DuReader_QG问题生成数据集

DuReader robust旨在利用真实应用中的数据样本来衡量阅读理解模型的鲁棒性，评测模型的过敏感性、过稳定性以及泛化能力，是首个中文阅读理解鲁棒性数据集。 DuReader_QG是从DuReader robust中选择的问题生成任务子集DuReader robust is designed to measure the robustness of reading comprehension models using real-world application data samples, and evaluate three key properties of the models: over-sensitivity, over-stability and generalization ability. It is the first Chinese reading comprehension robustness dataset. DuReader_QG is a subset selected from DuReader robust for the question generation task.