ART Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://arxiv.org/abs/2002.03159
下载链接
链接失效反馈官方服务：
资源简介：
该数据集专为归纳常识推理而设计，旨在对部分可观察的过去和未来生成可能解释的有效假设。在ART数据集中，每个示例都配有1至5个参考答案。平均而言，输入和输出的长度分别为17.4和10.8个词。数据集的规模包括50,481个训练样本，1,779个开发样本以及3,560个测试样本。其任务定位于归纳常识推理。
This dataset is designed for inductive commonsense reasoning, with the objective of generating valid hypotheses that provide plausible explanations for partially observable past events and future situations. In the ART dataset, each example is paired with 1 to 5 reference answers. On average, the input and output sequences have lengths of 17.4 and 10.8 words respectively. The dataset comprises 50,481 training samples, 1,779 development samples, and 3,560 test samples. Its targeted task is inductive commonsense reasoning.
提供机构：
Benchmark
