Bayes Analysis

From Roaring Penguin
Jump to: navigation, search

Bayes analysis is the learning system built into CanIt. There is a brief description of the system in the User Guide in Chapter "Bayesian Filtering", Section "Introduction to Bayesian Filtering".

In terms of user input to train and maintain the system to adapt to the actual spam and non-spam your organization receives there are two areas to be aware of. The first is mail actually trapped in one of the quarantines. Mail accepted/released for delivery will also be voted as non-spam for the purpose of Bayes while those rejected will also be voted as spam. Note that messages rejected from the pending quarantine will not be immediately discarded but are rather placed in the spam quarantine to allow a second chance to reset their disposition. Any messages that are just left in either the pending or spam quarantines are expire out after the set durations and are not voted.

The second area is the voting links that can be placed at the end of incoming message. This is controlled by the setting "S-2500 Add links to messages to train Bayesian analyzer" on the Preferences->Quarantine Settings page of the webui. Upon installation this is set to Inline by default which means the voting links will appear on incoming messages. Clicking on a link will allow you to do the Bayes voting without having to long into the CanIt webui.

One point worth emphasis is that care should be taken in voting messages. If messages are incorrectly marked as spam or non-spam this can "poison" the Bayes database and degrade the effectiveness of the analyzer. You are better to delete or allow a message to expire than voting it incorrectly.

For more specific detail consult either the User Guide or the Administrators Guide documentation.