POPQUORN

POPQUORN是由密歇根大学信息学院创建的大型NLP数据集，包含45,000条来自1,484名代表性样本的注释，涵盖性别、年龄和种族。数据集包含四个多样化的NLP任务：攻击性检测、问题回答、文本重写和礼貌评级。创建过程考虑了注释者的背景多样性，以减少数据集偏见。该数据集旨在解决NLP任务中的注释者背景影响问题，提高模型的公平性和准确性。

POPQUORN is a large-scale NLP dataset developed by the School of Information at the University of Michigan. It encompasses 45,000 annotations sourced from 1,484 representative annotators, with demographic coverage across gender, age and race. The dataset features four distinct NLP tasks: aggressive detection, question answering, text rewriting, and politeness rating. The construction of this dataset prioritizes annotator background diversity to mitigate dataset bias. Its primary goal is to address the confounding effect of annotator demographic backgrounds on NLP task outcomes, and enhance the fairness and accuracy of trained models.