查找个人数据
阅读约需 116分钟 分钟,13节中含11个示例。
读完本章,你将能够在数据平台上自动查找个人数据,并说明漏掉了多少。你会根据列名、类型和可复现的取值样本,为每一列提出分类建议;在自由文本中逐个片段地找出个人数据;并在任何人依赖某个检测器之前,用带标注的数据对它进行度量。在这个过程中,你会离线运行 Presidio 的模式识别器,其阈值由度量结果确定;会把它们与一个专门训练来查找个人数据的模型所记录的检测结果相比较;还会只对模式变更新增的列重新分类。
平台上的每一列都已经有了分类。第 2 章把分类写进了 governance/metadata.toml,这是一个供人审阅的文件,而经过审阅的文件记录的只是审阅者认为某一列包含什么。在这个在线学习平台上(本书治理的正是它的数据仓库),有三项这样的判断从未被检验过。billing.payments.card_number 是一个遗留列,其描述称它存放完整的银行卡号,但没有人查看过。crm.learners.notes 是客服人员输入的自由文本,因此它的名称无法说明里面有什么。学习者在工单和课程评价中想写什么就写什么。对整列做出的任何分类都无法保护这些文本。在对其进行掩码之前,平台必须逐个片段地找出其中的每个姓名、电子邮件地址和电话号码,就像第 9 章在课程助手学习评价内容之前对评价所做的那样。
本章为这项工作构建本书共用的模块 lib.pii,其中包括基于列名的规则、带随机种子的列取值样本、配置为离线运行的 Presidio 检测器,以及用带标注的数据为任意检测器打分的代码。本章中的每个检测器都以同一个文件为标准来度量,我们称之为基准真值:data/platform/pii_spans.parquet。平台的数据生成器会把它与工单和评价一起写出,其中记录了生成器放入文本的每一项个人数据及其准确的起止位置。发票号、订单号以及问题发生的日期则不做标注,因为平台不把它们视为个人数据。
如数据卡所述,本书中的电话号码、银行卡号和 IBAN 都是随机生成的,所有示例都不会完整打印其中任何一个:它们只保留最后四位数字,其余均被掩去。
根据表结构对列分类
关于一列,成本最低的证据是它的表结构:列名和类型。基于列名的规则不读取任何数据,因此可以在一次拉取请求中遍历整个目录,并找出那些名称本身就说明内容的列,例如 email 或 date_of_birth。目录工具也是这样起步的:OpenMetadata 的自动分类首先使用列名扫描器,用正则表达式检查列名(OpenMetadata 文档)。接下来怎么做,由类型决定。文本列的值可以抽样并扫描,但时间戳不可能存放电子邮件地址,用日期模式去扫描时间戳,会把每一行都标记出来。因此,数值、日期、时间戳和布尔值只按列名分类。
平台把规则保存在 governance/pii.toml 中。该文件包含检测器的全部设置,并与元数据一样接受评审。这些规则是 8 条针对小写列名的正则表达式,每条都指明它所提示的实体,并以第一个匹配为准。同一文件还把每种实体映射到一个分类:电话号码对应 personal,银行卡号或 IBAN 对应 sensitive。lib.pii.by_name 负责应用这些规则。规则会遍历目录中的每一列,结果与经过评审的元数据并列存放:
1"""Runs the name rules over every column of the catalog and compares each column they mark with the reviewed metadata.2The prose rests on the count of marked columns that agree, and on the personal columns no rule marks."""3from pyiceberg.types import StringType45from lib import metadata, pii6from lib.catalog import open_catalog78catalog = open_catalog()9described = metadata.load() # the reviewed metadata, keyed by table and then by field ID10# every column as (table, field): the field carries the ID the metadata is keyed by, so a renamed column still matches11columns = [(table, f) for (namespace,) in sorted(catalog.list_namespaces())12 for _, name in sorted(catalog.list_tables(namespace))13 for table in [f"{namespace}.{name}"] for f in catalog.load_table(table).schema().fields]14print("columns in the catalog:", len(columns))15print("name rules:", len(pii.NAME_RULES))1617named = [(table, f, pii.by_name(f.name)) for table, f in columns if pii.by_name(f.name)]18print("columns a name rule marks:", len(named))19for table, f, entity in named:20 print(f" {table}.{f.name}: {entity}, so {pii.CLASSIFICATION[entity]}")21agree = sum(pii.CLASSIFICATION[entity] == described[table].columns[f.field_id].classification22 for table, f, entity in named)23print("of them, agreeing with the metadata:", agree)2425# Only text is sampled: numbers, dates, timestamps and booleans are classified by name alone.26text = [(table, f) for table, f in columns if isinstance(f.field_type, StringType)]27print("columns of text, whose values can be sampled:", len(text))2829# Columns the metadata, reviewed by people, calls personal or sensitive, that no name rule marks.30missed = [f"{table}.{f.name}" for table, f in columns if not pii.by_name(f.name)31 and described[table].columns[f.field_id].classification in ("personal", "sensitive")]32print("personal columns no name rule marks:", ", ".join(missed))
python column-name-rules.pycolumns in the catalog: 64
name rules: 8
columns a name rule marks: 11
billing.payments.card_number: CREDIT_CARD, so sensitive
crm.instructors.full_name: PERSON, so personal
crm.instructors.email: EMAIL_ADDRESS, so personal
crm.instructors.payout_iban: IBAN_CODE, so sensitive
crm.learners.full_name: PERSON, so personal
crm.learners.email: EMAIL_ADDRESS, so personal
crm.learners.phone: PHONE_NUMBER, so personal
crm.learners.city: LOCATION, so personal
crm.learners.street_address: LOCATION, so personal
crm.learners.date_of_birth: DATE_TIME, so personal
learning.events.ip_address: IP_ADDRESS, so personal
of them, agreeing with the metadata: 11
columns of text, whose values can be sampled: 29
personal columns no name rule marks: crm.learners.notes, learning.reviews.body, support.tickets.body
凡是规则触发的地方,结果都与元数据一致,但规则会漏掉自由文本。在目录的 64 列中,规则标记了 11,其中 11与人工给出的分类一致。规则漏掉了评审人员分类为 personal 的三列:crm.learners.notes、learning.reviews.body 和 support.tickets.body。这三列都是自由文本,列名描述的是容器,而不是内容。平台最担心的被标记列是 billing.payments.card_number,它仅凭列名就被归为 sensitive,却没有人了解其中实际存放什么。完整的卡号、末四位数字或支付服务商的令牌,都会匹配 card_?(number|num|no)。
列名规则的脆弱性是已知的。用于检测列类型的模型 Sherlock 的作者指出,列名匹配对脏数据不够稳健,而且只能检测少数几种类型(Hulsebos 等,2019)。因此,平台也读取值,范围是其目录中的 29 文本列。
工作准则: 先对每一列运行列名规则,因为它们没有成本,并且只允许它们提高分类级别,不允许降低。规则要保持简短,写在代码中,并像其他规则一样接受评审。任何自由文本列,无论叫什么名字,都应视为需要对其值抽样的列;其他类型的列只按列名分类。
这部分尚未翻译,因此以英文显示。
Run Presidio offline
To read values, the platform uses Presidio, an open-source library for finding and masking personal data. Presidio's analyzer runs recognizers: regular expressions with checksums, phone-number parsing, and a named-entity model reached through an NLP engine. The NLP engine does two jobs, named-entity recognition and feature extraction for rules, such as the context words that raise a score. Its default model is spaCy's en_core_web_lg (Presidio docs). At 2.2.366, AnalyzerEngine() with no arguments downloads that model when it isn't installed (source). So a pipeline without network access fails on its first line, and one with access installs a model of about 400 MB as a side effect (en_core_web_lg 3.8.0).
lib.pii.detector configures the engine itself, with NoOpNlpEngine, which loads no language model (source). What that costs is context. With a model, Presidio raises a pattern's score when a word such as "phone" appears near the match, by comparing the lemma of each nearby word with each recognizer's context terms (Presidio's tutorial on context). With no model there are no lemmas, so the engine can't use the text's context words at all. The engine has one more way onto the network. Its email recognizer checks domains with tldextract, which fetches the Public Suffix List the first time it's used (tldextract 5.4.0's README). Importing Presidio imports tldextract, whose README says an empty TLDEXTRACT_PUBLIC_SUFFIX_LIST_URLS turns the fetch off if it's set before the import. So lib/__init__.py, which Python runs before any module of the book's lib, empties the variable. If an example imported Presidio first, the same file replaces tldextract's shared extractor with one that has no URLs. lib.pii refuses to load if any remain, and tldextract uses the copy of the list it ships with, so the same text gives the same answer on every machine.
The defaults are tested on real text: load every predefined recognizer on one of the platform's support tickets, which holds a name, an email, a phone number and a date:
1"""Loads Presidio's predefined recognizers on an engine that loads no language model, and runs them on one support2ticket. The prose's point rests on the detections printed under the ticket, and on the assert that the model isn't3installed."""4import spacy56from lib import pii7from lib.platform import load89# AnalyzerEngine() with no arguments loads spaCy's en_core_web_lg, and downloads it when it isn't installed.10installed = spacy.util.is_package("en_core_web_lg")11assert installed is False # the pipeline assumes no model: a machine that has one fails here, not silently12print("en_core_web_lg installed:", installed)13print("URLs tldextract would fetch the Public Suffix List from:", len(pii.suffix_list_urls()))1415detector = pii.detector() # every predefined recognizer, on an NLP engine that loads no model16names = sorted(r.name for r in detector.registry.recognizers)17print("recognizers loaded:", len(names))18print("recognizers:", ", ".join(names))19print("default score threshold:", detector.default_score_threshold)2021# The ticket holds a phone number, which may be in service: masked() prints it with only its last four digits.22ticket = pii.texts("tickets")[1421] # one ticket with a name, an email, a phone number and a date of the problem23found = sorted(pii.detect(detector, ticket))24print("ticket 1421:", pii.masked(ticket))25for start, end, entity, score in found:26 print(f" {entity} on {pii.masked(ticket[start:end])}: {score:.2f}")2728# Where is the phone number valid? The recognizer reads it in each of its regions.29phone = next(ticket[start:end] for start, end, entity, _ in found if entity == "PHONE_NUMBER")30[learner] = [row["learner_id"] for row in load("support.tickets").to_pylist() if row["ticket_id"] == 1421]31[country] = [row["country"] for row in load("crm.learners").to_pylist() if row["learner_id"] == learner]32print("the learner's country:", country)33# The recognizer's built-in regions, not the learner's country.34print("default regions where it's valid:",35 ", ".join(r for r in pii.PhoneRecognizer.DEFAULT_SUPPORTED_REGIONS if pii.valid_phone(phone, r)))
python presidio-offline-engine.pyen_core_web_lg installed: False
URLs tldextract would fetch the Public Suffix List from: 0
recognizers loaded: 17
recognizers: CreditCardRecognizer, CryptoRecognizer, DateRecognizer, EmailRecognizer, IbanRecognizer, IpRecognizer, MacAddressRecognizer, MedicalLicenseRecognizer, NhsRecognizer, PhoneRecognizer, UrlRecognizer, UsBankRecognizer, UsItinRecognizer, UsLicenseRecognizer, UsPassportRecognizer, UsSsnRecognizer, UuidRecognizer
default score threshold: 0
ticket 1421: I can't open the videos in Introduction to Design Systems, they stop after a few seconds. It happened on 12/03 around 9pm. My colleague Jennifer Abbott (jennifer.abbott@example.net) wants to enrol as well. You can reach me on •••••••0440 after 6pm.
DATE_TIME on 12/03: 0.10
EMAIL_ADDRESS on jennifer.abbott@example.net: 1.00
URL on example.net: 0.50
PHONE_NUMBER on •••••••0440: 0.40
US_BANK_NUMBER on •••••••0440: 0.05
US_DRIVER_LICENSE on •••••••0440: 0.01
the learner's country: US
default regions where it's valid: BR
Presidio's defaults on this engine see the email and the phone number, and miss the colleague's name. en_core_web_lg isn't installed, so a bare AnalyzerEngine() would try to download it, and tldextract has 0 URLs to fetch its list from. The no-model engine loads 17 recognizers, and none of them looks for names, so Jennifer Abbott, the colleague in the ticket, goes unnoticed. Her email is found with a score of 1.00, and its domain is found again as a URL, at 0.50. The learner's phone number, an 11-digit number typed without separators, is found three times: as a phone number at 0.40, a US bank number at 0.05 and a US driver's licence at 0.01. Even the phone number is found by chance. The learner's country is US, but of Presidio's default regions, the US among them, the digits are valid only in BR. The date of the problem, 12/03, is read as a date too, which says nothing about anyone.
Those scores aren't probabilities. Each pattern carries a constant score, and weak patterns get low ones, as Presidio's tutorial explains for a five-digit zip code (Presidio docs). A US bank number is any run of 8 to 17 digits (supported entities), and the FAQ admits that a driver's licence "could be any 9-digit number" (FAQ). The analyzer's default threshold is 0, so every weak match comes back.
Working rule: Configure Presidio's engine explicitly: NoOpNlpEngine when you run patterns only, or a pinned model you install yourself when you need its names. Never call AnalyzerEngine() bare in a pipeline, and pin anything a recognizer would download, such as the Public Suffix List. Read each score as a constant set for its pattern, never as a probability.
每次都以相同方式抽样一列的值
对每个文本列的每个值都做扫描,对任何数据仓库而言都太慢,因此各类工具都采用抽样。OpenMetadata 的工作流会抽取若干行,并用实体识别引擎对其进行检查(OpenMetadata 文档)。抽样需要做三项决定。抽哪些行:应当随机抽取,而不是取前几行,也不是从随机起点开始取一个连续块。Google 的 Sensitive Data Protection 提供了一种随机起点模式 RANDOM_START,它从随机的某一行开始,向后连续读取(Google Cloud 文档)。在按注册日期排序的表上,这样的连续块只是一批学习者。每次运行抽到相同的行,这样分类结果可以复现,结果一旦变化,就说明数据发生了变化。最后是抽多少,这取决于必须捕获的数据有多稀少。
DuckDB 的蓄水池抽样可以接受种子,但正如 DuckDB 2.0.0 文档所警告的,只有 DuckDB 在单线程下运行时,种子才能给出可复现的样本。lib.pii.sample 会将线程数设为 1,滤除 NULL,然后用种子抽取蓄水池样本。
稀少的值需要更大的样本,具体大多少可以由算式给出。设含有个人数据的值所占比例为 p。n 个抽样值中至少包含一个这类值的概率为
概率 = 1 − (1 − p)ⁿ
使该概率达到 99% 的 n 为 ln(0.01) / ln(1 − p)。这是一条经验法则:它把各次抽取视为相互独立,而从数千行中不放回抽样,几乎就是如此。
下面用这个公式来实际计算:从 crm.learners 中两次抽取 20 条备注,然后比较按全部行计算的概率与只按备注计算的概率。
1"""Draws the same twenty notes twice from a seeded reservoir sample, then counts the notes that hold personal data, to see2how many sampled values a rare value needs. The prose rests on the assert that the draws match, the two chances and the sample sizes."""3import math45from lib import pii6from lib.catalog import open_catalog7from lib.warehouse import connect89con = connect(open_catalog(["crm.learners"]))10first = pii.sample(con, "crm.learners", "notes", rows=20)11second = pii.sample(con, "crm.learners", "notes", rows=20) # the same call: the seed and one thread (set in lib.pii.sample) make the draws match12assert second == first13print("second draw, same notes as the first:", "yes" if second == first else "no")14# Notes can hold a phone number, which may be in service: masked() prints it with only its last four digits.15print("first sampled notes:", " | ".join(pii.masked(note) for note in first[:3]))1617# How rare is personal data in the column? Run the detector over every note, once, to know.18detector = pii.platform_detector()19[(rows, notes)] = con.sql("select count(*), count(notes) from crm.learners").fetchall() # count(notes) skips NULLs20every_note = [note for (note,) in con.sql("select notes from crm.learners where notes is not null").fetchall()]21holding = sum(bool(pii.detect(detector, note)) for note in every_note)22print("learners:", f"{rows:,}")23print("learners with notes:", notes)24print("notes holding personal data the detector finds:", holding)2526# The chance that n sampled values hold at least one of them is 1 - (1 - p)^n, for a share p of the values.27for unit, p in (("rows", holding / rows), ("notes", holding / notes)): # p counted over all learners, then over the notes28 print(f"chance 20 sampled {unit} catch one:", f"{1 - (1 - p) ** 20:.1%}")29 print(f"sampled {unit} needed for a 99% chance:", math.ceil(math.log(0.01) / math.log(1 - p))) # solves for n30print("values the platform samples per text column:", pii.ROWS)31# The rarest share p that a sample of ROWS values still catches 99% of the time, printed as 1 in a whole number.32print("rarest share that sample catches 99% of the time:", f"1 in {math.floor(1 / (1 - 0.01 ** (1 / pii.ROWS)))}")
python seeded-column-sample.pysecond draw, same notes as the first: yes
first sampled notes: VIP: bought a team licence. | Asked about invoices for their employer. | Called about a refund, call back on (•••)•••-1366x•••.
learners: 5,000
learners with notes: 527
notes holding personal data the detector finds: 138
chance 20 sampled rows catch one: 42.9%
sampled rows needed for a 99% chance: 165
chance 20 sampled notes catch one: 99.8%
sampled notes needed for a 99% chance: 16
values the platform samples per text column: 200
rarest share that sample catches 99% of the time: 1 in 43
两次抽取的结果相同,因此样本可以审计。种子相同,第二次抽取返回的仍是同样的 20 条备注,其中第三条备注已经含有一个电话号码,输出中对它做了掩码。
在 5,000 名学习者中,有 527 人有备注,而这些备注中有 138 条含有个人数据,能被平台的检测器发现,该检测器将在本章后面调优。NULL 会稀释样本,因为 NULL 不可能含有个人数据。抽取 20 行时,捕获个人数据的概率为 42.9%,要达到 99% 的概率,需要 165 行。若只对实际存在的备注计数,抽取 20 条捕获的概率为 99.8%,有 16 条就足以达到 99%。
**工作准则:**对非空值抽样,使用固定的种子,在单线程下运行,并将种子与结果一并保存。样本量应按必须捕获的最稀少的个人数据来确定,达到 99% 时取 n = ln(0.01) / ln(1 − p)。平台对每个文本列抽取 200 个值,该数值设在 governance/pii.toml 中,对于有 1 in 43 或更多的值含有个人数据的情形,捕获概率为 99%。对于更稀少的数据,应扫描整列,或者按规则对其分类。
根据取值对列分类
名称规则和可复现的抽样合在一起,为每一列给出一个分类建议。lib.pii.classify取二者所支持的最严格分类:名称所暗示的实体,以及在任一抽样取值中发现的每一个实体。没有任何证据的列得到internal,因为检测器什么也没找到,就等于什么也没有证明,而只有人才能判定某一列是public。
采用最严格的分类是一种选择。Presidio 自己用于表格的工具 Presidio structured 只为每一列给出一个实体,其默认方式 Most Common 选取出现次数最多的实体(Presidio 文档)。这适用于决定如何对一列做掩码。但用于分类时,一列在 200 个取值中只要有一个卡号,它就仍然含有卡号。
下面用平台的检测器为目录中每一列生成建议,所用的识别器由后面的一节选定,并显示各列的建议与已审核元数据在哪些地方不同:
1"""Proposes a classification for every column, the strictest of what its name and its sampled values support, and compares2each proposal with the reviewed metadata. The prose's table rests on the proposed and found lines of each column, and3its counts on the totals printed after them."""4from lib import metadata, pii5from lib.catalog import open_catalog6from lib.warehouse import connect78catalog = open_catalog()9findings = pii.classify(catalog, connect(catalog), pii.platform_detector()) # the platform's tuned detector, not the defaults10print("columns classified:", len(findings))11print("columns sampled:", sum(f.sampled > 0 for f in findings))1213holding = [f for f in findings if f.found] # columns where sampled values hold personal data14print("columns whose sampled values hold personal data:", len(holding))15# A column counts when its values hold an entity other than the one its name suggests (named is None when no rule matches).16print("of them, holding something their name doesn't suggest:", sum(set(f.found) != {f.named} for f in holding))17for f in holding:18 print(f"{f.table}.{f.column}, proposed: {f.classification}")19 print(f"{f.table}.{f.column}, found: " + ", ".join(f"{e} in {n} of {f.sampled}" for e, n in f.found.most_common()))20[cards] = [f for f in findings if f.column == "card_number"] # the one card_number column in the catalog21# The card recognizer's count: values that match its pattern and pass its Luhn check, out of the values sampled.22print("sampled card numbers the card recognizer accepts:", f"{cards.found['CREDIT_CARD']} of {cards.sampled}")2324# Classifications the values raise above what the name alone gives: the floor is used for a name no rule matches.25named_only = [f for f in findings if pii.CLASSIFICATION.get(f.named, pii.FLOOR) != f.classification]26print("columns the values classify more strictly than the name:", ", ".join(f"{f.table}.{f.column}" for f in named_only))2728# Compare with the metadata people reviewed. A column the detector found nothing in proposes only the floor, internal.29described = metadata.load()30evidenced = [f for f in findings if f.named or f.found]31print("columns with evidence:", len(evidenced))32print("of them, equal to the metadata:",33 sum(f.classification == described[f.table].columns[f.field_id].classification for f in evidenced))34for f in evidenced:35 reviewed = described[f.table].columns[f.field_id].classification36 if metadata.stricter(f.classification, reviewed) != reviewed: # the stricter of the two: a lower proposal prints nothing37 print(f"stricter than the metadata, {f.table}.{f.column}: {f.classification}, reviewed as {reviewed}")
python classify-sampled-columns.pycolumns classified: 64
columns sampled: 29
columns whose sampled values hold personal data: 9
of them, holding something their name doesn't suggest: 5
billing.payments.card_number, proposed: sensitive
billing.payments.card_number, found: CREDIT_CARD in 151 of 200, PHONE_NUMBER in 3 of 200
crm.instructors.email, proposed: personal
crm.instructors.email, found: EMAIL_ADDRESS in 50 of 50
crm.instructors.payout_iban, proposed: sensitive
crm.instructors.payout_iban, found: IBAN_CODE in 50 of 50
crm.learners.email, proposed: personal
crm.learners.email, found: EMAIL_ADDRESS in 200 of 200
crm.learners.phone, proposed: personal
crm.learners.phone, found: PHONE_NUMBER in 175 of 200
crm.learners.notes, proposed: personal
crm.learners.notes, found: PHONE_NUMBER in 29 of 200, EMAIL_ADDRESS in 18 of 200
learning.events.ip_address, proposed: personal
learning.events.ip_address, found: IP_ADDRESS in 200 of 200, PHONE_NUMBER in 39 of 200
learning.reviews.body, proposed: personal
learning.reviews.body, found: EMAIL_ADDRESS in 4 of 200
support.tickets.body, proposed: sensitive
support.tickets.body, found: EMAIL_ADDRESS in 47 of 200, PHONE_NUMBER in 37 of 200, DATE_TIME in 29 of 200, IBAN_CODE in 21 of 200, CREDIT_CARD in 12 of 200
sampled card numbers the card recognizer accepts: 151 of 200
columns the values classify more strictly than the name: crm.learners.notes, learning.reviews.body, support.tickets.body
columns with evidence: 14
of them, equal to the metadata: 13
stricter than the metadata, support.tickets.body: sensitive, reviewed as personal
名称没有说明的部分,由取值来确定。在样本中含有个人数据的 9 列里,有一些只含有其名称所说的内容,例如crm.learners.email。其余的 5 列含有名称并未暗示的内容,或者名称本身什么也没有暗示:
| 列 | 建议 | 在样本中发现 |
|---|---|---|
billing.payments.card_number |
sensitive | CREDIT_CARD in 151 of 200, PHONE_NUMBER in 3 of 200 |
crm.learners.notes |
personal | PHONE_NUMBER in 29 of 200, EMAIL_ADDRESS in 18 of 200 |
support.tickets.body |
sensitive | EMAIL_ADDRESS in 47 of 200, PHONE_NUMBER in 37 of 200, DATE_TIME in 29 of 200, IBAN_CODE in 21 of 200, CREDIT_CARD in 12 of 200 |
learning.reviews.body |
personal | EMAIL_ADDRESS in 4 of 200 |
learning.events.ip_address |
personal | IP_ADDRESS in 200 of 200, PHONE_NUMBER in 39 of 200 |
旧版卡号列的描述写着“Should not be stored at all”(根本不应存储),而样本表明它含有完整的卡号,该表的所有者team-billing可以据此采取行动。在抽样得到的卡号中,有151 of 200符合 Presidio 的卡号模式,并通过了它所检查的 Luhn 校验和(源码),其余的是该模式未覆盖的格式(参见模式在哪些地方失效)。仅凭取值,备注列就是personal。工单文本是sensitive,因为学习者会把卡号和 IBAN 粘贴到工单中,所以这是唯一一个比元数据更严格的建议,而人工审核时把它定为personal。在有证据的 14 列中,有 13 列与已审核的分类一致。
要阅读证据,而不只是结论。有些 IP 地址也会被识别为电话号码。这在此处没有影响,因为 IP 地址无论如何都是personal,但检测器的标签说明的是一个值看起来像什么,而不是它是什么。
工作准则: 提出名称或任一抽样取值所支持的最严格分类,以internal为下限,并随建议一起给出证据。比已审核元数据更严格的建议,对表的所有者而言是一项发现,由其提高分类,或记录不提高的原因;决不能让检测器降低分类。OpenMetadata 会跳过任何已带有 PII 标签的列;我们建议改为每次运行都对每个文本列抽样,因为已审核的标签正是需要被检查的对象。
这部分尚未翻译,因此以英文显示。
Measure a detector against labelled spans
A column-level answer is enough to classify support.tickets.body, but not to mask it: masking replaces each span of personal data in each text, and a span the detector misses stays readable. To know what a detector misses, compare its spans with the truth, the labelled spans of every ticket and review. Two matching rules are standard, and lib.pii applies both.
Exact matching, from the CoNLL-2003 shared task, counts an entity "correct only if it is an exact match of the corresponding entity in the data file" (Tjong Kim Sang and De Meulder, 2003). Overlap matching asks only for one shared character, with a label of its type, which the SemEval-2013 task on drug interactions calls "Type matching" (Segura-Bedmar et al., 2013). Neither credits a span hidden under another label, so lib.pii adds a third measure, coverage: whether every character of a labelled span is inside some detection, whatever its label. Exact asks whether the detector labelled the whole span right, overlap whether it noticed the span as what it is, and coverage whether masking its detections hides the span.
lib.pii.score computes exact and overlap matching as recall, the share of labelled spans of each type that a detector finds with a label of that type, and precision, the share of each label's detections that match a labelled span. lib.pii.coverage gives the share of each type's spans covered. score also counts the texts in which every labelled character falls inside some detection, the texts that masking would leave with nothing personal. The Text Anonymization Benchmark protects a person only when all of their mentions are masked (Pilán et al., 2022), and the platform counts whole texts, which is stricter. Recall comes first: Presidio's evaluation guide says it often matters more than precision, because missing personal data is the error to avoid (Presidio docs).
Presidio's defaults, scored against the labelled spans of every ticket and review, give the baseline the rest of the chapter improves on:
1"""Runs Presidio with its defaults over every ticket and review, and scores its detections against the truth's labelled2spans. The prose rests on the recall and precision lines, the count of texts left with nothing personal, and the two3date lines that test the truth's limits."""4import re56from lib import pii78labelled = pii.truth()9print("texts:", f"{sum(len(pii.texts(source)) for source in pii.SOURCES):,}")10print("labelled spans:", f"{len(labelled):,}")1112detector = pii.detector() # Presidio's defaults: every recognizer, threshold 013found = pii.find(detector)14# scoring needs a label of the span's type; coverage doesn't15scores, covered = pii.score(found, labelled), pii.coverage(found, labelled)16print("detections:", f"{len(found):,}")1718for entity, (n, exact, overlap) in scores.recall.items():19 print(f"{entity}, labelled: {n:,}")20 print(f"{entity}, recall, exact: {exact / n:.2f}")21 print(f"{entity}, recall, overlap: {overlap / n:.2f}")22 print(f"{entity}, covered, any label: {covered[entity][1] / n:.2f}") # covered[entity] is (spans, covered), so n is the same count2324for label, (n, exact, overlap) in scores.precision.items(): # precision is per label: what each detection was called25 print(f"{label}, detections: {n:,}")26 print(f"{label}, precision, exact: {exact / n:.2f}")2728# every bank number and driver's licence detection is a false alarm: none matches a labelled span29assert all(scores.precision[entity][1] == 0 for entity in ("US_BANK_NUMBER", "US_DRIVER_LICENSE"))30print("precision of all detections, exact:", f"{pii.precision_overall(scores):.2f}")31print("texts holding personal data:", f"{scores.texts:,}")32print("texts with every labelled span masked:", f"{scores.masked:,}")3334# The truth's limits. The generator builds every text from a few sentence templates, and writes every date of birth35# one way. Presidio's date patterns don't read every way of writing one.36print("sentence templates the texts are built from:", len(pii.templates()))37births = [pii.text_of(s) for s in labelled if s.entity == "DATE_TIME"] # the truth's dates are the dates of birth: dates of a problem are unlabelled38written_ymd = sum(re.fullmatch(r"\d{4}-\d\d-\d\d", d) is not None for d in births) # year first, the generator's one form39assert written_ymd == len(births) # so every date of birth is written this way40print("dates of birth written like 1985-03-12:", written_ymd)41for written in ("1985-03-12", "12 March 1985"):42 text = f"My date of birth on the account is wrong, it should be {written}." # a sentence in the generator's shape43 print(f"dates found in a date of birth written {written}:", sum(e == "DATE_TIME" for _, _, e, _ in pii.detect(detector, text)))
python measure-pattern-recognizers.pytexts: 5,000
labelled spans: 4,292
detections: 3,902
CREDIT_CARD, labelled: 213
CREDIT_CARD, recall, exact: 0.71
CREDIT_CARD, recall, overlap: 0.71
CREDIT_CARD, covered, any label: 0.89
DATE_TIME, labelled: 196
DATE_TIME, recall, exact: 1.00
DATE_TIME, recall, overlap: 1.00
DATE_TIME, covered, any label: 1.00
EMAIL_ADDRESS, labelled: 543
EMAIL_ADDRESS, recall, exact: 1.00
EMAIL_ADDRESS, recall, overlap: 1.00
EMAIL_ADDRESS, covered, any label: 1.00
IBAN_CODE, labelled: 236
IBAN_CODE, recall, exact: 1.00
IBAN_CODE, recall, overlap: 1.00
IBAN_CODE, covered, any label: 1.00
LOCATION, labelled: 210
LOCATION, recall, exact: 0.00
LOCATION, recall, overlap: 0.00
LOCATION, covered, any label: 0.00
PERSON, labelled: 2,455
PERSON, recall, exact: 0.00
PERSON, recall, overlap: 0.00
PERSON, covered, any label: 0.00
PHONE_NUMBER, labelled: 439
PHONE_NUMBER, recall, exact: 0.78
PHONE_NUMBER, recall, overlap: 0.79
PHONE_NUMBER, covered, any label: 0.91
CREDIT_CARD, detections: 151
CREDIT_CARD, precision, exact: 1.00
DATE_TIME, detections: 421
DATE_TIME, precision, exact: 0.47
EMAIL_ADDRESS, detections: 543
EMAIL_ADDRESS, precision, exact: 1.00
IBAN_CODE, detections: 236
IBAN_CODE, precision, exact: 1.00
MEDICAL_LICENSE, detections: 18
MEDICAL_LICENSE, precision, exact: 0.00
PHONE_NUMBER, detections: 384
PHONE_NUMBER, precision, exact: 0.89
UK_NHS, detections: 8
UK_NHS, precision, exact: 0.00
URL, detections: 751
URL, precision, exact: 0.00
US_BANK_NUMBER, detections: 689
US_BANK_NUMBER, precision, exact: 0.00
US_DRIVER_LICENSE, detections: 683
US_DRIVER_LICENSE, precision, exact: 0.00
US_ITIN, detections: 2
US_ITIN, precision, exact: 0.00
US_PASSPORT, detections: 8
US_PASSPORT, precision, exact: 0.00
US_SSN, detections: 8
US_SSN, precision, exact: 0.00
precision of all detections, exact: 0.38
texts holding personal data: 2,586
texts with every labelled span masked: 425
sentence templates the texts are built from: 30
dates of birth written like 1985-03-12: 196
dates found in a date of birth written 1985-03-12: 1
dates found in a date of birth written 12 March 1985: 0
| Entity type | Labelled | Recall, exact | Recall, overlap | Covered |
|---|---|---|---|---|
PERSON |
2,455 | 0.00 | 0.00 | 0.00 |
EMAIL_ADDRESS |
543 | 1.00 | 1.00 | 1.00 |
PHONE_NUMBER |
439 | 0.78 | 0.79 | 0.91 |
IBAN_CODE |
236 | 1.00 | 1.00 | 1.00 |
CREDIT_CARD |
213 | 0.71 | 0.71 | 0.89 |
LOCATION |
210 | 0.00 | 0.00 | 0.00 |
DATE_TIME |
196 | 1.00 | 1.00 | 1.00 |
The defaults find every email, IBAN and date of birth, most phone and card numbers, and no names or street addresses at all, which is most of what the tickets hold. Most of what they flag is wrong: 0.38 of all detections match a labelled span. The 689 US bank numbers, 683 driver's licences and 751 URLs are all wrong. Phone, card and invoice numbers, all runs of digits, come out as bank numbers and licences, and pieces of emails and a few addresses come out as URLs. A wrong label still hides the digits, so phone and card numbers are covered more often than they're found.
Masking would leave personal data in most texts that hold it. Only 425 of the 2,586 texts would come out with nothing personal left.
The truth has its own limits, because the text is generated. Every ticket and review is built from 30 sentence templates in prep/generate.py, so each hard case is one sentence frame, and all 196 dates of birth are written like 1985-03-12. The date recall of 1.00 therefore measures one format: in the generator's own sentence, Presidio finds 1 date when the date of birth is written 1985-03-12, and 0 when it's written 12 March 1985. A measure on this text tells you how a detector behaves on these shapes, not how it will do on yours.
Working rule: Measure every detector against labelled spans before it masks anything: recall per entity type by exact and by overlap matching, both with a label of the right type, coverage whatever the label, precision per label, and the share of texts with nothing personal left. Weigh recall first. Before you trust a detector on your text, label a few hundred real records by hand and measure on those too.
依据真值调优识别器
Presidio 的默认设置会返回每一个弱匹配。FAQ 给出了两种补救办法:调整接受阈值,以及移除数据集不需要的识别器;它同时指出,误报与漏报之间存在权衡(FAQ)。Presidio 2.2.364 增加了按识别器和按实体设置的分数阈值,并以分析器的默认值作为回退(变更日志,发布说明)。对于学习者分布在多个国家的平台,电话识别器还有一项同样重要的设置:它读取其编号方案的地区(源码)。
governance/pii.toml 保留了 6 个识别器,对应平台文本可能包含的各类标识符,每个识别器各有自己的阈值。电话地区不是一项设置:lib.pii.countries 从 crm.learners 中读取学习者所在的国家,因此地区随数据而定。四种设置依据同一份真值打分:
1"""Scores four detector settings against the same truth: Presidio's defaults, one threshold for every recognizer, the kept2recognizers with their own thresholds, and those with phone numbers read in the learners' countries. The prose rests on3the precision, phone recall, coverage and fully masked texts of each setting."""4from lib import pii56labelled = pii.truth()7print("recognizers kept:", len(pii.THRESHOLDS)) # read from governance/pii.toml8print("recognizers kept, with their thresholds:", ", ".join(f"{name} {t}" for name, t in pii.THRESHOLDS.items()))9print("phone regions, Presidio's default:", ", ".join(pii.PhoneRecognizer.DEFAULT_SUPPORTED_REGIONS)) # Presidio's own list10print("phone regions, the learners' countries:", ", ".join(pii.countries())) # read from crm.learners1112# In the order the prose builds them: the defaults, one threshold, the kept recognizers, then their regions.13settings = {14 "Presidio's defaults": pii.detector(),15 "one threshold of 0.5": pii.detector(default_threshold=0.5),16 "kept recognizers, own thresholds": pii.detector(pii.THRESHOLDS),17 "kept recognizers, learners' phone regions": pii.platform_detector(),18}19for name, detector in settings.items():20 found = pii.find(detector)21 scores, covered = pii.score(found, labelled), pii.coverage(found, labelled)22 n, exact, _ = scores.recall["PHONE_NUMBER"] # the table reads exact recall, so the overlap count is left out23 print(f"{name}, detections: {len(found):,}")24 print(f"{name}, precision, exact: {pii.precision_overall(scores):.2f}")25 print(f"{name}, phone recall, exact: {exact / n:.2f}")26 # coverage takes a detection under any label27 for entity, short in (("PHONE_NUMBER", "phones"), ("CREDIT_CARD", "cards")):28 print(f"{name}, {short} covered, any label: {covered[entity][1] / covered[entity][0]:.2f}")29 print(f"{name}, texts with every labelled span masked: {scores.masked:,}")
python tune-pattern-recognizers.pyrecognizers kept: 6
recognizers kept, with their thresholds: EmailRecognizer 0.5, PhoneRecognizer 0.4, CreditCardRecognizer 0.5, IbanRecognizer 0.5, DateRecognizer 0.5, IpRecognizer 0.5
phone regions, Presidio's default: US, GB, DE, FR, IL, IN, CA, BR
phone regions, the learners' countries: AU, BR, CA, DE, ES, FR, GB, IN, JP, MX, PL, US
Presidio's defaults, detections: 3,902
Presidio's defaults, precision, exact: 0.38
Presidio's defaults, phone recall, exact: 0.78
Presidio's defaults, phones covered, any label: 0.91
Presidio's defaults, cards covered, any label: 0.89
Presidio's defaults, texts with every labelled span masked: 425
one threshold of 0.5, detections: 1,903
one threshold of 0.5, precision, exact: 0.59
one threshold of 0.5, phone recall, exact: 0.00
one threshold of 0.5, phones covered, any label: 0.01
one threshold of 0.5, cards covered, any label: 0.71
one threshold of 0.5, texts with every labelled span masked: 284
kept recognizers, own thresholds, detections: 1,510
kept recognizers, own thresholds, precision, exact: 0.97
kept recognizers, own thresholds, phone recall, exact: 0.78
kept recognizers, own thresholds, phones covered, any label: 0.78
kept recognizers, own thresholds, cards covered, any label: 0.72
kept recognizers, own thresholds, texts with every labelled span masked: 394
kept recognizers, learners' phone regions, detections: 1,539
kept recognizers, learners' phone regions, precision, exact: 0.97
kept recognizers, learners' phone regions, phone recall, exact: 0.84
kept recognizers, learners' phone regions, phones covered, any label: 0.84
kept recognizers, learners' phone regions, cards covered, any label: 0.72
kept recognizers, learners' phone regions, texts with every labelled span masked: 405
| 指标 | 默认设置 | 单一阈值 0.5 | 保留的识别器 | 再加学习者所在地区 |
|---|---|---|---|---|
| 检出数 | 3,902 | 1,903 | 1,510 | 1,539 |
| 精确率(完全匹配) | 0.38 | 0.59 | 0.97 | 0.97 |
| 电话召回率(完全匹配) | 0.78 | 0.00 | 0.78 | 0.84 |
| 已覆盖的电话号码 | 0.91 | 0.01 | 0.78 | 0.84 |
| 已覆盖的卡号 | 0.89 | 0.71 | 0.72 | 0.72 |
| 完全脱敏的文本 | 425 | 284 | 394 | 405 |
最直观的办法行不通。单一的 0.5 阈值去掉了弱匹配,也去掉了所有电话号码,因为在没有上下文词可以提高分数时,电话号码的得分为 0.40。只保留 6 个识别器,让每个识别器使用各自的阈值,并把电话阈值设为电话号码实际得到的分数,精确率就提高到 0.97,电话召回率则保持不变。按学习者所在地区而不是 Presidio 的默认地区来读取电话号码,电话召回率提高到 0.84。
有一个数字朝着不利的方向变化:完全脱敏的文本比使用默认设置时更少。被去掉的识别器曾把一些电话号码和银行卡号以错误的标签脱敏,而这样仍然遮住了数字。去掉它们之后,电话号码的覆盖率从 0.91 降到 0.78,银行卡号的覆盖率从 0.89 降到 0.72。我们认为精确率值得这个代价,因为检出结果是审阅者要读的证据;而在默认设置下,工单和评价中含有 689 个银行账号和 683 个驾驶证号,其中没有一个是真实的。governance/pii.toml 没有把这两种实体映射到任何分类,所以它们不会改变任何提案,但每一个都是需要有人阅读并驳回的发现。去掉美国识别器也是对平台文本的一项断言:真值中不含美国银行账号或驾驶证号。如果平台日后需要其中任何一种,这项断言和测量都必须随之改变。
工作准则: 切勿使用 Presidio 默认的 0 阈值,也不要让所有识别器共用一个阈值。只保留文本可能包含的标识符所对应的识别器,依据每个识别器自身的分数设定其阈值,并按用户所在的地区读取电话号码,地区由你的数据得出。把这一切都写进代码,并在每次变更后重新测量。
这部分尚未翻译,因此以英文显示。
See where patterns fail
The tuned recognizers score well on totals, and totals hide where they fail. A pattern detector fails in three ways: there's no pattern for the thing at all, its pattern or its validation rejects the value, or it matches something that only looks personal. The platform's tickets plant cases of each kind, as its data card says: names typed in lowercase, initials, first names that are ordinary words, phone numbers typed without separators, and invoice numbers, order numbers and dates of a problem, which the truth leaves unlabelled. Count each case with the platform's detector. A case counts as caught when some detection overlaps it, whatever its label:
1"""Counts each hard case the tickets and reviews plant, and each look-alike number or date the platform's detector flags,2then checks the card and phone numbers it misses. The prose rests on the caught counts for names and invoice numbers,3and on the misses' lengths, first digits and validity."""4from collections import Counter56from lib import pii7from lib.platform import load89labelled = pii.truth()10found = pii.find(pii.platform_detector())1112cases = pii.hard_cases(labelled) # a case is caught when some detection overlaps it, whatever its label13for case, spans in cases.items():14 print(f"{case}, labelled: {len(spans)}")15 print(f"{case}, caught: {pii.caught(spans, found)}")16print("hard cases of a name, caught:", pii.caught([s for spans in cases.values() for s in spans if s.entity == "PERSON"], found))1718# Look-alikes are numbers and dates that aren't personal data, so any detection on one is a false alarm.19for kind, spans in pii.lookalikes().items():20 print(f"{kind}, in the tickets: {len(spans)}")21 print(f"{kind}, flagged as personal data: {pii.caught(spans, found)}")2223# Card numbers the card pattern misses: too short or long for its 13 to 17 digits, or a first digit it doesn't accept.24# Out of range, the length is the reason; in range, the first digit is, since the pattern accepts only some of them.25missed = pii.missed([s for s in labelled if s.entity == "CREDIT_CARD"], found, {"CREDIT_CARD"})26why = lambda digits: f"{len(digits)} digits" if not 13 <= len(digits) <= 17 else f"{len(digits)} digits starting with {digits[0]}"27print("CREDIT_CARD, missed:", len(missed))28for reason, n in Counter(why(pii.text_of(s)) for s in missed).most_common():29 print(f"CREDIT_CARD, missed, {reason}: {n}")30print("IBAN_CODE, missed:", len(pii.missed([s for s in labelled if s.entity == "IBAN_CODE"], found, {"IBAN_CODE"})))3132# Are the phone numbers the patterns miss valid where the ticket's learner lives? Phones appear only in tickets,33# so each phone's learner comes from its ticket.34learner_of = {row["ticket_id"]: row["learner_id"] for row in load("support.tickets").to_pylist()}35country_of = {row["learner_id"]: row["country"] for row in load("crm.learners").to_pylist()}36phones = pii.missed([s for s in labelled if s.entity == "PHONE_NUMBER"], found, {"PHONE_NUMBER"})37print("phone numbers missed:", len(phones))38print("of them, invalid in the learner's country:",39 sum(not pii.valid_phone(pii.text_of(s), country_of[learner_of[s.row_id]]) for s in phones))
python pattern-hard-cases.pynames in lowercase, labelled: 211
names in lowercase, caught: 0
initial and surname, labelled: 163
initial and surname, caught: 0
first names that are words, labelled: 187
first names that are words, caught: 0
instructors named in reviews, labelled: 798
instructors named in reviews, caught: 0
phones typed as digits only, labelled: 257
phones typed as digits only, caught: 220
hard cases of a name, caught: 0
invoice numbers, in the tickets: 224
invoice numbers, flagged as personal data: 31
order numbers, in the tickets: 212
order numbers, flagged as personal data: 0
dates of a problem, in the tickets: 225
dates of a problem, flagged as personal data: 0
CREDIT_CARD, missed: 62
CREDIT_CARD, missed, 19 digits: 23
CREDIT_CARD, missed, 12 digits: 22
CREDIT_CARD, missed, 16 digits starting with 2: 10
CREDIT_CARD, missed, 15 digits starting with 2: 7
IBAN_CODE, missed: 0
phone numbers missed: 65
of them, invalid in the learner's country: 63
Names are the gap, since no pattern matches a name. Across the names in lowercase, the initials with a surname, the first names that are ordinary words and the instructors named in reviews, the patterns catch 0 of them. Phones typed as digits only do well, 220 of 257. The price is the look-alikes: 31 of the 224 invoice numbers are flagged as personal data.
The platform's detector flags 0 of the 212 order numbers and 0 of the 225 dates of a problem. With Presidio's defaults, 12/03 was a date at 0.10, and the date recognizer returned 421 dates where the truth labels 196. The threshold in governance/pii.toml keeps such weak matches out.
The card pattern's shape causes the card misses. It accepts 13 to 17 digits starting with 1, 3, 4, 50 to 55 or 6, but no 13-digit number starting with 1 (source), and every card number in the data passes the Luhn checksum, so validation rejects none. Of the 62 card numbers it misses, 23 have 19 digits and 22 have 12, and the rest start with 2, which the pattern doesn't accept.
Validation causes the other misses. The IBAN recognizer checks each country's format, length and check digits (source). Every labelled IBAN passes, so it misses 0, but a learner who mistypes one digit writes a value it rejects. The phone recognizer accepts only numbers that python-phonenumbers calls valid (source), and the generator doesn't always write those. Of the 65 phone numbers the detector misses, 63 aren't valid in the country the ticket's learner lives in, so most of the phone misses are the generator's, not the detector's.
Working rule: Read a detector's misses by case, not by total, and keep a list of the cases your text holds. Use patterns for what has a checkable shape, such as emails, IBANs, card numbers, phone numbers and dates, and never for names or addresses. Keep the checksums and numbering plans, because they're what makes the patterns precise. Check the card pattern against the card numbers your payment provider issues, and the date patterns against the ways your users write dates.
这部分尚未翻译,因此以英文显示。
Compare a PII model
The patterns have nothing for names or addresses, and those are most of the personal data in the platform's text: 2,455 names and 210 addresses among the 4,292 labelled spans. Finding them takes a model that reads context, not a general named-entity model. The authors of the SPY dataset define PII as the entities that can identify, contact or locate a specific individual, and found that dedicated named-entity models have limits in PII contexts (Savkin et al., 2025). We use a PII model, one trained for personally identifiable information: OpenAI's Privacy Filter, an Apache-2.0 model with eight fixed labels, such as private_person, private_address and account_number (model card).
The labels are the model's own. OpenAI's card says its label policies can't be set while the model runs (OpenAI, 2026, section 4.2). We chose it over GLiNER2-PII, the runner-up among open PII models, for those fixed labels. GLiNER2-PII takes the labels to look for at inference time, along with a threshold (model card), so a recording of it would have to pin both. Privacy Filter has 1.5 billion parameters, 50 million of them active, and runs on a CPU.
The book's examples never run a model, so they read a recording of its output. The recording, data/recorded/pii-entities/, which this chapter and chapter 9 use, holds every span the model found in every ticket and review, with its offsets, its text and the model's label. recordings/pii_entities.py made it on a CPU with the model's reference decoder, opf (code), which chooses the best sequence of labels for a whole text, so most spans come back in one piece rather than token by token. It used the default operating point, with all six transition biases at 0 (calibration file); the model can also be set "for improved recall" (section 2.3.2). The manifest pins the model's revision and the SHA-256 of its weights, so anyone can record it again and compare, and lib.pii.recorded refuses spans that don't match the text at their offsets. The recording keeps the model's labels, and the example maps them to the truth's types, so a different mapping never needs a new recording. Score the model, the platform's patterns and both together, masking whatever either finds:
1"""Scores the recorded spans of OpenAI's Privacy Filter against the truth, alone and beside the platform's patterns. The2prose rests on the recall lines for names, addresses, phones and cards, the precision lines, and the texts left with3nothing personal."""4from lib import pii56# Privacy Filter's eight labels, mapped to the truth's entity types. A label the truth has no type for matches nothing.7MATCHES = {"private_person": {"PERSON"}, "private_email": {"EMAIL_ADDRESS"}, "private_phone": {"PHONE_NUMBER"},8 "private_address": {"LOCATION"}, "account_number": {"CREDIT_CARD", "IBAN_CODE"},9 "private_date": {"DATE_TIME"}, "private_url": set(), "secret": set()}1011model, manifest = pii.recorded("pii-entities") # the model's spans, from a recording: no model runs here12print("model:", manifest["model"], "decoded with", manifest["settings"]["decode_mode"])13labelled = pii.truth()14detectors = {"patterns": pii.find(pii.platform_detector()), "model": model}15# both lists joined: a span either one finds is masked, and a span both find is simply counted twice16detectors["both"] = detectors["patterns"] + model1718for name, found in detectors.items():19 scores = pii.score(found, labelled, MATCHES) # Presidio's labels aren't in MATCHES, so match themselves20 print(f"{name}, detections: {len(found):,}")21 for entity, (n, exact, overlap) in scores.recall.items():22 print(f"{name}, {entity}, recall, exact: {exact / n:.2f}")23 print(f"{name}, {entity}, recall, overlap: {overlap / n:.2f}")24 print(f"{name}, precision, exact: {pii.precision_overall(scores):.2f}")25 print(f"{name}, texts with every labelled span masked: {scores.masked:,} of {scores.texts:,}")2627# Precision by the model's own label: a label that's right for one truth type can still be wrong for another28for label, (n, exact, overlap) in pii.score(model, labelled, MATCHES).precision.items():29 print(f"model, {label}, detections: {n:,}")30 print(f"model, {label}, precision, exact: {exact / n:.2f}")31 print(f"model, {label}, precision, overlap: {overlap / n:.2f}")
python privacy-filter-spans.pymodel: openai/privacy-filter decoded with viterbi
patterns, detections: 1,539
patterns, CREDIT_CARD, recall, exact: 0.71
patterns, CREDIT_CARD, recall, overlap: 0.71
patterns, DATE_TIME, recall, exact: 1.00
patterns, DATE_TIME, recall, overlap: 1.00
patterns, EMAIL_ADDRESS, recall, exact: 1.00
patterns, EMAIL_ADDRESS, recall, overlap: 1.00
patterns, IBAN_CODE, recall, exact: 1.00
patterns, IBAN_CODE, recall, overlap: 1.00
patterns, LOCATION, recall, exact: 0.00
patterns, LOCATION, recall, overlap: 0.00
patterns, PERSON, recall, exact: 0.00
patterns, PERSON, recall, overlap: 0.00
patterns, PHONE_NUMBER, recall, exact: 0.84
patterns, PHONE_NUMBER, recall, overlap: 0.85
patterns, precision, exact: 0.97
patterns, texts with every labelled span masked: 405 of 2,586
model, detections: 4,424
model, CREDIT_CARD, recall, exact: 0.69
model, CREDIT_CARD, recall, overlap: 0.69
model, DATE_TIME, recall, exact: 0.97
model, DATE_TIME, recall, overlap: 0.97
model, EMAIL_ADDRESS, recall, exact: 1.00
model, EMAIL_ADDRESS, recall, overlap: 1.00
model, IBAN_CODE, recall, exact: 0.99
model, IBAN_CODE, recall, overlap: 1.00
model, LOCATION, recall, exact: 0.67
model, LOCATION, recall, overlap: 0.79
model, PERSON, recall, exact: 0.84
model, PERSON, recall, overlap: 0.84
model, PHONE_NUMBER, recall, exact: 0.94
model, PHONE_NUMBER, recall, overlap: 0.94
model, precision, exact: 0.84
model, texts with every labelled span masked: 2,095 of 2,586
both, detections: 5,963
both, CREDIT_CARD, recall, exact: 0.92
both, CREDIT_CARD, recall, overlap: 0.92
both, DATE_TIME, recall, exact: 1.00
both, DATE_TIME, recall, overlap: 1.00
both, EMAIL_ADDRESS, recall, exact: 1.00
both, EMAIL_ADDRESS, recall, overlap: 1.00
both, IBAN_CODE, recall, exact: 1.00
both, IBAN_CODE, recall, overlap: 1.00
both, LOCATION, recall, exact: 0.67
both, LOCATION, recall, overlap: 0.79
both, PERSON, recall, exact: 0.84
both, PERSON, recall, overlap: 0.84
both, PHONE_NUMBER, recall, exact: 0.99
both, PHONE_NUMBER, recall, overlap: 0.99
both, precision, exact: 0.88
both, texts with every labelled span masked: 2,120 of 2,586
model, account_number, detections: 746
model, account_number, precision, exact: 0.51
model, account_number, precision, overlap: 0.51
model, private_address, detections: 177
model, private_address, precision, exact: 0.79
model, private_address, precision, overlap: 1.00
model, private_date, detections: 384
model, private_date, precision, exact: 0.50
model, private_date, precision, overlap: 0.50
model, private_email, detections: 543
model, private_email, precision, exact: 1.00
model, private_email, precision, overlap: 1.00
model, private_person, detections: 2,110
model, private_person, precision, exact: 0.98
model, private_person, precision, overlap: 0.98
model, private_phone, detections: 464
model, private_phone, precision, exact: 0.89
model, private_phone, precision, overlap: 0.89
| Measure | Patterns | Model | Both |
|---|---|---|---|
Recall, exact, PERSON |
0.00 | 0.84 | 0.84 |
Recall, exact, LOCATION |
0.00 | 0.67 | 0.67 |
Recall, exact, PHONE_NUMBER |
0.84 | 0.94 | 0.99 |
Recall, exact, CREDIT_CARD |
0.71 | 0.69 | 0.92 |
Recall, exact, IBAN_CODE |
1.00 | 0.99 | 1.00 |
| Precision, exact | 0.97 | 0.84 | 0.88 |
| Texts fully masked | 405 of 2,586 | 2,095 of 2,586 | 2,120 of 2,586 |
The model finds most of the names and addresses the patterns can't, and the two detectors complement each other on card numbers. The model and the patterns find about the same share of cards, 0.69 and 0.71, but not the same cards, so together they find 0.92. A later section shows why the model's share is low.
The model pays in precision, 0.84 overall, and only 0.51 for account_number and 0.50 for private_date. That partly measures a difference in label policy. The truth leaves invoice numbers and dates of a problem unlabelled, and OpenAI's card says appropriate label policies can differ (section 4.2).
Together, the two detectors leave 2,120 of 2,586 texts with nothing personal in them, against 405 of 2,586 for the patterns alone.
这部分尚未翻译,因此以英文显示。
Read the model's misses
OpenAI's card says where the model fails: it misses rare names, names from other naming traditions, initials, names wrapped in titles and domain-specific identifiers, and it over-masks public entities, locations and common nouns when the local context is unclear (OpenAI, 2026, section 4.3). Check that list against the hard cases the tickets plant:
1"""Counts how many of each hard case the model's recorded spans catch, and reads the names it misses by case and length.2The prose rests on the missed-name counts and on the first lowercase ticket that both detectors miss."""3from collections import Counter45from lib import pii67model, _ = pii.recorded("pii-entities")8labelled = pii.truth()9cases = pii.hard_cases(labelled)1011for case, spans in cases.items():12 print(f"{case}, labelled: {len(spans)}")13 print(f"{case}, caught by the model: {pii.caught(spans, model)}")14for kind, spans in pii.lookalikes().items():15 print(f"{kind}, flagged by the model: {pii.caught(spans, model)}")1617# The names the model misses, by case: read them before trusting a recall figure.18missed = pii.missed([s for s in labelled if s.entity == "PERSON"], model, {"private_person"})19print("names the model misses:", len(missed))20for case in ("first names that are words", "instructors named in reviews"):21 often = Counter(pii.text_of(s) for s in pii.missed(cases[case], model, {"private_person"})).most_common(3)22 print(f"{case}, missed most often:", ", ".join(name for name, _ in often))2324# A first name on its own says less than a full name: how many of the misses are one word?25print("missed names that are one word:", sum(len(pii.text_of(s).split()) == 1 for s in missed))26print("missed names of two words or more:", sum(len(pii.text_of(s).split()) > 1 for s in missed))27# islower() counts only names typed entirely in lowercase, so a name with a capital letter, such as "Jane", doesn't.28print("of them, typed in lowercase:", sum(len(pii.text_of(s).split()) > 1 and pii.text_of(s).islower() for s in missed))2930# A lowercase name that neither the model nor the platform's patterns cover any part of, in the first ticket with one.31# The model's spans and the patterns' together: a name is missed only when both miss it. The lowest ticket ID is taken,32# so the same ticket is printed on every run.33first = min(pii.missed(cases["names in lowercase"], model + pii.find(pii.platform_detector())), key=lambda s: s.row_id)34print("first ticket with a lowercase name both detectors miss:", first.row_id)35print("the name, as typed:", pii.text_of(first))
python privacy-filter-hard-cases.pynames in lowercase, labelled: 211
names in lowercase, caught by the model: 167
initial and surname, labelled: 163
initial and surname, caught by the model: 162
first names that are words, labelled: 187
first names that are words, caught by the model: 128
instructors named in reviews, labelled: 798
instructors named in reviews, caught by the model: 535
phones typed as digits only, labelled: 257
phones typed as digits only, caught by the model: 256
invoice numbers, flagged by the model: 224
order numbers, flagged by the model: 125
dates of a problem, flagged by the model: 188
names the model misses: 387
first names that are words, missed most often: Summer, Will, Hope
instructors named in reviews, missed most often: Marie, Lori, Augustin
missed names that are one word: 337
missed names of two words or more: 50
of them, typed in lowercase: 43
first ticket with a lowercase name both detectors miss: 19
the name, as typed: juan fonseca
The misses are first names that are also words, and an instructor's first name alone in a review. The model catches 128 of the 187 first names that are words, and 535 of the 798 instructors' names.
Initials aren't the problem: the model catches 162 of 163 initials with a surname, and lowercase names are mostly caught, 167 of 211.
Most often it misses the words Summer, Will, Hope and the instructors' names Marie, Lori, Augustin. The model also flags far more look-alikes than the tuned patterns: 224 invoice numbers, 125 order numbers and 188 dates of a problem, against 31, 0 and 0.
How much the misses matter depends on their length. Of the 387 names the model misses, 337 are a single word, a first name on its own, and 50 are longer, 43 of those typed in lowercase. The authors of the Text Anonymization Benchmark draw the same distinction: missing a direct identifier, such as a full name, is much more harmful than missing a quasi-identifier (Pilán et al., 2022). A first name alone still needs masking, since a review carries its learner's ID, and an instructor's first name in a review of their own course points at one person.
The card recommends testing the model in your own domain against local policy references before production use (section 4.5). It also lists education among the settings that need extra caution (section 4.4), and calls the model "a redaction and data minimization aid, not an anonymization, compliance, or a safety guarantee" (section 4.1).
Working rule: Mask free text with a PII model and the tuned patterns together, whatever either finds: the model for names and addresses, the patterns' checksums for card numbers and IBANs. Record the model's output with its revision, weights and operating point pinned, keep its own labels, and map them in code. Measure it on your own text, by hard case, before use and after every model change; if it misses too much, try an operating point that leans toward recall and record again. Treat masked text as reduced, never as anonymous.
这部分尚未翻译,因此以英文显示。
Count what a wrong label costs
The comparison with the patterns scored the model by exact and overlap matching, which both need a label of the span's type. Masking doesn't read the label: a card number the model labels private_phone is still hidden. Classification does, because governance/pii.toml maps each entity type to a classification: a phone number to personal, a card number to sensitive. Measure the model's coverage beside its recall, and the labels it gives the platform's card numbers, phone numbers and addresses:
1"""Measures the model's coverage of card numbers, phone numbers and addresses whatever their label, and counts the labels2that would misclassify them. The prose rests on the coverage lines, the mislabelled counts and the uncovered addresses."""3from lib import pii45# What the model's labels cost. A span is covered when every character of it is inside some detection, whatever its6# label: that's what masking needs. Exact and overlap matching also need a label of the span's type.7model, _ = pii.recorded("pii-entities")8labelled = pii.truth()9covered = pii.coverage(model, labelled)10for entity in ("CREDIT_CARD", "PHONE_NUMBER", "LOCATION"):11 n, fully = covered[entity]12 print(f"{entity}, covered by the model, any label: {fully / n:.2f}")1314spans = {entity: [s for s in labelled if s.entity == entity] for entity in ("CREDIT_CARD", "PHONE_NUMBER", "LOCATION")}15print("card numbers labelled private_phone:", pii.caught(spans["CREDIT_CARD"], model, {"private_phone"}))16print("phone numbers labelled account_number:", pii.caught(spans["PHONE_NUMBER"], model, {"account_number"}))1718# For each address, the labels of the detections that overlap it, and the characters no detection covers.19on = pii.labels_on(spans["LOCATION"], model)20print("addresses with no detection at all:", sum(not labels for labels in on))21print("addresses read, at least in part, as a person:", sum("private_person" in labels for labels in on))22# A person label and no address label: the address was read as a name, which overlap matching can't credit.23print("of them, with no address label:", sum({"private_person"} <= labels and "private_address" not in labels for labels in on))24left = [pii.uncovered(address, model) for address in spans["LOCATION"]]25print("addresses not fully covered:", sum(bool(chars) for chars in left))26print("of them, with only spaces and punctuation left:", sum(bool(chars) and not any(c.isalnum() for c in chars) for chars in left))
python privacy-filter-labels.pyCREDIT_CARD, covered by the model, any label: 0.87
PHONE_NUMBER, covered by the model, any label: 1.00
LOCATION, covered by the model, any label: 0.67
card numbers labelled private_phone: 40
phone numbers labelled account_number: 27
addresses with no detection at all: 24
addresses read, at least in part, as a person: 28
of them, with no address label: 20
addresses not fully covered: 70
of them, with only spaces and punctuation left: 21
A wrong label costs classification more than it costs masking. The model covers 0.87 of card numbers, but labels 40 of the 213 as phone numbers, so its recall for cards is only 0.69. If the platform classified columns from the model's labels, a column of those card numbers would be proposed personal, not sensitive. That's what the patterns add: their Luhn check keeps card numbers labelled as cards.
The mistake also runs the other way, toward the stricter side. The model labels 27 phone numbers account_number, which maps to card numbers and IBANs, so its phone coverage is 1.00 and its recall 0.94.
Addresses show what each measure counts. The model's exact recall for them is 0.67 and its overlap recall 0.79, and its coverage is 0.67. The model finds nothing at all in 24 addresses.
The model reads part of 28 addresses as a person, since Faker builds many street names from surnames, as the platform's data card notes. It gives 20 of those no address label at all, so overlap matching misses them too. Other addresses come back in pieces with the right label, such as a street and a town split at a comma. Overlap matching credits those, and coverage doesn't. Of the 70 addresses not fully covered, 21 have every letter and digit masked and only a comma or a space left. Coverage is the measure masking needs, and it's strict.
Working rule: Report coverage, whatever the label, beside recall with the right label: masking needs the first, classification the second. Read which labels a detector gives each type of span, and keep a recognizer with a checksum for every identifier whose label decides a classification, such as card numbers.
这部分尚未翻译,因此以英文显示。
Classify again when the schema changes
A classification describes a column at one moment, and chapter 2 showed how schemas move under it. So the question is when to classify again, not on every schema ID: in PyIceberg, editing a column's description gives the schema a new ID with no new column. What needs classifying is a field ID, the number Iceberg gives each column, that the metadata has never seen. Iceberg gives every added column a new one, and a renamed column keeps its ID and so its reviewed classification. Values can also change without a schema change, so the samples of text columns need a rerun on a schedule too. Vendors promise this as a service: Databricks says its classification scans incrementally and classifies new data with no manual setup (Databricks docs), a claim we don't test.
The support team wants a number to call learners back on for tickets opened by phone, so it adds a column called callback_number to support.tickets. The change comes in four steps: edit a description, add the column, fill it from the learners' phone numbers, then classify only the field IDs that governance/metadata.toml lacks:
1"""Edits the tickets table twice, first a column's description and then a new column, and after each change counts the field2IDs the metadata lacks. The prose rests on the schema IDs, the field counts and the proposal for the new column."""3from pyiceberg.types import StringType45from lib import metadata, pii6from lib.catalog import fill_column, open_catalog7from lib.platform import load8from lib.warehouse import connect910catalog = open_catalog(["support.tickets"])11tickets = catalog.load_table("support.tickets")12reviewed = metadata.load()["support.tickets"].columns # field ID -> the column as people reviewed it13# A field ID the reviewed metadata doesn't have: Iceberg gives each added column a new one.14unclassified = lambda: {f.field_id for f in tickets.schema().fields if f.field_id not in reviewed}15print("schema ID at the start:", tickets.schema().schema_id)1617with tickets.update_schema() as update: # a doc edit: a new schema, no new column18 update.update_column("body", doc="What the learner wrote, as they wrote it.")19print("schema ID after a doc edit:", tickets.schema().schema_id)20print("fields to classify after the doc edit:", len(unclassified()))2122# The support team starts storing the number to call back for tickets opened by phone: the learner's phone number.23with tickets.update_schema() as update:24 update.add_column("callback_number", StringType(), doc="Where to call the learner back.")25phone_of = {row["learner_id"]: row["phone"] for row in load("crm.learners").to_pylist()}26# tickets from other channels have no callback number, so their value is None27fill_column(tickets, "callback_number", lambda row: phone_of[row["learner_id"]] if row["channel"] == "phone" else None)28print("schema ID after adding a column:", tickets.schema().schema_id)29print("fields to classify after adding a column:", len(unclassified()))3031# only= keeps the proposals to the field IDs the metadata lacks.32for f in pii.classify(catalog, connect(catalog), pii.platform_detector(), only={"support.tickets": unclassified()}):33 found = ", ".join(f"{entity} in {n} of {f.sampled}" for entity, n in f.found.items())34 print(f"{f.table}.{f.column}, field ID: {f.field_id}")35 print(f"{f.table}.{f.column}, entity its name suggests: {f.named}")36 print(f"{f.table}.{f.column}, found in its sampled values: {found}")37 print(f"{f.table}.{f.column}, proposed classification: {f.classification}")
python reclassify-new-columns.pyschema ID at the start: 0
schema ID after a doc edit: 1
fields to classify after the doc edit: 0
schema ID after adding a column: 2
fields to classify after adding a column: 1
support.tickets.callback_number, field ID: 9
support.tickets.callback_number, entity its name suggests: None
support.tickets.callback_number, found in its sampled values: PHONE_NUMBER in 154 of 173
support.tickets.callback_number, proposed classification: personal
A doc edit moves the schema to a new ID and leaves 0 fields to classify, since it adds no column. Adding the column moves the schema to 2, with 1 field to classify: field 9.
No name rule matches callback_number, since it contains none of the words the phone rule looks for, which is how a name rule misses a column nobody wrote it for. Its values settle it. The sample finds PHONE_NUMBER in 154 of 173, and the proposal is personal.
The pull request that adds the column is where its classification belongs, beside the metadata entry it needs: chapter 10 fails that pull request until the column has one. A proposal from values needs a column that holds values, so this one comes from the scheduled sample.
Working rule: Classify every field ID the metadata doesn't have, in the pull request that adds it, and don't let it merge without a reviewed classification. Key the trigger on field IDs, never on schema IDs. Sample every text column again on a schedule too, because values change without a schema change.
这部分尚未翻译,因此以英文显示。
What goes wrong in production
A detector's worst failure is a miss, and its output always looks like a list of findings: what it misses never appears in it:
- Phone numbers read as bank numbers. With Presidio's defaults, every run of 8 to 17 digits is a US bank number, and so is every phone number typed without separators. The evidence for the ticket table lists 689 bank numbers that aren't there, a report of "bank numbers in tickets" sends someone to the wrong team, and mapping that entity to a classification would raise columns for the wrong reason. Keep only the recognizers your text needs, each with its own threshold, as the tuning section does, and count detections per label: a label with hundreds of hits and no labelled spans is a false positive.
- Names in free text missed. Patterns find no names, and the model misses some lowercase names, first names that are also words, and first names standing alone. Ticket 19 still says "this is juan fonseca, i still can't log in" after both detectors have run, and passes every check that counts detections. You notice only by measuring recall against labelled text, per hard case, and by reading a sample of masked output.
- Personal data in a column called
notes. No name rule marks it, and a reviewer who never read it classifies itinternal. Sample every text column's values, and treat free-text columns as personal until a sample says otherwise. - A column more sensitive than its metadata. Learners paste card numbers into tickets, so
support.tickets.bodyholdssensitivedata under apersonalclassification. Compare every proposal with the reviewed metadata, and route a stricter one to the table's owner. - A detector that reaches the network.
AnalyzerEngine()downloads a model, and tldextract fetches the Public Suffix List. Offline, the pipeline fails; online, the same text can be classified differently on two machines. A setting read at import holds only if it's made before anything imports the library. Configure the engine, pin what it would fetch, check that the setting held, and run the detector's tests with the network off. - Truth shaped by its generator. The platform's text comes from 30 sentence templates filled by Faker, which writes every date of birth one way, phone numbers that aren't valid where the learner lives, and street names built from people's names. Read the misses before trusting a recall figure, and measure again on labelled samples of your own text.
这部分尚未翻译,因此以英文显示。
What this chapter decided
- Columns are classified from their name, their type and their values. Name rules run on every column; text columns are sampled; the strictest evidence wins, with
internalas the floor, and only a person setspublic. - Samples are reproducible: 200 non-null values per text column, by a seeded reservoir sample on one DuckDB thread.
- Presidio runs offline, on
NoOpNlpEngine, with 6 recognizers, each with its own threshold, phone regions read from the learners' countries, and tldextract's own copy of the Public Suffix List, checked whenlib.piiloads. - Detectors are measured against labelled spans: recall per entity type by exact and overlap matching with a label of the right type, coverage whatever the label, precision per label and texts left with nothing personal, with recall weighed first.
- A proposal stricter than the metadata is a finding:
support.tickets.bodygoes toteam-supportassensitive, andbilling.payments.card_numbertoteam-billing, with the evidence that it holds full card numbers. - Free text is masked with two detectors: OpenAI's Privacy Filter, recorded at a pinned revision and operating point, for names and addresses, and the tuned patterns for card numbers, IBANs and phone numbers, whose checksums keep card numbers labelled as cards. Masking hides whatever either finds; neither is trusted to make text anonymous.
- No example prints a whole phone number, card number or IBAN:
lib.pii.maskedkeeps the last four digits. - Reclassification follows field IDs: every field the metadata lacks is classified in the pull request that adds it, and text columns are sampled again on a schedule. A detector is only as good as the measurement behind it, and what it misses never appears in its output. The measurement is what tells a team which of its classifications it can rely on, and which it has to check by hand.
本章,在本版次中
版次 2026.11.31示例于 2026年11月19日 通过本章 11个示例 个示例中有 11 个通过
译自英文版图书,版次 2026.11.31。