Ask HN: Is there a ready-to-go solution to parse documents content? | Hacker News Reader

Ask HN: Is there a ready-to-go solution to parse documents content?

27 pointsfpd444417 comments

Hey there

I'm looking for a ready-to-go solution to parse documents (such as pdf, docx, pptx and others). By 'parse' I mean text extraction, including OCR if needed. I know about Tika and tried it already, but are there any more reliable alternatives, maybe based on Tika? I'd like to interact with it via REST API.

Thnx

View on HN

Tika has a REST server built in https://wiki.apache.org/tika/TikaJAXRS

Is there some functionality that you need and is not covered there? Some specific document extraction feature?

ElasticSearch has support for ingesting a bunch of document formats if you're already using it / looking at using it in your stack:

https://www.elastic.co/guide/en/elasticsearch/plugins/5.x/in...

Elastic is using Tika.

Yep Tika is pretty much the choice in this situation. It's not perfect but it's good enough for most purposes.

Yes, the https://www.ibm.com/watson/developercloud/document-conversio... Watson Document Conversion service meets those requirements. It's not free, and it's not popular, but it's reliable.

Good one, will look on it

I use 'poppler' or Apache's PDFBox for text extraction from PDF. They both can write HTML or their own XML format. In addition, they keep the absolute positioning of the layout.

For XML files, there is XSL-T. A simple run with the default template will give you all strings in the document, if you really want just the paragraph text, you will need to find/create an XSL transform.

None of these is ready to go, but very close to it. Epecially in the case of poppler and pdfbox.

Thanks guys. But none of your suggestions solve the whole problem (some don't include OCR, some support only limited file types and other). I'd like to have a black box that does everything for me (does OCR if needed, extracts pds, docs, txts and others).

But I'm afraid there's no such solution...

I'm afraid I may be late to the party, but I've seen https://github.com/openpaperwork/paperwork before and it looked like a good solution for this. Never tried though.

It's not an enterpise class solution ;(

What makes an enterprise class solution ?

We've been using Google Cloud Vision's OCR service with pretty good accuracy (varies from ~80-99%).

https://cloud.google.com/vision/docs/

We've used Aspose for manipulating PDFs but they work with "over 100 file formats". Offers both SDKs and RESTful APIs

https://www.aspose.com

Maybe this https://rawtext.ambar.cloud/ ?

Tika, pdftotext, lynx (html), tesseract (ocr)

It's very difficult to combine it together, as documents in different types has a lot of edge cases

Yeah, it's a beat effort deal, but a pretty reliable one (except maybe tesseract-ocr)

Reply on news.ycombinator.com