0
$\begingroup$

I'm working on a project to recognize confidential info like social security number, name, driver license number, financial details(like credit card number, account details etc), address, certain confidential info on legal and medical documents from a user-uploaded pdf, my question is let's say I collect some 2k records on financial details, 3k records on legal related terms, can I train only one model to do all these tasks? or separate models for each domain? for e.x: for finance separate model, for legal separate model, etc

I'm very new to the NLP and I don't have much idea, any suggestions on where can I get the data? and what techniques in NLP I can use for this task?

p.s: this problem is both cv and nlp related, cv for the ocr part and nlp for rest, please read the question completely and mention in comments if you have any doubts before downvoting.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

From the NLP side, i.e. assuming you have the data as text, this would be typically treated as a NER problem, i.e. with a sequence labeling model.

The task look similar to text anonymization. This is a area of active research. If the goal is to recognize all the potentially sensitive information in a text, it's important to keep in mind that there's no perfect method. In general performance is higher with a specific domain than with a broad domain.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.