On Wed, 2004-05-12 at 00:50, Tom Diehl wrote: > On Wed, 12 May 2004, You wrote: > > I am looking to set Spam Assassin up for site-wide Spam tagging, but it > > seems that the SA-Learn utility builds user specific Bayesian databases. > > Can anyone enlighten me on how I can use Auto-Learn and sa-learn to teach > > my system sitewide? > > There are several ways listed on the spamassassin web page. One way is to > use postfix/amavisd-new/and spamassassin together. spamassassin gets run > as the amavis user and as such stores all of its information in the amavis > home dir. > I used procmail/sendmail to set this up. In the procmailrc file I used spamc with the -u option to specify a spamuser. The database is kept under that users directory. This works very well. > > Also I currently have one Linux Login which recieves mail for several > > e-mail address' specifically one for each list that I am on, is using the > > practice of pointing sa-learn to these list e-mails a bad practice? I > > took it as an easy way to feed sa-learn HAM messages but now I am thinking > > that it may just be negatively skewing messages sent to these address' > > such as this one (fedoraAAATTTb-dub.org). > > you shouls feed the Bayes filter with equal numbers of spam and ham based > on a sample of ALL of the different types of mail you receive, not just > ML messages. On my system at home I use spamassassin in a filter. But I dump mailing list mail directly into selected folders prior to the spamassassin filter. Non-mailing list email is then processed by spamassassin. This tends to speed up the processing of incoming email since I get more mailing list email than other things. I have also found that most mailing lists have little if any spam. (besides a few flame wars but those would be difficult to pre-filter :). -- Scot L. Harris <webid@xxxxxxxxxx>
