It is more precise to say that Bayesian filters are best suited to individual mailboxes, and on the opposite scale they are not effective when the message base is random.
Bayesian filters need to be trained, and for that you need a corpus of messages that is spam and another that is ham. The better the training, the better the result, and the reverse is true: garbage in, garbage out. Likewise, you need something or someone to keep feeding the algorithm: what were the false positives and what were the false negatives. This makes Bayes ideal for a single user yet makes it poorly suited to an ISP. If you want to implement Bayes for a corporation, you will do better, because more messages will be "on topic" and more and more we are all receiving similar spam. The catch is in training. You may find that Bayes is not worth using, but that the filters in SpamAssassin are worth keeping. Andrew 8) -----Original Message----- From: Nick Hayer [mailto:[EMAIL PROTECTED] Sent: Wednesday, February 04, 2004 8:05 AM To: [EMAIL PROTECTED] Subject: Re: [Declude.JunkMail] [IMail Forum] Continuous statistical filter updates? <thread originally from imal list> Scott - & others regarding SpamAssassin In your opinion: > Correct. That's why for statistical filtering to be effective, you > need to have very small "groups" that receive similar E-mails. > Ideally, each user will have their own statistical database. If not, > per-domain can sometimes be acceptable. Server-wide statistical > databases fare worse. I have baynesian filtering enabled on Sandy's implimentation of SpamAssassin server wide. Am I just wasting cpu cycles/decreasing SA effectiveness by including this? Thanks -Nick Hayer --- [This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)] --- This E-mail came from the Declude.JunkMail mailing list. To unsubscribe, just send an E-mail to [EMAIL PROTECTED], and type "unsubscribe Declude.JunkMail". The archives can be found at http://www.mail-archive.com. --- [This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)] --- This E-mail came from the Declude.JunkMail mailing list. To unsubscribe, just send an E-mail to [EMAIL PROTECTED], and type "unsubscribe Declude.JunkMail". The archives can be found at http://www.mail-archive.com.
