Names with Presidio

The built-in recognisers find names after a title (Hr. Keller), as first name plus surname from a list of common first names, and a surname again later in the same text. For other names in running text you can add Presidio (MIT, runs locally). It is not a dependency of Isomorphism and nothing is downloaded by this package.

pip install presidio-analyzer
python -m spacy download en_core_web_lg    # or the model for your language

from presidio_analyzer import AnalyzerEngine
from isomorphism import anonymize
from isomorphism.presidio import presidio_detector

detector = presidio_detector(AnalyzerEngine(), language="en")
safe, mapping = anonymize(text, detectors=[detector])

Presidio hits show up as PERSON or ORG findings with detector == "presidio". A hit that overlaps a better hit loses against it, and names of authorities are still protected. For German, French or Italian texts configure the AnalyzerEngine with the spaCy model of that language and pass the matching language.

The quality of the names depends on the model: small models (*_sm) confuse persons and companies. Large models do better. The tests of Isomorphism use a fake analyzer; the adapter was also run once against Presidio with en_core_web_sm.