THE PRE-CONFIGURED DEFAULT BAYESIAN TOKEN DATABASE

Automatically upon Praetor's installation, spam is filtered using the preconfigured default Bayesian token database. This contains information processed from CMS' study of thousands of spam and non-spam email messages.  With this database, Praetor will instantly filter over 90% of all spam messages. 

TRAINING THE BAYESIAN TOKEN DATABASE

The Bayesian training process starts with Praetor receiving messages and using your pre-existing rules or the supplied default rules to make the first classification of the email.  Is an email message spam (unwanted) or ham (good)?

Once 1000 or more samples are captured, review them to determine if the classifications are correct.  Usually this can be determined from the summary line shown in the Log Viewer, but if more details are needed, the message can be opened.  For any misclassified message (spam accepted as ham or ham quarantined as spam), press the appropriate button to remove the wrong classification.

After the review is complete, CMS recommends waiting until normal mail traffic of the business day has subsided.  Then use the training wizard.  This analyzes each message, breaks it down into tokens and updates the frequency count of each token that is stored in an SQL database.

This training process could take several hours.  Praetor will receive messages, but the CPU utilization will stay high during this training period.  For high volume sites (even after business hours), CMS recommends that the Praetor box be temporarily taken off-line by stopping the SMTP server.

For tips and suggestions on how to train Praetor's Bayesian filter, read the "Bayesian Training Tips" page.