MultiWOZ 2.3

MultiWOZ 2.3是由华为技术有限公司人工智能应用研究中心和清华大学合作开发的多域任务导向对话数据集，包含超过10,000条对话，覆盖八个不同领域。该数据集通过纠正对话状态和对话行为的标注错误，并引入共指标注，提高了数据集的质量。创建过程中，研究团队对原始数据集进行了详细的错误分析和修正，确保了对话状态和对话行为之间的一致性。MultiWOZ 2.3主要应用于自然语言理解和对话状态跟踪领域，旨在解决现有模型在处理多域交互时性能不佳的问题。

MultiWOZ 2.3 is a multi-domain task-oriented dialogue dataset co-developed by the AI Application Research Center of Huawei Technologies Co., Ltd. and Tsinghua University. It contains over 10,000 dialogues spanning eight distinct domains. This version improves dataset quality by correcting annotation errors in dialogue states and dialogue acts, and introducing coreference annotations. During its development, the research team conducted detailed error analysis and corrections on the original dataset to ensure consistency between dialogue states and dialogue acts. Primarily applied in the fields of natural language understanding and dialogue state tracking, MultiWOZ 2.3 aims to address the underperformance of existing models when handling multi-domain interactions.