NoRA (Non-Path Reasoning with Ambiguous Facts)
收藏资源简介:
NoRA 数据集是一个用于测试系统神经关系推理能力的新基准。该数据集旨在挑战现有模型,要求它们能够处理更加复杂的关系推理任务,如考虑多个关系路径、处理模糊事实等。数据集中的实例是基于故事和事实生成的,每个故事都包含至少一个答案集,模型需要从中推断出实体之间的关系。该数据集旨在推动系统神经关系推理领域的发展,并评估模型的泛化能力。
The NoRA Dataset is a novel benchmark designed to test systematic neural relational reasoning capabilities. This dataset aims to challenge existing models by requiring them to handle more complex relational reasoning tasks, such as considering multiple relational paths and dealing with ambiguous facts. Instances in the dataset are generated based on stories and facts, with each story containing at least one answer set, from which models are required to infer the relationships between entities. This dataset is intended to advance the development of the systematic neural relational reasoning field and evaluate the generalization ability of models.




