Depth of structure for optimal find?

Hi all.

I am still trying out DT personal, and every day I love it more, but I have one question…

When I first started importing my local files and indexing the remotely-stored files I need to access, I built a pretty deep structure for my documents. For example, if I want to store a document about how to draw a horse, I would file it in Personal -> Hobbies -> Drawing -> Instruction. That seems like a lot of levels, although some of those 4th-level-down groups have many documents in them.

Is there some sweet spot to how many levels are useful to the AI for classification, beyond which it’s not helpful anymore?

Many thanks,

Hi, Robin. I don’t believe that the depth of hierarchical nesting is particularly relevant.

What is relevant is the degree to which the contextual usage of words in a new document is similar to the contextual relationships of words used in the documents in each of the groups in your database.

In your example, it’s likely that a document describing the techniques for drawing a horse would result in a suggestion by the Classify AI that a place for your document might be in the group named Instruction. That suggestion would be made whether Instruction is located at the top level, or at the seventh level in your organizational scheme. It would be made because of the terms and their frequencies common to articles about drawing things, whether you are drawing people, houses, birds or horses.

But if your database also includes a group containing documents about the care and feeding of horses, and if your document about drawing a horse frequently uses the term “horse”, that second group might also appear in the list of suggested locations for organization, regardless of its position in your organizational structure.

Hope that helps.

Thanks Bill. That actually helps a lot, because right now the depth of the hierarchy is making me a little nuts, as I was trying to be as specific as possible, but now it sounds like I don’t have to have so many levels in order to benefit the AI.

  • Robin