知识卡片
隐私标签体系需要任务约定
命名实体不全是需要保护的个人信息,哪些属性、人物与上下文应标记也可能有不同判断。应先写清标签和例外规则,再用标注分歧发现歧义;不能把文化或规范分歧全当作标注员失误。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 预训练数据
原书出处 · 1 处
- · 查看原书依据
structured tags in this list like emails and government IDs can be identified using regular expressions. For other tags, we annotated datasets that could then be used to train Transformer-based NER-like models. Interestingly, we observed a very high degree of inter-annotator disagreement (same example being annotated differently by different people) that underscored the cultural nuances of the definition of privacy and what constitutes personal information.
参考来源
- 位置:《Designing Large Language Model Applications》第02章《预训练数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html)
- 结论依据:摘录中关于隐私标签体系需要任务约定的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:structured tags in this list like emails and government IDs can be identified using regular expressions. For other tags, we annotated datasets that could then be used to train Transformer-based NER-like models. Interestingly, we observed a very high degree of inter-annotator disagreement (same example being annotated differently by different people) that underscored the cultural nuances of the definition of privacy and what constitutes personal information.