JAQKET_1000 / README.md
masa-ok's picture
クレジット情報を更新
b9d4fbd verified
metadata
license: cc-by-sa-4.0
dataset_info:
  features:
    - name: question
      dtype: string
    - name: answer_entity
      dtype: string
    - name: label
      dtype: int64
    - name: answer_candidates
      dtype: string
  splits:
    - name: train
      num_bytes: 272455
      num_examples: 1000
    - name: validation
      num_bytes: 56309
      num_examples: 200
  download_size: 212204
  dataset_size: 328764
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*

ランダムに抽出した1000件の学習データセットへ変換

元データ
# DatasetDict({
#     train: Dataset({
#         features: ['qid', 'question', 'answer_entity', 'label', 'answer_candidates', 'contexts'],
#         num_rows: 13061
#     })
#     validation: Dataset({
#         features: ['qid', 'question', 'answer_entity', 'label', 'answer_candidates', 'contexts'],
#         num_rows: 271
#     })
# })

変換データ
# DatasetDict({
#     train: Dataset({
#         features: ['question', 'answer_entity', 'label', 'answer_candidates'],
#         num_rows: 1000
#     })
#     validation: Dataset({
#         features: ['question', 'answer_entity', 'label', 'answer_candidates'],
#         num_rows: 200
#     })
# })
データ元のクレジット
lisence : cc-by-sa-4.0
name : kumapo
URL : https://huggingface.co/datasets/kumapo/JAQKET

@InProceedings{Kurihara_nlp2020,
  author =  "鈴木正敏 and 鈴木潤 and 松田耕史 and ⻄田京介 and 井之上直也",
  title =   "JAQKET: クイズを題材にした日本語 QA データセットの構築",
  booktitle =   "言語処理学会第26回年次大会",
  year =    "2020",
  url = "https://www.anlp.jp/proceedings/annual_meeting/2020/pdf_dir/P2-24.pdf"
  note= "in Japanese"}'