Taglocity General Discussion Forum

I spent the afternoon tagging my emails in my inbox and creating my tag cloud.  I have roughly 2000 emails now tagged .  I didn't have the auto tag option turned on for my tags as I was doing this because I thought I would go back and tag all my historical data first and then turn on the auto tag feature and train it. 

I am not however finding this training option.  Does the program only train / learn when an email comes in and is first tagged?  Is there a way I can make it analyze my folder and relearn based on the tags that have been manually done?  Basically I want to point it to a folder and say, these are all correct tags, and I want it to train on them.

I see the import from folders feature, but that is suppose to be for people who already had emails seperated out by a subject and what to train a specific tag on it.  Figured this out because when I tried it on my inbox for a tag, it went through and tagged a bunch of emails with the tag I tried.

So do I need to make some rules that go around and make copies of every email and put them into a folder for that tag, and then run the folder to auto tag on each folder that contains all the emails that contain that tag?

BTW, I am finishing my masters in CS in intelligent systems this next semester and thus we have studied learning networks etc... quite a bit and I find this very impressive and joyful that someone has applied to this email.  Its actually how I ran into this program was after building a part of speech tagger and phrase tagger for sentences, it hit me email contains context etc... and I thus began my search. :)

Thanks!
Michael

Hi Michael,

First off, thanks for trying out Taglocity - I must admit of all the different parts of it, I do like the AI side and could talk about this aspect for hours :-)

The assumption is normally that tag is set as an AutoTag before it is applied to Outlook items. In the beginning it is 'empty' of statistics pool, and that each time it is used then it 'learns' a little more. For results to converge it's also useful for that default algorithm to 'unlearn' incorrectly assigned content - which is why large batch training tends to be less accurate. It's still a good start though.

I understand what you've done already though, and I can see the need for a retrospective 'Learn Now' menu option. We can look at adding that in the next update.

The workaround for now is just to 're-tag' as in apply the tag in batch again to the same content. Now they are set as AutoTags then they will learn when applied.

If you select a batch of existing ones with the tag, untag and then tag again the 'unlearn' and 'learn' will then fire. The unlearn won't really matter as the first step, as the stats are empty anyway.

On a new system I only tend to select 30 or so examples for the existing training pool, so you might not have to select/retag that many to get going.

PS The default AutoTag CRM114 OSBF algorithm is something that we are planning to expand upon in the upcoming releases. If you're interested in trying out a few more derivatives out in 'beta' then just send an email via [email protected] and we can set something up...