Hierarchy-Transformers/Schemaorg

该数据集是从Schema.org的子集层次结构（TBox）创建的混合跳预测数据集的集合，用于评估层次嵌入模型。它是一个仅用于评估的数据集，仅包含验证和测试集。数据集包含多个子集，每个子集根据任务类型、负样本类型和样本结构进行命名。混合跳预测任务旨在评估模型在确定任意实体对之间的子集关系存在性方面的能力，这些实体在训练期间不一定被看到。

This dataset is a collection of Mixed-hop Prediction datasets created from Schema.orgs subsumption hierarchy (TBox) for evaluating hierarchy embedding models. It is an evaluation-only dataset consisting of just validation and test splits. The subsets of the dataset are named according to the task type (Mixed-hop), negative sampling strategy (random negatives or hard negatives), and sample structure (triplets or pairs). Each sample can be structured as a triplet (child, parent, negative) or a pair (child, parent, label), where label=1 denotes a positive subsumption and label=0 denotes a negative subsumption. The dataset is used for both evaluation and training, depending on the models design and specific requirements.