Hierarchy-Transformers/WordNetNoun

该数据集是基于WordNet名词的上下位关系（超类层次结构）构建的，主要用于训练和评估层次嵌入模型。数据集包含多跳推理和混合跳预测任务，支持随机负样本和困难负样本，样本格式包括三元组和配对形式。多跳推理任务旨在评估模型从直接的一跳上下位关系推断间接的多跳上下位关系的能力，以模拟传递推理。混合跳预测任务旨在评估模型在任意实体对之间确定上下位关系存在的能力，其中实体在训练期间不一定被看到。

This dataset is a collection of Multi-hop Inference and Mixed-hop Prediction datasets created from WordNets subsumption (hypernym) hierarchy of noun entities for training and evaluating hierarchy embedding models. The dataset includes multiple configurations for Multi-hop Inference and Mixed-hop Prediction tasks, with different negative sampling strategies and sample structures. Each subset follows the naming convention `TaskType-NegativeType-SampleStructure`, where `TaskType` indicates the type of hierarchy evaluation task, `NegativeType` specifies the strategy used for negative sampling, and `SampleStructure` indicates the format of the samples. The dataset is suitable for training and evaluating hierarchy embedding models, supporting various task types and sample structures.