zenless-lab/jsem

JSeM数据集是一个用于日语语义测试的基准数据集，主要用于测试叙述文之间的蕴含关系。数据集包含三个主要特征：前提（premise）、假设（hypothesis）和标签（label），标签分为三类：蕴含（entailment）、中立（neutral）和矛盾（contradiction）。数据集分为训练集、验证集和测试集，分别包含12667、1583和1584个样本。数据集的创建者是DaisukeBekki，语言为日语，采用BSD 3-Clause许可证。数据集的设计基于FraCaS测试套件，并扩展了日语特有的语言现象。

The JSeM dataset is a benchmark dataset for Japanese semantic testing, primarily used to test the entailment relationships between narrative texts. The dataset includes three main features: premise, hypothesis, and label, with the label categorized into three types: entailment, neutral, and contradiction. The dataset is divided into training, validation, and test sets, containing 12667, 1583, and 1584 samples respectively. The dataset was created by DaisukeBekki, is in Japanese, and is licensed under BSD 3-Clause. The design of the dataset is based on the FraCaS test suite and extends to include phenomena unique to the Japanese language.