e-QRAQ

e-QRAQ是由IBM T. J. Watson研究中心和马萨诸塞大学阿默斯特分校联合开发的多轮推理数据集，旨在测试代理在处理含糊文本时的能力，包括提问、推理和回答，并解释其推理过程。数据集通过模拟用户提供简短、含糊的故事和挑战性问题，故事中的实体被变量替换，增加了推理难度。数据集创建过程中，通过模拟器动态生成解释，帮助理解代理的响应是否正确及其原因。e-QRAQ数据集适用于开发和测试可解释的机器学习算法，特别是在需要透明推理过程的领域，如医疗诊断和法律合规。

e-QRAQ is a multi-turn reasoning dataset jointly developed by IBM T. J. Watson Research Center and the University of Massachusetts Amherst. It is designed to test agents' capabilities in dealing with ambiguous texts, including question asking, reasoning, answering, and explaining their own reasoning processes. The dataset provides short, ambiguous stories and challenging questions via simulation, where entities within the stories are replaced with variables to increase the difficulty of reasoning. During the dataset construction, a simulator dynamically generates explanations to help assess whether an agent's response is correct and the rationale behind it. The e-QRAQ dataset is applicable to the development and testing of explainable machine learning algorithms, particularly in domains that demand transparent reasoning procedures, such as medical diagnosis and legal compliance.