On Tue, Mar 18, 2003 at 09:25:39AM -0800, Ryan Finnie wrote:
> On Tue, 18 Mar 2003, Sam Phillips wrote:
> > I currently use procmail + bogofilter which gets a good percentage of
> > the spam that I receive. The beauty of it is the more spam you get the
> > less you have to look at.
>
> I've been playing around with Bayes, the bayesian classifier included with
> SpamAssassin 2.50, which gets it's same roots from as bogofilter (why they
> didn't just use bogofilter, I don't know),
Don't get me wrong, bogofilter is a great unix tool. But there are
other "bayesian" filters that are a bit more experimental, and have more
promise. I'm thinking spambayes may be cool.
http://spambayes.sourceforge.net/
> and I've learned something. Specifically, it takes a TON of email and
> time to properly train a bayesian filter. I've fed it about 4000
> emails and about 1000 spams, and it's still not performing optimally.
> In a couple cases, SA by itself would have classified the message as
> spam, but Bayes didn't think it was spam and lowered the score just
> enough to make it through.
The more spam you get the less you have to look at. Set up an email
alias like [EMAIL PROTECTED] and forward it to your
regular email account. Then have a procmail rule like
:0:
* [EMAIL PROTECTED]
| bogofilter -S
Hooray!
Also bounce your spam into a spam mailbox. I currently have a 40MB spam
archive that I can use for training/testing "bayesian" filters. I think
I have several hundred megabytes of non-spam to train/test with too.
> I'll keep on working on training it better, though. I consider it a
> challenge to get rid of those 3-4 daily spams that make it through
> spamassassin.
I haven't kept metrics on how much spam I get and how much of it gets
filtered, but looking into my procmail log bogofilter catches 55 message
a day. I've been running it since November-ish. The only false
positives I've gotten are when I try to send myself (or a list I'm on)
spammy email.
--
Sam Phillips <[EMAIL PROTECTED]> http://www.dasbistro.com
Reno Nevada
_______________________________________________
RLUG mailing list
[EMAIL PROTECTED]
http://www.rlug.org/mailman/listinfo/rlug
