From: "Brandon D. Valentine" <brandon@...>
Date: 2003-08-01T07:13:26+09:00
Subject: Re: Spam at ruby-talk

On Fri, Aug 01, 2003 at 06:46:18AM +0900, Daniel Carrera wrote:
> Thanks I'll keep that in mind.  I'm using spamassassin because that's what 
> the sysadmin has installed.  The latest version has a Bayes filter which 
> seems to help a lot.
> 
> There's probably an advantage to using different programs for filtering 
> mail.  Maybe one can pickup some kinds of spam that the other cannot.
> 
> On Fri, Aug 01, 2003 at 04:26:24AM +0900, Seth Kurtzberg wrote:
> > I've recently switched from spamassassin to spambayes with much improved results.
> > 

Bayesian and other probabilistic classifiers require training and would
be a poor choice to employ in an automated remailer.  To implement a
Bayesian filter would require a large corpus of presorted spam and ham
to bootstrap the classifier and willing volunteers to continue training
the classifier over time to respond to changes in spam content.  That
job is only marginally less intensive than just moderating the list to
begin with.  The filter administrator would need to read each message
which came across ruby-talk to figure out whether it should be reported
to the filter as spam.  Heuristic analysis like that of SpamAssassin is
better suited to an automated system.

Brandon D. Valentine
-- 
brandon@dvalentine.com                                 http://www.geekpunk.net
Pseudo-Random Googlism:  texas is required to implement an on