Add Presidio integration

[julian-risch]

Summary and motivation

Presidio is an open-source framework for detecting, redacting, masking, and anonymizing sensitive data (PII) across text, images, and structured data.

Detailed design

The idea is to have a new Haystack component called PresidioDocumentCleaner that accepts a list of Documents as input and transforms the text of each document's text into redacted output text. It should return a list of new documents and leave the inputs unchanged. Details on why we need to avoid inplace mutation are here: https://docs.haystack.deepset.ai/docs/custom-components#requirements
I believe we should have another component, PresidioEntityExtractor. With that approach PresidioEntityExtractor corresponds to a Presidio Analyzer and PresidioDocumentCleaner corresponds to a Presidio Anonymizer.

For example an input document with the the text

should be transformed into an output document with the text

Haystack recently added a similar integration with Tonic Textual: https://haystack.deepset.ai/integrations/tonic-textual and the components TonicTextualDocumentCleaner and TonicTextualEntityExtractor can serve as a reference for PresidioDocumentCleaner and PresidioEntityExtractor.
We should add a thrid component PresidioTextCleaner analogue of the PresidioDocumentCleaner. It should take strings as input instead of documents and also return strings instead of documents.

Checklist

Follow https://github.com/deepset-ai/haystack-core-integrations/blob/main/CONTRIBUTING.md#create-a-new-integration if you would like to contribute. In particular, use the scaffolding script:

python scripts/create_new_integration.py --name presidio --type preprocessors

Ensure the following checklist is complete before closing this issue.

Tasks

• The code is documented with docstrings and was merged in the main branch
• Docs are published at https://docs.haystack.deepset.ai/
• There is a Github workflow running the tests for the integration nightly and at every PR
• A new label named like integration:<your integration name> has been added to the list of labels for this repository
• The labeler.yml file has been updated
• The package has been released on PyPI
• An integration tile with a usage example has been added to https://github.com/deepset-ai/haystack-integrations
• The integration has been listed in the Inventory section of this repo README
• The feature was announced through social media

[sjrl]

Additionally I think making a PresidioTextCleaner analogue of the PresidioDocumentCleaner would be good to have to remove PII from incoming user queries. This request has popped up in the past from clients since they've had users who put in PII even when they aren't requested to, due to confusion on the usage of the client's platform.

[julian-risch]

Additionally I think making a PresidioTextCleaner analogue of the PresidioDocumentCleaner would be good to have to remove PII from incoming user queries. This request has popped up in the past from clients since they've had users who put in PII even when they aren't requested to, due to confusion on the usage of the client's platform.

Same inputs and outputs as Haystack's built-in TextCleaner, which is list[str], correct? Or for a query, did you have in mind just str? https://docs.haystack.deepset.ai/docs/textcleaner

[sjrl]

Additionally I think making a PresidioTextCleaner analogue of the PresidioDocumentCleaner would be good to have to remove PII from incoming user queries. This request has popped up in the past from clients since they've had users who put in PII even when they aren't requested to, due to confusion on the usage of the client's platform.

Same inputs and outputs as Haystack's built-in TextCleaner, which is list[str], correct? Or for a query, did you have in mind just str? https://docs.haystack.deepset.ai/docs/textcleaner

Oh interesting, I would have said for a query so just str

[julian-risch]

Smart connections would make connections still work if PresidioTextCleaner had list[str] and other components before or after it had only str. So for consistency I'd suggest we go with list[str].

[srini047]

Working on this, will try to raise PR soon.

[srini047]

@julian-risch @sjrl
One question based on the current analysis: Should we have support for multiple languages, default supported by presidio is English en?

If yes:

• Should we make it dynamic where the user can bring in the langauage(s) and the model(s) via a configuration file?
• [OR] Should the integration natively support some default languages, which can be tweaked based on request?
• [OR] Should the user provide the instance to Engine and we can directly make use of it. Incase a user doesn't provide we can stick to default AnalyzerEngine