>> I beg to differ with you. On the contrary, there have been a number >>of recent posts to the SA list from admins all of whose spam is now >>identified as BAYES_00 and AWL, because they enabled autolearn and let >>it run on its own while a lot of spam was slipping through. >> >> > >This was my issue and the reason I turned off Bayes and DSPAM, because of >our lack of spending the time manually training it and leaving it on >autolearn, it was causing more harm than good. We have a large user base, >5K+ users in different states/countries, it was not clear cut to implement >a user mechanism where they could manually flag email as spam or ham and >train the DB, additionally we use Notes on the back end. > > > > >> If significant amounts of spam are being missed in the first place, >>they are ipso facto being scored low, so if you aren't *manually* >>training them as spam they are more likely to get learned as ham than >>as spam. >> >> >>
We do not use AWL; this has proven to be dangerous. We use dspam and Bayes in parallel, train Bayes with hamp/spam levels of 0.0 and 5.0 and score both with 3.5 on spam and -1.0 on nonspam. In conjunction with the "postfix mini-greylister" aka reject_unverified_sender and some other sanity rule on the domain name, mx etc. we have very rare cases of spam slipping through with no manual intervention. We are rejecting about 95 % of mail so it's not because we dont't get spam.... But I agree mileages might vary widely. The last argument might be true but in practice I found we catch most things because at least dspam starts to score mails when several similar ones have arrived and with that score bayes will soon start to score too. Yours, Jakob Curdes ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ AMaViS-user mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/amavis-user AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3 AMaViS-HowTos:http://www.amavis.org/howto/
