StereoSet

该数据集是一个大规模的自然英语数据集，涵盖了性别、职业、种族和宗教四个领域，旨在衡量语言模型中存在的刻板印象偏差。 StereoSet 数据集不仅包含了开发集，还包含了一个隐藏的测试集，其重点关注在不对数据集进行微调的情况下，测量预训练语言模型中的偏差。这一数据集的规模属于大规模，任务则是衡量语言模型中的刻板印象偏差。

StereoSet is a large-scale natural English dataset covering four domains: gender, occupation, race, and religion, which is designed to measure stereotype bias in language models. It not only includes a development set but also a hidden test set, which focuses on evaluating bias in pre-trained language models without fine-tuning on the dataset. The core task of this dataset is to measure stereotype bias in language models.

搜集汇总

背景与挑战

背景概述

StereoSet是一个用于测量预训练语言模型中刻板偏见的工具和数据集，提供了评估偏见的方法和代码，旨在促进对语言模型偏见的进一步研究和减轻。