[NCLUG] Beating spammers with SpamAssassin 2.63

[jbass at dmsd.com]

Hi Guys,

If you are using a well known email address under linux and have
a spam problem, you might want to upgrade to SpamAssassin 2.63
and take the time to train the BAYES filter with sa-learn.
SpamAssassin also now has RBL support by default, so you don't
have to configure it into the SMTP deamon.

   http://www.spamassassin.org/index.html

They are sourceforge hosted, and in the Debian archive. Older
versions are shipped with RedHat and Fedora - if you are using
those I suggest upgrading. There are RH9 rpm downloads available
which I suggest installing from rpm source to avoid version skew
issues. There are some missing dependancies for the source rpm,
which assume a build on a fairly complete development machine.
If rpmbuild -bb fails, consider building on a development desktop.

After training Bayes I now get about 1% false negatives on
200-500 spam emails per day ... which is much better than the
20-50% that was going thru prior versions of SpamAssassin. I've
only seen a few false positives on SpamAssassin over the last
couple years, mostly ebay vendors that use a form and html email.
A script dynamically whitelisting ebay trading partners from the
ebay notices cures that problem if it becomes a problem for you.
I haven't seen it at all using 2.63 with the despam utility to
clean my last six months of archived email.

If you don't get the much email, but still a fair bit of spam,
you might want to tune the RBL weights up a bit until the BAYES
filter starts to kick in.

I've used this email address since 01-Jul-1991 and haven't really
wanted to give in to the spammers like a number of my associates,
and change email addresses every year or two.

Spamassassin has worked pretty good over the years, along with
RBL's. Over the last year the mail going thru spamassassin and
the RBL's has continue to rise, until taking the time to upgrade
to SpamAssassin 2.63 and taking the several days to go thru the
last six months of archived email and train the new BAYES filter.

have fun,
John