Does it really check the context words while detecting PII data? #1634

[pythodev]

Suppose I am detecting a credit card number in a text. Does it check the context words to detect the card number?

For example: (#1607

I noticed that it's detecting timestamps as credit card numbers if they pass both the regex and the Luhn check.

I want to include context words in the detection logic. For example, in this case:

uid: 3714 4963 5398 431

Even if it passes the regex and Luhn check, it should not be detected as a credit card number because of the context word uid.

How can I include context word checks so that detection only happens when appropriate?

[omri374]

Hi, apologies for the late reply. The current logic for the CreditCardRecognizer, is that values passing the checksum validation result in a score of 1.0. This can be changed by tweaking the CreditCardRecognizer validate method and adding this custom recognizer instead of the predefined credit card recognizer.

With regards to the example you've given: Presidio looks for context words before and after an entity. You can potentially create more sophisticated context mechanisms and pass those to Presidio (see this class), to handle specific cases like ENTITY_NAME: ENTITY_VALUE. Note that in real life text you could also have things like uid, 3714 4963 5398 431 is my card where there's only a difference of a punctuation mark (, instead of :) but the context is completely different.