How about doing OCR of paper PDFs to allow for entire dataset collection of only its text?
#8
by
g370
- opened
Its a no brainer that PDF files will be much bigger than the raw text, one could plainly extract raw text from the pdf, but you can't guarantee that ALL pdfs submitted will have raw text in a same format that is would be easy to extract.
If you guys add OCR parsing of whole pdf as it processes each of them, the extracted text can then be compressed decently and the final dataset would be considerably less than 1TB?