cos_e

  • 説明:

Common Sense Explanations (CoS-E) を使用すると、言語モデルをトレーニングして、新しい Commonsense Auto-Generated Explanation (CAGE) フレームワークでトレーニングおよび推論中に使用できる説明を自動的に生成できます。

スプリット
'train' 9,741
'validation' 1,221
  • 機能構造:
FeaturesDict({
    'abstractive_explanation': Text(shape=(), dtype=string),
    'answer': Text(shape=(), dtype=string),
    'choices': Sequence(Text(shape=(), dtype=string)),
    'extractive_explanation': Text(shape=(), dtype=string),
    'id': Text(shape=(), dtype=string),
    'question': 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
abstractive_explanation文章ストリング
答え文章ストリング
選択肢シーケンス(テキスト) (なし、)ストリング
extractive_explanation文章ストリング
ID文章ストリング
質問文章ストリング
  • 引用
@inproceedings{rajani2019explain,
     title = "Explain Yourself! Leveraging Language models for Commonsense Reasoning",
    author = "Rajani, Nazneen Fatema  and
      McCann, Bryan  and
      Xiong, Caiming  and
      Socher, Richard",
      year="2019",
    booktitle = "Proceedings of the 2019 Conference of the Association for Computational Linguistics (ACL2019)",
    url ="https://arxiv.org/abs/1906.02361"
}