CODAH

CODAH数据集是由西北大学计算机科学系创建的一个对抗性构建的常识问答数据集。该数据集包含2801个多选句子完成问题，旨在测试和挑战现有的神经问答系统。数据集的创建过程中，参与者被鼓励针对最先进的神经问答系统的弱点设计问题，并通过实时反馈机制来优化问题的难度。CODAH数据集特别关注于提高机器的常识推理能力，尤其是在量化推理、否定和对象引用等方面。该数据集的应用领域主要集中在提升自然语言处理系统的常识推理能力，以解决现有系统在处理复杂常识问题时的局限性。

The CODAH dataset is an adversarially constructed commonsense question answering dataset created by the Department of Computer Science at Northwestern University. It contains 2,801 multiple-choice sentence completion questions designed to test and challenge existing neural question answering systems. During the dataset creation process, participants were encouraged to design questions targeting the weaknesses of state-of-the-art neural question answering systems, and optimize the difficulty of the questions through a real-time feedback mechanism. The CODAH dataset specifically focuses on enhancing machines' commonsense reasoning capabilities, particularly in areas such as quantitative reasoning, negation, and coreference resolution of objects. The primary applications of this dataset center on improving the commonsense reasoning capabilities of natural language processing systems, to address the limitations of existing systems when handling complex commonsense problems.