マックタコ

  • 説明:

MC-TACO は、一時的な常識理解を必要とする 13,000 の質問と回答のペアのデータセットです。データセットには、次の 5 つの時間プロパティが含まれています。

  1. 期間 (イベントの所要時間)
  2. 時間的順序 (イベントの一般的な順序)
  3. 通常時(イベント発生時)
  4. 頻度 (イベントが発生する頻度)
  5. 定常性 (状態が非常に長い間維持されるか、無期限に維持されるか)

このデータセットが、この特定のクラスの推論問題の将来の調査を促進できることを願っています。

スプリット
'test' 9,442
'validation' 3,783
  • 機能構造:
FeaturesDict({
    'answer': Text(shape=(), dtype=string),
    'category': ClassLabel(shape=(), dtype=int64, num_classes=5),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=2),
    'question': Text(shape=(), dtype=string),
    'sentence': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
答え文章
カテゴリークラスラベルint64
ラベルクラスラベルint64
質問文章
文章
  • 引用
@inproceedings{ZKNR19,
    author = {Ben Zhou, Daniel Khashabi, Qiang Ning and Dan Roth},
    title = {"Going on a vacation" takes longer than "Going for a walk": A Study of Temporal Commonsense Understanding },
    booktitle = {EMNLP},
    year = {2019},
}