Henrik Krohns wrote: > Any system that requires feedback from the users is flaved IMO. I'm sure > our workers have better things to do than train bayes. We just use a global > autolearning database, and in our case it works even better as we don't > communicate in english.
IMHO this is a weak argument: A system that does not adapt to your habbits and relies on regularly updated rules from third parties may be considered as broken, too. Of course, training efforts should be minimized. With DSPAM and some shared profiles, you can get very good results without much work for the employees. Finally: No flame intended, just my opinion. > The main point is that SpamAssassin bayes has very small part in scoring > (atleast in our setup). Probably 90% of spam is catched with DNSBLs and > other rules (gotta love FuzzyOcr!), only few messages are helped with the > extra bayes score. Our hit/kill level is as high as 10, and vast majority > of the spam has score of 20+. Maybe its just me, but SpamAssassin only "works" for me. If I would use SpamAssassin, I would still get ~5-10 spam messages per day. With DSPAM, I only get 1-3 per week (currently, I get ~2500 spam messages per week). The only thing I would like to see implemented in DSPAM is a some kind of OCR. fs ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ AMaViS-user mailing list AMaViS-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/amavis-user AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3 AMaViS-HowTos:http://www.amavis.org/howto/
