StepGame

该数据集名为StepGame，是一个合成的文本问答数据集，其关系数量从1到10不等。对于每个k值从1到5，数据集包含10,000个训练/验证的干净样本；同时，对于k值从1到10，数据集还包含10,000个带噪声的测试样本。测试集中包含三种类型的干扰噪声：不连贯的、不相关的和支持性的。数据规模方面，包含10,000个训练样本，1,000个验证样本以及10,000个测试样本。该数据集的任务是进行多跳空间推理。

This dataset, named StepGame, is a synthetic text-based question answering dataset where the number of relations ranges from 1 to 10. For each k value from 1 to 5, the dataset contains 10,000 clean training/validation samples; meanwhile, for k values from 1 to 10, it also includes 10,000 noisy test samples. The test set encompasses three types of interfering noise: incoherent, irrelevant, and supportive. In terms of data scale, the dataset consists of 10,000 training samples, 1,000 validation samples and 10,000 test samples. The task of this dataset is multi-hop spatial reasoning.