The rules are static, and hand made. The "AI" part is done by the developers to evolve the scores applied to rules, and is done prior to release.
/usr/share/spamassassin is the default location for the standard ruleset. You can add your own custom rules in /etc/mail/spamassassin/local.cf for site-wide application, or in <user's home dir>/.spamasssassin/user_prefs for the user that SpamAssassin is run as. You *can* edit the standard ruleset, but it is generally better to add your own rules, and score over-rides to one of the other locations since they will not be replaced when you upgrade your version of SA. SA rules are in standard perl regexp format. If you have SA installed on your system man Mail::SpamAssassin::Conf. There's also a copy of this info at http://spamassassin.taint.org/doc/Mail_SpamAssassin_Conf.html In general your own rules are of the format body TEST_NAME <some regexp to match> describe TEST_NAME <some message to include in report> score TEST_NAME <some score to apply> instead of body you can also do header, rawbody and full to check the message in different places or different stages of decoding. Here's a simple rule: body SPAMASSASSIN_MENTIONED /\bSpamAssassin/i describe SPAMASSASSIN_MENTIONED mentions SpamAssassin in body score SPAMASSASSIN_MENTIONED -5.0 This will search the decoded message body text and match any word-break (ie space, tab, newline, etc) followed by "spamassassin" in a case insensitive manner. Any message matching the rule gets 5 points knocked off it's score. (I use this rule to help off-list discussions of SA from being tagged as spam) At 04:14 PM 7/17/2002 +0800, Sophia wrote: >I'm considering whether to incoporate spamassassin >into our main mailgateway (which is running amavis as email virus >scan. The good thing is the new amavis has this spamassassin >inclusion option) or not. My main concern is how difficult it is >to configure its so called "rule base" which i think identifies >how probable the mail is spam or not. > >I've read thru roughly the README and FAQ of SA on its web site. It does >not mention anything about how to define the rule base. Is it really that >intelligent (AI involved :-) that one usually doesn't need to customize >the rules at all , and its default rules have very good guess on detecting >spam ? > >Pui (Hongkong) > > > >------------------------------------------------------- >This sf.net email is sponsored by:ThinkGeek >Welcome to geek heaven. >http://thinkgeek.com/sf >_______________________________________________ >Spamassassin-talk mailing list >[EMAIL PROTECTED] >https://lists.sourceforge.net/lists/listinfo/spamassassin-talk ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
