no filtering without ham

[Matthias Andree]

On Sun, 18 Jan 2009, RW wrote:

> > I'd discourage that. It unnecessarily pollutes the database.
> 
> Not if you learn a single fake ham, with the minimal content required
> to create a database and turn-on classification. 

If that actually helps - and it might give users pretty many false
positives at first, which may turn out to be even more annoying than
false negatives.

> This seems a pretty good solution because it conforms to the
> expectations and needs of someone who has a hand-waving idea of what a
> Bayesian filter is, doesn't want to to any reading, and just wants to
> get on with it. My guess is that that describes most kmail wizard users.

I'd agree on the wizard user and "doesn't want to do any reading" party,
but not on the solution.

> If they understand that such filters need to be trained at all, they'll
> understand it might be erratic to start with - it's just common sense.

I'd rather have it err on the safe side and let spam slip, rather than
ditch solicited messages into the trash bin (this is in the light of
Sven's posts that KMail defaults to 'move spam into trash').

-- 
Matthias Andree