CLUTRR v0.1
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/koustuvsinha/clutrr
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个用于关系推理的基准数据集，它涉及基于从亲属关系图中提取的文本故事来预测亲属关系。该数据集特意设计成需要组合泛化能力，包含了干扰信息和更长的推理链。规模上，对于每个可能的路径长度k，从k=3到k=6，都包含了5000个故事。任务则是基于一个文本故事，预测两个实体之间的关系。
This dataset is a benchmark dataset for relational reasoning, which focuses on predicting kinship relationships based on textual stories extracted from kinship graphs. It is deliberately designed to require compositional generalization capabilities, and contains distractor information and longer reasoning chains. For each possible path length k ranging from 3 to 6, the dataset includes 5000 stories. The task is to predict the relationship between two entities based on a given textual story.
