JMultiWOZ

JMultiWOZ是首个大规模日语多领域任务导向对话数据集，由名古屋大学信息科学研究生院创建。该数据集包含4246个对话，涵盖旅游相关的六个领域，旨在推动日语任务导向对话系统的研究和开发。数据集通过Wizard of Oz方法收集，每个对话由两个人类参与者进行，一人扮演旅行者（用户），另一人作为信息提供者（向导）。JMultiWOZ不仅提供了每轮对话的状态，还包括各领域的数据库，用于实现和基准测试任务导向对话模型。该数据集的应用领域主要集中在对话状态跟踪（DST）和响应生成（RG），旨在解决日语环境下任务导向对话系统的复杂性和多样性问题。

JMultiWOZ is the first large-scale Japanese multi-domain task-oriented dialogue dataset, created by the Graduate School of Informatics, Nagoya University. It contains 4246 dialogues spanning six tourism-related domains, and is designed to advance the research and development of Japanese task-oriented dialogue systems. The dataset is collected via the Wizard of Oz method, where each dialogue involves two human participants: one acting as a traveler (user) and the other as an information provider (wizard). JMultiWOZ not only provides the dialogue state for each conversational turn, but also includes domain-specific databases for implementing and benchmarking task-oriented dialogue models. Its main application focuses on dialogue state tracking (DST) and response generation (RG), and it aims to address the complexity and diversity of task-oriented dialogue systems in Japanese contexts.