2
$\begingroup$

(If this is not the right forum for this, I apologize)

I'm trying to write a program that can automatically name scanned PDF files depending on what text they contain on their first page.

I have ~10,000 PDFs that are already named the way they should be; there are rules to how the files are named (that I don't know many of myself). So, I want to make a program that can detect patterns/clues/whatever the correct term is, in the content of each file, and see how it relates to the naming of the file.

I have a working solution to parse the text from the PDFs. It works pretty good, with the exception that it can't parse handwriting very good, if at all. 80-90% of the scans are printed forms, letters, or other official documents, and it can parse those fine. I used .Net C# for this.

I have no experience with ML. Frankly, I don't even know if it applies here. Just thought this could be a fun project to try my luck with. At least it's more interesting than going through the documents and finding the naming rules manually...

Is this something that could be accomplished? Any advice or input on where I could start reading up on an area that covers this, if nothing else? Thanks!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

It is indeed possible by NLP.

You can try 2 approaches:

  • Try unsupervised topic modelling. Since you have annoted 10000 documents ,use them to evaluate your models and choose best performing one .

  • Since rules are working fine. You can training on 10000 docs using NAvier bayes .

Improve this answer
$\endgroup$

Your Answer

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

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