Submit Hint Search The Forums LinksStatsPollsHeadlinesRSS
14,000 hints and counting!


Click here to return to the 'The filter is ALWAYS learning' hint
The following comments are owned by whoever posted them. This site is not responsible for what they say.
The filter is ALWAYS learning
Authored by: notmatt on Apr 18, '03 11:17:35AM

The junk email filter always 'learns' according to manual junk/not junk assignments, which is nice (do a quick search of the spam articles in the Apple KB for confirmation).

While I can't say for sure, it seems likely that dragging messages to the junk box will do the same; when they're dragged in, they're assigned junk status automatically.

As for the difference between automatic and training modes, if I had to guess, I'd assume that training mode used ALL mail, both automatically-assigned and manually to train on, while automatic mode used only the manually-assigned messages to alter the filter model. Might also be less sophisticated, and simply change a couple of constants.



[ Reply to This | # ]
The filter is ALWAYS learning
Authored by: aranor on Apr 18, '03 12:51:53PM

From what I understand, Training mode is just like Automatic mode, except that instead of moving messages to the junk mailbox, it simply colors them tan. This way you can train the junk mail filter to act appropriately before having it auto-move stuff to the Junk mail folder.

If you'll notice, the junk mail filter settings (i.e. Training vs. Automatic) simply change a rule in Mail.app's rules list.



[ Reply to This | # ]
The filter is ALWAYS learning
Authored by: notmatt on Apr 18, '03 01:43:56PM
If you'll notice, the junk mail filter settings (i.e. Training vs. Automatic) simply change a rule in Mail.app's rules list.

Yeah, but it's the "this message is junk mail" bit where all the magic happens, and anything could be going on behind the scenes.

What makes me think there IS something is that training sets for latent semantic analysis are generally a LOT bigger than the number of mails the typical person will click on as "junk" or "not junk" before turning it to automatic. Training with that level of supervision is also notoriously slow, and even though Apple gives a pretty good default model, it seems like you'd need something more intensive at first. Similarly, you would actually want to stop intensive training at some point, or it becomes very hard to adapt to 'new' junk mail. All this makes me think that there's something else going on behind the scenes besides switching what happens to the positive results.

I tried searches on google and citeseer, but couldn't find anything. I'm getting more and more curious.



[ Reply to This | # ]