None),
   'html_answer_start': Value(dtype='int64', id=None),
   'html_answer_text': Value(dtype='string', id=None)},
 'url': Value(dtype='string', id=None),
 'raw_html': Value(dtype='string', id=None)
}

Data Splits

Train : 83486
Validation: 10165

Dataset Creation

See the paper

Additional Information

Dataset Curators

Authors crowd-sourced the annotators to build the dataset. See the paper for more information.

Licensing Information

The license is originally distributed with CC-BY-ND-2.0 License. We redistribute the data for the convenience of model evaluation. We are currently asking Authors for further plans if they will directly distribute the data on huggingface.

Citation Information

@article{NODE09353166,
    author={Youngmin Kim,Seungyoung Lim;Hyunjeong Lee;Soyoon Park;Myungji Kim},
    title={{KorQuAD 2.0: Korean QA Dataset for Web Document Machine Comprehension}},
    booltitle={{Journal of KIISE 제47권 제6호}},
    journal={{Journal of KIISE}},
    volume={{47}},
    issue={{6}},
    publisher={The Korean Institute of Information Scientists and Engineers},
    year={2020},
    ISSN={{2383-630X}},
    pages={577-586},
    url={https://koreascience.kr/article/CFKO201930060807852.page}
}

Contact Information

If you find something wrong happened while we upload the dataset to huggingface hub, contact [email protected].

Otherwise, see the homepage.

Contributions

See the homepage for more information.