GraphWOZ

该数据集名为GraphWOZ，包含了人类参与者与扮演接待员角色的机器人进行对话的“巫师奥兹”式对话，这些对话使用了作为动态知识图谱呈现的对话状态。数据集包含了用户发言的WAV文件、自动语音识别（ASR）转录文本以及JSON格式的标注，并按照大约50%的训练集、25%的开发集和25%的测试集进行划分。规模上，该数据集共有119个对话，每个对话平均包含11-12个回合。任务方面，包括对话管理任务，涉及会话实体链接和响应排序。

This dataset is named GraphWOZ, which contains Wizard-of-Oz style dialogues where human participants converse with robots acting as receptionists, with dialogue states presented as dynamic knowledge graphs. The dataset includes WAV files of user utterances, automatic speech recognition (ASR) transcripts, and JSON-formatted annotations, and is split into training, development, and test sets at an approximate ratio of 50%, 25%, and 25% respectively. In total, the dataset consists of 119 dialogues, with each dialogue containing an average of 11 to 12 conversational turns. The included tasks cover dialogue management, specifically conversational entity linking and response ranking.