Re: [Evolution] Re: junk filtering

[Srinivasa Ragavan <sragavan novell com>]

On Tue, 2006-01-10 at 21:34 +0100, guenther wrote:
> > > The question is not whether you can use the buttons. Yes, they are
> > > always enabled and do "something".
> > >
> > > The problem is when you want to use it to learn ham. Apparently it was
> > > not designed for that, it looks like it was designed to undo a previous
> > > "junk"-button-press, because only then something actually happens, being
> > > the bayes database being updated. If you push "not junk" on a random
> > > message, the message doesn't get learned as "ham".
> >
> > So this is a big issue, currently on HEAD, we have either the "Junk" or
> > the "Not Junk" button, based on wether the messages has been marked junk
> > or not. We actually needs to discuss this further. Am CCing Srini for
> > his comments on this.
>
> Revert that commit immediately.
There is no issue in reverting it, but why do we want a 'All the time
disabled - Not Junk' button on toolbar. It is enabled just in Junk
Folder. If the junk implementation extends learning not-junk, then it
would make sense. 

> This is a bad idea. Really. All you gain is a few pixels in the toolbar.
> But if this UI change is introduced in 2.6, it will be harder for the
> developers and confusing to the users to get the other button in again.
>
> Questions regarding "Junk filtering not working" majorly impacted by a
> badly (if at all) trained Bayes are common these days. To make SA more
> effective, the user needs to train Bayes. So there *must* be a way to
> train it. The user has to be able to make SA learn *any* mail marked as
> either Ham or Spam previously.
>
> Automatic learning is *not* done for any mail. The score has to be above
> (or lower than) a threshold, which is way more extreme than the "spam
> limit".
>
> For Bayes to kick in SA needs to be trained 200 Spam and Ham *each*.
>
> These buttons should not be "mark as" Ham/Spam, but "learn as" Ham/Spam.
>
> ...guenther