Build More Accurate Custom Sensitive Information Types in Microsoft Purview
Diesen Beitrag auf Deutsch lesen
Combine regex, supporting keywords, proximity settings and real sample testing to reduce false positives in custom Microsoft Purview sensitive information types.
TL;DR
Microsoft Purview custom sensitive information types use structured regex patterns, nearby supporting keywords and proximity settings to detect proprietary identifiers. Keyword dictionaries group related terms for reuse, while positive, negative and borderline sample files help validate accuracy before enforcement. Loose patterns and generic terms can create noisy alerts.
Original on M365 FM. Read the original · suggested by Mirko Peters
This is our own summary, not a republication or full translation.
Governance takeaway
- Makers: Use structurally precise regex patterns for proprietary identifiers, because loose numeric matches can flag phone numbers, invoices and routine spreadsheets.
- Admins/CoE: Maintain specific keyword dictionaries and define tight proximity windows, because unrelated terms or distant context can create false positives.
- Security/Compliance: Test custom types with real positive, negative and borderline documents before enforcement, then refine patterns based on observed misses and false matches.
Frequently asked questions
How do custom sensitive information types detect internal employee IDs in Microsoft Purview?
They combine a regex pattern describing the identifier structure with supporting evidence such as staff number or HR record. This adds business context to the match.
What does proximity control change in a Microsoft Purview sensitive information type?
Proximity controls how close supporting keywords must be to the regex match. Tight proximity helps prevent distant document text from creating unrelated detections.
How should Microsoft Purview custom sensitive information types be tested before enforcement?
Upload real business documents and emails containing positive, negative and borderline examples. Review false matches and misses, then refine the regex, keywords or proximity settings.