SabrinaSadiekh/mixed_hate_dataset
收藏Hugging Face2025-11-11 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/SabrinaSadiekh/mixed_hate_dataset
下载链接
链接失效反馈官方服务：
资源简介：
Mixed Harm–Safe Statements Dataset是一个用于分析大型语言模型中潜在极性编码的数据集，包含1244个独特的观察值，分为622个有害-安全配对。每个配对表达了从相反伦理视角的相同情况或主张。数据集通过并发基础和否定基础两种策略构建，以测试CCS和PA-CCS对语义和句法极性线索的敏感性。
The Mixed Harm–Safe Statements Dataset is a dataset for analyzing latent polarity encoding in large language models, containing 1,244 unique observations grouped into 622 harm–safe pairs. Each pair expresses the same situation or claim from opposite ethical perspectives. The dataset is constructed using two complementary strategies: concurrent-based and negation-based, to test the sensitivity of CCS and PA-CCS to semantic and syntactic polarity cues.
提供机构：
SabrinaSadiekh
