Taglocity General Discussion Forum

I have just installed Taglocity, but I don't appear to be able to get AutoTag to work for me. I've done a whole load of manual tagging. I've tried selecting a number of untagged emails. I toggle autotag 0n/Off ect  Either I get nothing or outlook hangs.

I've tried the set-up autoTag. i.e. it suggest I point it to a folder (my inbox), but then it asks me to name a tag (??), defaults to InBox. I accept that but all it does is tag the 50 samples with the tag 'Inbox. This is not what I expected. I expected that it would analyse 50 emails and suggest a whole bunch of tags.

Does this point to the fact that the AI part is not functioning?

Hi Graham,

Autotags are working in the Beta 2, or rather we've had no bug reports so far. We've had some Beta 2 users who are seeing some good Autotag classification results, so up to know we saw no problems.

I think there is a misunderstanding on what the 'Tags from Folders' start-up screen helper does. This is used where you have a folder that contains information on a single topic and you would like to create a single trained AutoTag from it - it then goes through and reads all the content to 'teach' it some starting statistics. It is not used to suggest tags (a harder problem, but interesting idea), but populate a single one. I hope that makes sense.

For any Outlook.exe hangs, the 'in-application' 'Taglocity->Report a Beta Bug' diagnostic information would really help us here.

If you are not seeing AutoTags come on it may be that the classifier hasn't got enough good hits yet. If you look at the text file 'classify.out' in your <install dir>\Classifiers\crm114 you can see the details of each scoring.

Thanks it does appear to be working, I'm getting a CRM.out file generated with some content. I just don't think I have enough data yet for it to work with.

How much manual training is needed. 50/100 emails? I understand that you also need to train the AI as you go along.

Thanks for the explanation for the initial set-up. It makes sense now in relation to helping someone who has already setup a lot of sub-folders.

The amount of manual training is more to do with what 'statistically identifies' the AutoTag as unique rather than the amount of content.

If you have 10 AutoTags that have some very similar content then the 'decision recognizer' value becomes very small.

Put another way, if the default filter is having a hard time recognizing between the AutoTag stats then it will not return a positive hit as it's confidence will be too small.

Converserly, if you ever just set up one AutoTag then you will find it *always* hits, which isn't that useful either.

The default algorithm in Beta 2 is 'phrase based' and something called 'Orthogonal Sparse Bigram' (OSB). A really nice reference on this algorithm can be found here: (warning 2mb PDF download) http://crm114.sourceforge.net/CRM114_Revealed_20051207.pdf

I hope this helps.

Err not really, but I do recognise a 'it depends' answer :-).

One of the good things about beta testing someone else's product is that you can act like a real user and be an dumb as you like.

I'll try to record some metrics for my scenario, maybe we can work out some yardsticks. I would suspect that people's expectations need to be set.