AI Optimization

Could you please direct me to a previous discussion regarding the best DB structure such that it optimizes the AI algorithm? I’m looking for issues such as the best use of tags vs nested tags, file name, metadata, etc. to get the optimal results from the AI classification suggestions.

Thanks in advanced

BTW, kudos to the MacSparky DT Fieldguide. It has rekindled my desire to take a much deeper dive into DT3. For anyone thinking about buying it, I personally give it a 5-star rating, I just wish he also offered a more advanced course too!!!

From Take Control of DEVONthink, p. 28:

About DEVONthink’s Artificial Intelligence (AI)
When DEVONthink says a feature uses “artificial intelligence,” what
does that mean? To oversimplify somewhat, DEVONthink has a fairly
sophisticated algorithm that examines the text in a given document
(along with document properties and custom metadata), compares it
to the contents of all other documents, and mathematically computes
the degree of similarity for each one. (It’s analogous to the process a
spam filter uses to determine, based on an analysis of an email
message’s contents, whether it’s most likely spam or legitimate.) This
helps you determine where a document belongs and which docu-
ments might be related to it.
Note that DEVONthink’s AI does not examine filenames or tags, so
you can’t use them to give DEVONthink “hints” as to what you think
should go where. And, because DEVONthink’s decisions are largely
based on word similarity, if you have multiple groups whose files
contain quite similar words, DEVONthink may suggest the wrong
destination. Remember, it’s not magic—just statistics!
DEVONthink’s AI engine is also used by the Summarize feature, and
helps DEVONthink suggest names for scanned documents.

I don’t believe this has changed. Hope that helps.

1 Like

Anything our late, beloved Bill Deville said on the subject re: organization and AI is still good…