arXiv：cs.LG· Qiming Bao, Sherry J. H. Feng, Kim Chester Eugenio, Meng Fon·· 4 小时前
替代替换保持 PHI 可检测性：多检测器等价性研究
Surrogate Substitution Preserves PHI Detectability: A Multi-Detector Equivalence Study
AI 导读
一项 arXiv 研究提出"可检测性保持"指标，在 7 个临床与通用 PII 基准（7 种语言、1,750 篇文档、5,333 个被替换片段、11 个检测器）上检验结构保持去标识化：某商业转换后掩码片段召回率从 75.8% 降至 74.6%，变化 90% CI 为 [-1.62, -0.82] 点，后验概率超 0.99 落在 ±2 点内。
正文
Abstract:Structure-preserving de-identification replaces each detected identifier with a realistic same-type surrogate ("Anna S." -> "Maria S.") rather than a tag such as [NAME]. If downstream PHI detectors no longer recognise surrogates, hiding-in-plain-sight protection and second-pass safety nets weaken. We ask: on the spans a de-identifier actually replaced, do detectors find the surrogate as often as the original value? We call this detectability preservation and treat it as a proxy, not a full measure of utility. Our paired protocol (i) scores detectability only on replaced spans, separately from coverage; (ii) runs an identical heterogeneous detector panel on original and transformed text; and (iii) bounds the paired change with document-level cluster-bootstrap equivalence testing and posterior probabilities instead of significance tests. For one commercial transform on 7 clinical and general-PII benchmarks in 7 languages (1,750 documents, 5,333 replaced spans, 11 detectors), masked-span recall moves from 75.8% to 74.6% (90% CI of the change [-1.62, -0.82] points; posterior probability > 0.99 of lying within +/-2 points). The change reaches 2-3 points on dense Spanish and Dutch text, and replaced spans are easier to detect than unreplaced ones (+16.6 points), so the result describes the replaced population only. Coverage is a separate, weaker axis (48% of gold PHI on MEDDOCAN). On financial, legal and enterprise text (750 documents, 10 detectors) the pooled result replicates (72.4% -> 72.8%), but rule-based and fine-tuned detectors lose 2-4 points on finance, and the transform edits substantial non-PII text. Redaction and Faker baselines give reference points. We release per-span predictions and analysis code.
| Comments: | v2: corrects the gold-span re-projection and statistical analysis of v1 (see Appendix A) and adds financial, legal and enterprise experiments. 15 pages, 3 figures, 13 tables. Code, data and per-span predictions: this https URL ; dashboard: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2608.03172 [cs.AI] |
| (or arXiv:2608.03172v5 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2608.03172 arXiv-issued DOI via DataCite |
Submission history
From: Qiming Bao [view email]
[v1]
Tue, 4 Aug 2026 06:04:18 UTC (23 KB)
[v2]
Thu, 6 Aug 2026 08:29:28 UTC (23 KB)
[v3]
Thu, 13 Aug 2026 03:45:55 UTC (27 KB)
[v4]
Mon, 12 Oct 2026 09:26:12 UTC (28 KB)
[v5]
Wed, 14 Oct 2026 07:46:56 UTC (28 KB)
来源：arXiv：cs.LG · arxiv.org