RiSAWOZ中文多领域任务型对话数据集

RiSAWOZ是一个具有丰富语义信息标注的大规模中文多领域任务型对话数据集，它包含1.12万个已标注的人-人多轮对话，总对话轮数超过15万轮，覆盖12个领域。其中，单领域和多领域的对话分别占数据集的65%和35%。每个对话都进行了广泛的对话信息标注，包括以自然语言描述的对话目标、所属领域、对话状态以及用户和系统双方的对话动作信息。除了以上的对话标注信息之外，RiSAWOZ还提供了关于对话语篇现象（省略和共指）的语言学标注，该标注信息对解决对话中的省略和共指问题具有重要价值。

RiSAWOZ is a large-scale Chinese multi-domain task-oriented dialogue dataset with rich semantic annotations. It contains 11,200 annotated human-human multi-turn dialogues, with a total of over 150,000 dialogue turns, covering 12 domains. Single-domain and multi-domain dialogues account for 65% and 35% of the dataset respectively. Each dialogue is comprehensively annotated with diverse dialogue-related information, including dialogue goals described in natural language, affiliated domains, dialogue states, and dialogue act information of both users and systems. In addition to the above-mentioned dialogue annotation contents, RiSAWOZ also provides linguistic annotations for dialogue discourse phenomena (ellipsis and coreference), which holds significant value for resolving ellipsis and coreference issues in dialogues.