Training the Bayesian filter

Training the Bayesian filter

One of the methods SpamAssassin uses to classify email as possible spam is a Bayesian filter. It learns what spam looks like by being given examples of what is and is not spam. When it makes mistakes, it can be retrained on those mistakes to better its performance.

It is important that we train the filter based on as accurate copy of the email as possible. You should therefore resend the message to the spam filter training addresses rather than forward them. (Forwarding changes the characteristics of the mail enough that it's not useful for training purposes.)

Spam not tagged as spam should be resent to spam<AT>spamfilter.com. Mail erroneously tagged as spam should be resent to ham<AT>spamfilter.com. Mail resent to these addresses is automatically scanned and the Bayesian filter's database updated.


This page last updated October 7, 2007.