feat: pdfMiner Document Parser

[espoirMur]

Feature description

Create a new PDF parser based on the pdminer package.

Motivation

For simple PDF files and parsing, it's good to have a lightweight PDF parser.
It can handle small PDFs without images and tables.

Additional context

No response

[espoirMur]

I will have a look in implementing this.

[mhordynski]

@espoirMur thanks for submitting this issue.

I like the idea of having lightweight PDF parser. Looking forward to the implementation 🙌

[micpst]

@espoirMur why do you want to implement pymupdf parser? It has more restrictive license, is potentially slower and has some quality issues (https://arxiv.org/pdf/2408.09869). If you can prove that it works better than the default docling parser, we can discuss adding it to the lib.

[espoirMur]

Hi @micpst !

I agree with you pymupdf licence is restrictive. The parser I was referring to is not pymupdf but pdfminer.

It is the library openparse used to parse PDF without OCR capacity.

I have worked with Docling today, it is working fine for some of my PDFs but not others. In terms of the quality of the document chunk, it is better compared to pdminer.

I tried it on 700 documents in my documents folder, but it is very slow, uses a lot of memory, and sometimes crashes my laptop(a MacBook M1 16 GB of RAM).

I am trying to replicate the issue and isolate it. I will open another issue in Docling if I manage to isolate the issue.

[micpst]

There is an option to turn off OCR in docling. Maybe we can expose more params in our parser so that it can cover more use cases.