Selected Adversarial SemanticS (SASS)
收藏arXiv2023-01-05 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2301.01874v1
下载链接
链接失效反馈官方服务：
资源简介：
SASS数据集是由纽约大学开发的一个用于评估和挑战现有毒性语言检测工具的基准。该数据集包含250个自然语言示例，覆盖10个毒性类别，如刻板印象、阶级歧视和勒索等。每个示例约1-2句话，旨在揭示毒性检测系统中的漏洞。SASS数据集的创建旨在帮助研究人员和公司改进毒性内容检测技术，以更有效地管理和减少社交媒体和在线论坛中的有害内容。
The SASS dataset is a benchmark developed by New York University for evaluating and challenging existing toxic language detection tools. This dataset contains 250 natural language examples spanning 10 toxicity categories, including stereotypes, class discrimination, blackmail, and others. Each example consists of approximately 1 to 2 sentences, designed to uncover vulnerabilities in toxicity detection systems. The SASS dataset was created to help researchers and companies improve toxic content detection technologies, enabling more effective management and reduction of harmful content on social media and online forums.
提供机构：
纽约大学
创建时间：
2023-01-05
