ReCoPII: A Relation-Grounded Framework for Building Contextual PII Masking Datasets

查看原文 / View Original 编辑 / Edit

Computer Science > Computation and Language arXiv:2610.35925 (cs) [Submitted on 14 Oct 2026] Title: ReCoPII: A Relation-Grounded Framework for Building Contextual PII Masking Datasets Authors: Taehwan Lee.

计算机科学 > 计算与语言 arXiv:2610.35925 (cs) [提交于 2026 年 10 月 14 日] 标题：ReCoPII：一种用于构建上下文 PII 掩码数据集的关系基础框架作者：Taehwan Lee。

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 augmentation (SPY) labels mentions unrelated to the author as non-PII, presuming a single data subject and overlooking others.

摘要：训练用于个人身份信息 (PII) 的上下文掩码系统需要区分同一实体类型中可掩码与可保留提及的示例，但隐私限制限制了对真实数据的访问。基于作者的增强 (SPY) 将与作者无关的提及标记为非 PII，这预设了单一数据主体并忽略了其他人。

We propose ReCoPII (Relation-grounded Contextual PII masking), a framework that builds such data without this premise: it plans ontology-constrained entity relations, writes insertion sentences with placeholders, reconstructs annotations with code, checks the result with a separate LLM, and derives MASK/KEEP labels from entity relations.

我们提出了 ReCoPII（基于关系的上下文 PII 掩码），这是一个无需此前提即可构建此类数据的框架：它规划本体约束的实体关系，编写带有占位符的插入句子，通过代码重建注释，使用独立的 LLM 检查结果，并从实体关系中导出 MASK/KEEP 标签。

On Korean documents from six domains, it yields ReCoPII-Ko, a corpus with relation, privacy, and MASK/KEEP annotations, which we validate. Across 300 pairs, four annotators agree with 88.42% of its sampled KEEP labels, against 26.27% for SPY-style augmentation of the same sources.

在来自六个领域的韩语文档上，它生成了 ReCoPII-Ko，这是一个包含关系、隐私和 MASK/KEEP 注释的语料库，我们对其进行了验证。在 300 个配对中，四名标注员对其抽样的 KEEP 标签的一致性为 88.42%，而相同来源的 SPY 式增强的一致性仅为 26.27%。

Classifiers trained on it over-mask less than under that reference on a matched test. Contextual masking fine-tuned jointly with a PII detector removes most over-masking of detection alone within a fixed leakage budget; on human-verified entity records, the four detectors meeting it over-mask at most 5.04% with at most 4.18% leakage.

在匹配测试中，基于此训练的分类器比参考方法产生的过度掩码更少。与 PII 检测器联合微调的上下文掩码在固定的泄露预算内消除了大部分仅靠检测产生的过度掩码；在人工验证的实体记录上，满足该条件的四个检测器最多产生 5.04% 的过度掩码，且泄露率最高为 4.18%。

The detector is comparable to XLM-R-large on KDPII with about a quarter of its parameters. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI).

该检测器在 KDPII 上的表现与 XLM-R-large 相当，但参数量仅为其四分之一左右。学科：计算与语言 (cs.CL)；人工智能 (cs.AI)。