presidio sometimes misses very crucial information such telephone number

[sadathknorket]

Given that I used this tutorial https://github.com/microsoft/presidio/blob/main/docs/samples/python/example_pdf_annotation.ipynb

and this pdf example
personalpdf.pdf

The outputted pdf was
output_annotated.pdf

We could notice telephone number was missed to be annotated .

Is there any extra parameter or something I could do to avoid such mistakes ?

[omri374]

Hi, for telephone numbers specifically, Presidio is configured to search for numbers belonging to some countries but not all. See this issue, where the supported_regions param is updated: #1140

For addresses and others, I would suggest to look into transformers based models. See this example (still in PR) for a full presidio customization: https://github.com/microsoft/presidio-research/blob/omri/evaluator_updates/notebooks/5_Evaluate_Custom_Presidio_Analyzer.ipynb

[sadathknorket]

How does it know about the region ? Through the phone number itself ?

[omri374]

The user has to provide the regions it wants it to scan. There is no region detection here. If your data likely has Turkish phone numbers, you should add Turkey to the list of supported regions.