Re: Bayes DB's
Gray, Richard wrote: Surely that would only happen if there were equal amounts of Spam and ham passing through. Otherwise the token will have a tendency toward whichever the program has seen more of. From: Loren Wilton [mailto:[EMAIL PROTECTED] > Assuming that the same header values appear in both spam and ham, I'd expect that Bayes would conclude the token was useless for classification and ignore it. Loren Is there a reason why the trusted received headers would make good bayes tokens? I can't think of any. I can see the value in the first untrusted received header (which would be YOUR first server that the message hits), since you could tokenize data such as the received 'Date:' (and time of day) along with which MX the message came in on (most mail to a lower preference MX will be spam... but that could cause problems if your preferred MXes go down). Daryl
RE: Bayes DB's
Surely that would only happen if there were equal amounts of Spam and ham passing through. Otherwise the token will have a tendency toward whichever the program has seen more of. From: Loren Wilton [mailto:[EMAIL PROTECTED] Sent: 06 December 2004 10:50To: users@spamassassin.apache.orgSubject: Re: Bayes DB's Assuming that the same header values appear in both spam and ham, I'd expect that Bayes would conclude the token was useless for classification and ignore it. Loren --- This email from dns has been validated by dnsMSS Managed Email Security and is free from all known viruses. For further information contact [EMAIL PROTECTED]
Re: Bayes DB's
Assuming that the same header values appear in both spam and ham, I'd expect that Bayes would conclude the token was useless for classification and ignore it. Loren - Original Message - From: Gray, Richard To: users@spamassassin.apache.org Sent: Monday, December 06, 2004 1:46 AM Subject: Bayes DB's Our mailservers add their name to the received from header of every message. As far as I can see, SA detects this and uses it to create tokens when autolearning. Because our DB is shown more spam than ham, there are tokens in the DBase that identify messages coming from our server as being more likely to be spam than ham. This is quite bad. Is there any way to fix/prevent this happening? Thanks, Richard---This email from dns has been validated by dnsMSS Managed Email Security and is free from all known viruses.For further information contact [EMAIL PROTECTED]
