SocialStigmaQA
收藏arXiv2023-12-28 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2312.07492v4
下载链接
链接失效反馈官方服务：
资源简介：
SocialStigmaQA是一个专注于检测生成语言模型中社会偏见放大的综合基准数据集。该数据集由IBM研究院创建，包含约10,360个精心设计的提示，涉及93种美国特有的社会偏见情况。数据集通过问答形式，旨在测试模型在处理社会偏见和模型鲁棒性方面的表现。SocialStigmaQA的应用领域主要集中在评估和减少语言模型中的社会偏见，以避免对特定群体造成不公平的影响。
SocialStigmaQA is a comprehensive benchmark dataset dedicated to detecting social bias amplification in generative language models. Developed by IBM Research, it contains approximately 10,360 meticulously designed prompts covering 93 U.S.-specific social bias scenarios. Presented in a question-and-answer format, this dataset aims to evaluate models' performance in handling social biases and their robustness. The primary application areas of SocialStigmaQA focus on assessing and reducing social biases in language models, so as to avoid causing unfair impacts on specific groups.
提供机构：
IBM研究院
创建时间：
2023-12-13
