5 ms·
Seems like a large part of what's needed is just being able to make the pdfs machine-readable, by making decent plain-text versions of the text content. Right n
by generalizations 5y ago
Seems like a large part of what's needed is just being able to make the pdfs machine-readable, by making decent plain-text versions of the text content. Right now, IIRC, there's no hands-off way to get the text of a pdf. Especially if there's weirdness like multiple-columns (sometimes happens with this stuff).
- zapataband1 5y agoCurrently working in this field and this is actually the cutting-edge(!!) but it will be 100% possible/robust within the next year or so I believe. Really cool ML techniques being used for htis.
- sleepingsoul 5y agoHow well does it work with old OCR'd PDFs? :)
- generalizations 5y agoI had no idea. Is there anything that's open source, or is it all still being kept proprietary?
- zapataband1 5y agoLike you said the hard parts are the unstructured data/images/tables. There are pretty-good(80% of the way there) solutions tho. But nothing that could handle millions of paper without error