语言:

• 英文

数据集:

• winogrande

示例:

• 文本: "Rachel家的屋顶又旧又破，而Betty家的很新。因此， Rachel的房屋价值更低。"
• 文本: "我朋友工作场所的木门比我工作场所的木桌更差，因为 桌子的材料更便宜。"
• 文本: "如果邮政服务减少投递频率。 邮政服务可以降低投递频次。"
• 文本: "我把蛋糕放进冰箱了。它里面有很多黄油。 蛋糕含有大量黄油。"

基于Winogrande微调的RoBERTa Large模型

本模型在Winogrande数据集（XL版本）上以序列分类任务形式进行微调，原始成对句子及其对应填空选项被拆分、打乱后作为独立样本进行分类。

模型说明

使用范围与限制

使用方法

训练数据

WinoGrande-XL数据集经以下方式重构：

1. 每个句子以"_"占位符为界拆分
2. 每个选项与拆分后的第二部分拼接，将每个样本转化为两个文本段组合
3. 正确/错误选项对应的文本段分别标记为True/False
4. 文本段组合后续被打乱

例如：

{
  "answer": "2",
  "option1": "plant",
  "option2": "urn",
  "sentence": "The plant took up too much room in the urn, because the _ was small."
}

转化为：

{
  "sentence1": "The plant took up too much room in the urn, because the ",
  "sentence2": "plant was small.",
  "label": false
}

和

{
  "sentence1": "The plant took up too much room in the urn, because the ",
  "sentence2": "urn was small.",
  "label": true
}

这些句子组合随后被视为独立样本。

文献引用信息

@article{sakaguchi2019winogrande,
    title={WinoGrande: An Adversarial Winograd Schema Challenge at Scale},
    author={Sakaguchi, Keisuke and Bras, Ronan Le and Bhagavatula, Chandra and Choi, Yejin},
    journal={arXiv preprint arXiv:1907.10641},
    year={2019}
}

@article{DBLP:journals/corr/abs-1907-11692,
  author    = {Yinhan Liu and
               Myle Ott and
               Naman Goyal and
               Jingfei Du and
               Mandar Joshi and
               Danqi Chen and
               Omer Levy and
               Mike Lewis and
               Luke Zettlemoyer and
               Veselin Stoyanov},
  title     = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},
  journal   = {CoRR},
  volume    = {abs/1907.11692},
  year      = {2019},
  url       = {http://arxiv.org/abs/1907.11692},
  archivePrefix = {arXiv},
  eprint    = {1907.11692},
  timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}