This site has some nefarious cloudflare thingy on it that tells me that my machine is infected. I am quite sure that is not the case. It will not let me access the website in question unless I enable third party cookies which is something I don't want to do. Please don't use services like cloudflare.

Edit: Why the downvotes? This is what I am seeing: http://imgur.com/YlLWH.png

Is this just sort of generalized bayesian spam filter? I always thought that if it could learn to label some messages as spam, it would be able to apply any other label as well.

Indeed PG alludes to bayesian text classification being an old an established field (here: http://paulgraham.com/spamfaq.html and other posts about spam filtering)

Something like that. I've been doing this stuff since 2000: http://getpopfile.org/

Ahhh, I knew your name was familiar.

POPFile has an IMAP module that makes something like this pretty easy to set up. You just need to connect POPFile your GMail account, set up some buckets for POPFile, and then tell it how to map those buckets to IMAP folders.

It takes a little time to train the filters, but once that's done, the accuracy is pretty good. My only problem now is coming up with a good labeling system.

Yes, it does, and for power users they can certainly go that way. Although, POPFile doesn't support OAuth for IMAP at this point so you have to trust your configuration with your GMail password.

I started experimenting with something similar (although got nowhere) as a helper for getting back to Inbox-0 from an overflowing inbox.

The concept was fairly similar, go through the inbox and sort it into piles of important, replied, unreplied, deletable, twitter statuses etc, and then the user could deal with archiving/deleting/following up more quickly.

This looks like a much better solution however, something running the whole time and stopping me getting into that mess.

It doesn't look like anything that Mutt + Procmail couldn't do, other than the web interface. Is this something Google should consider setting up for Gmail?

That sounds pretty awesome. I've got an idea for something kind of similar which I might try, but yours would probably be way better so please do it :)

I handle my mail with imapfilter (http://imapfilter.hellug.gr/#overview). Can I reuse the machine learning part ?

Would you be willing to share the details of the algorithm?

PDF: http://www.extravalent.com/polymail-datasheet.pdf

naïve Bayesian classifier and an inverse chi­squared test

Woah. Congrats on finding that, it's pretty old. The actual classifier I'm using has moved on a bit and I'm currently using logistic regression. Here's a good paper on that: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.126...

Is the code available for this or is it something you have written yourself?

Looks like a fairly straightforward spam filter, but for a couple of "dimensions" (school/gaga-1) in parallel.

To be really lazy, perhaps the Google prediction API could be used.

Or directed edge :)

http://directededge.com/signup-developer.html

(don't forget it's free for noncommercial use)