微软顺序问题解答（SQA）数据集

为了探索对话式质量检查环境，我们提出了一个更为现实的任务：回答一系列简单但相互关联的问题。 我们通过要求众包工作者分解WikiTableQuestions（WTQ）中的2,022个问题来创建SQA，其中包含来自Wikipedia的表格的高度复杂的问题。 我们让三名工作人员分解每个WTQ问题，从而产生了6,066个序列的数据集，总共包含17,553个问题。 每个问题还与表格中单元格位置形式的答案相关联。 相关论文： https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/acl17-dynsp.pdf

To explore conversational quality inspection scenarios, we propose a more realistic task: answering a series of simple yet interconnected questions. We developed the SQA dataset by requiring crowdworkers to decompose 2,022 questions from WikiTableQuestions (WTQ), a dataset containing highly complex questions about tables sourced from Wikipedia. We had three workers decompose each WTQ question, resulting in a dataset of 6,066 sequences with a total of 17,553 questions. Each question is also associated with an answer represented by the positions of relevant cells in the table. Related paper: https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/acl17-dynsp.pdf