Training SpamAssassin

[Marc Schwartz]

On Sat, 2003-12-13 at 15:14, Alexandre Oliva wrote:
> On Dec 13, 2003, "Neil B. Cohen" <nbc at aikisoft.com> wrote:
> 
> > I have installed SpamAssassin on my mail server and it works very well.
> > But it is still picking up only 50-60% of the junk delivered to my
> > mailbox. So recently I started saving the junk that gets through into a
> > SPAM folder and then I tried running
> 
> > sa-learn --spam --file spambox
> 
> > on that folder. I had 250+ messages in the folder, but when it finishes,
> > it says something like "Learned from 1 message".
> 
> If it's an mbox file, use --mbox, not --file.  --file is for one
> message in a file.
> 
> > 2) Can I expect to get SpamAssassin to get up to the point of getting
> > rid of 90% or 95% of the junk that crosses the Net these days?
> 
> It does for me, but it's very important that you train it not only
> with spam, but also with good e-mail, otherwise you may skew the
> results or not even have enough data for it to kick in.


One other thing to do. In ~/.spamassassin/user_prefs, add the line:

score MICROSOFT_EXECUTABLE 5.0

That will take care of the MS executables that are contained in incoming
e-mails and force the score to 5, which is the default threshold for
spam.

Once I added that, it eliminated most of the virus/worm crap. I have an
evolution filter that deletes those messages specifically, whereas the
others I move to a spam folder for subsequent review.

HTH,

Marc Schwartz