BriefGPT.xyz
Jun, 2021
通过不变量合理化降低有害语言检测中的偏见
Mitigating Biases in Toxic Language Detection through Invariant Rationalization
HTML
PDF
Yung-Sung Chuang, Mingye Gao, Hongyin Luo, James Glass, Hung-yi Lee...
TL;DR
通过使用不变量理性化 (InvRat) 方法，我们可以降低对某些语法模式的误判，从而避免使用带有偏见的训练数据集导致毒性过滤器产生偏见，进而加剧群体边缘化的现象。
Abstract
Automatic detection of
toxic language
plays an essential role in protecting
social media
users, especially
minority groups
, from verbal ab
→