ReCoPII：基于关系的上下文PII掩码数据集构建框架
原标题：ReCoPII: A Relation-Grounded Framework for Building Contextual PII Masking Datasets
AI 摘要
研究者提出 ReCoPII 框架，通过本体约束的实体关系规划、LLM 生成插入句与独立 LLM 复核，从 PII 标注文档构建上下文掩码数据集，不再预设单一数据主体。在六个领域韩语文档上构建 ReCoPII-Ko 语料，四名标注者对其 KEEP 标签一致率达 88.42%，远高于 SPY 式增强的 26.27%。基于该数据训练的模型过度掩码更少，联合微调的检测器在固定泄漏预算下过度掩码不超过 5.04%。
正文节选
[ BoldFont=texgyretermes-bold.otf, ItalicFont=texgyretermes-italic.otf, BoldItalicFont=texgyretermes-bolditalic.otf ] [Path=fonts/,BoldFont=NanumGothicBold.ttf] ReCoPII: A Relation-Grounded Framework for Building Contextual PII Masking Datasets Abstract Training contextual masking systems for personally identifiable information (PII) requires examples distinguishing maskable from preservable mentions of one entity type, but privacy restrictions limit access to real data. Author-based augmentatio