Spam protection strategies, was:x ot, spamassassin question

[Chuck Hallenbeck <chuckh@ftml.net>]

Greg, John, and all

On Thu, Jan 21, 2010 at 12:56:02PM -0700, Gregory Nowak wrote:
> 
> yes, like I said before, I have my script run every 24 hours on the
> messages my mom left as misidentified spam, or ham, so sa-learn is
> being trained that way. Also, yes, the messages are being segregated
> based on the x-spam-status header using maildrop actually, since I
> personally prefer that to procmail. Thanks again.
> 
> Greg

Congratulations on taming your spamassassin configuration. I just
finished installing a different spam protection here, much easier, and
with a different strategy. It uses vipul's "razor" and integrates
nicely with procmailrc and mutt. Razor is a collaborative spam
protection scheme, and works a lot like "denyhosts" in that false
positives and false negatives are reported to a distributed set of
hosts one by one as they are encountered, and each email the user
checks for spam is tested against the collective data base. It is the
consensus of razor users that distinguishes between ham and spam, and
that consensus is constantly being updated as the FP's and FN's are
reported. Also, each user acquires a "trust level" as his reports are
made, based on the soundness of his judgment in making the report.

Advantages: It's a smart filter out of the box, already trained by
other users.

Disadvantages: It doesn't work too well if an individual user has
ideosyncratic criteria for distinguishing ham/spam, since it stresses
group consensus on that question.

I divert incoming mail that is flagged by razor to a spam folder, and
occasionally scan it for false positives, reporting them when I find
them, by a keypress in mutt. As I examine mail that has not been
flagged and diverted to my spam folder, I occasionally find a false
negative, and simply report that to the network by another mutt
keypress.

I wonder what your thoughts are about such a collaborative data base
for spam protection?  It seems to me that it has some strengths that
"going it alone" with the bayesian model for isolated individuals may
lack.


Chuck



-- 
The Moon is Waxing Crescent (35% of Full)
      Either of these web addresses will take you to my web site:
          www.mhcable.com/~chuckh, or www.hallenbeck.ftml.net
                Audio editor weblog: edway.wordpress.com
               Or jabber 1on1 with me, chuckh1@jabber.org
                                --------
 People in general do not willingly read if they have anything else to
amuse them. -- S. Johnson