Smart Scan: Auto notebook selection and tagging recommendations based on AI/ML

[Bill Ferry 1]

It would really be powerful if Evernote could use Artificial Intelligence/Machine Learning to make notebook and tagging suggestions based on how previous notes have been placed (notebook wise) and tagged.  An example process would be:

• New document scanned
• Evernote recommends, based on the contents of the newly scanned document and similar documents that have been scanned in the past, a notebook and possible tags for that new note.
• The user can then confirm/validate the recommended selection.  Maybe, at some point, the user can select an option to just accept the smart recommendations automatically

[PinkElephant 6,211]

Thinking about the post i see 3 scanning options:

1. Using the build in scanner
2. Using (on iOS) the Scannable app
3. Using a 3rd party scanner, with his own scanning software

Which of the options do you mean ?

[Bill Ferry 1]

I think you could use a variety of scanning methods.  I use Scansnap to get my docs into Evernote. 

[PinkElephant 6,211]

Yes, but you make a proposal. The 3 methods mentioned above have different ways how they create content in EN.

I don‘t think you can easily say „OK, lets put an AI actively sort and tag the scans“, and it works on each method the same.

[Bill Ferry 1]

Maybe I don't understand how it works but I would assume that Evernote has access to the data derived from a document via OCR that allows it to search documents (at least in the paid versions) regardless of the method that is used to import the document.  If that is the case, it can use that data plus past notebook/tagging done by the user to make suggestions on new documents that get scanned in.

[PinkElephant 6,211]

OK, this could work, but: Not all scans get OCRed. If the pdf already contains a significant text layer, it will not be OCRed again.

And OCR as a server side function happens with a time gap to all other note actions. It works in a queue, so it may happen within minutes, but sometimes it is delayed several hours, if the queue is busy.

Because no AI is perfect, it would have to apply a proposal tag and notebook, and the user would have to confirm it. This means that compared to today any note with a scan or picture (some scans come as a picture file) would have to be touched twice.

To avoid touching notes twice, the AI process needed to happen in the upload. But then it is often not yet OCRed. 

There is already an AI based approach for web clips and email. These come computer legible and don’t rely on a server OCR. Maybe check them out - it may help to refine the initial idea.

[Bill Ferry 1]

Thanks so much for your thoughtful comments.  You raise some good points.  I'll have to check out the web clips and email suggestion.

You got me thinking about the delay.  Not sure about others but I would actually be ok to have the AI and proposal portion done on a delayed basis as I oftentimes upload/import many documents at once.  I would love to be able to come back at a later time to do a quick overview of the proposed notebooks/tagging and confirm.  I still think this is a time saver and would only get better with more data that gets processed.

[PinkElephant 6,211]

Thanks for the exchange of ideas. I am not opposing it, I want to sharpen the focus.

The forum is user2user. To reach out to EN PM , you can use the feedback function build into the clients.

[Bill Ferry 1]

Thanks for the clarification on this being user to user.  Thought it would be something EN would be reviewing but I can use that feedback function.