ResQ
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/hlr/spartun
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为ReSQ，是一个众包基准测试，其中只包含是非问题，分别有1008、333和610个示例用于训练、验证和测试。由于数据集由人类生成，它能反映出真实世界中空间描述的自然复杂性。数据规模包括1008个训练示例、333个验证示例以及610个测试示例，任务类型为多跳空间推理。
This dataset, named ReSQ, is a crowdsourced benchmark exclusively containing yes-no questions, with 1008, 333, and 610 examples assigned to the training, validation, and test splits respectively. Since the dataset is generated by human annotators, it captures the natural complexity of spatial descriptions in real-world settings. The full dataset consists of 1008 training examples, 333 validation examples, and 610 test examples, with its task type being multi-hop spatial reasoning.
