Custom and Default Entities Not Fully Anonymizing Text

[cristianosticca-pagopa]

Context
I'm using Presidio Analyzer and Anonymizer to detect and anonymize personally identifiable information (PII) in Italian text. My setup includes both default entities (e.g., IT_FISCAL_CODE) and a custom entity for Italian vehicle plates (IT_VEHICLE_PLATE).

Issue
The following text is analyzed:

"My license plate number is AB123CD and my fiscal code is DNTCRL65S67M126L."
However, only the fiscal code (DNTCRL65S67M126L) is detected and anonymized. The custom entity IT_VEHICLE_PLATE (AB123CD) is not recognized.

Code Overview (see attached code)
test-presidio.txt

NLP Engine (Spacy) is correctly configured for Italian
Custom pattern recognizer for IT_VEHICLE_PLATE is added to the registry
Analyzer is set to detect both custom and default entities

Expected Behavior
Both AB123CD (license plate) and DNTCRL65S67M126L (fiscal code) should be identified and anonymized.

Question
Could anyone support me? Any guidance would be greatly appreciated! Thanks! 🙏

[omri374]

Hi @cristianosticca-pagopa
Everything is correct except for one small change needed. If you don't specify the language for the new custom recognizer, it would assume it supports English. Here's the fixed definition:

plate_recognizer = PatternRecognizer(patterns=[plate_pattern], 
                                     supported_entity="IT_VEHICLE_PLATE", 
                                     name="IT_VEHICLE_PLATE", 
                                     supported_language="it")

I would further suggest to add context words as these could boost the confidence of this being an Italian license plate and not a different entity.

[cristianosticca-pagopa]

Hi @omri374 thank you very much for your help!
Really appreciated.