Bayesian Filtering

Statistics from an Agent Beta user

Agent's junk detection utilizes Bayesian filtering to classify junk messages. Bayesian filters are ideal for personal email because you train the filter to recognize exactly what you consider to be junk.

Agent's adaptive junk filter is constantly learning to classify new words. Agent also implements "forgetting", a technique for expiring words which are no longer relevant. The result is a junk filter that keeps up with the evolving flow of spam messages.

The rules-based filters used by ISPs can't match the precision of Agent's junk filters. Rules work well for the obvious junk. But, rules often generate false positives, valid emails classified as junk.

Junk Training

Junk training is easy in Agent. You simply move messages to the Junk folder by dragging and dropping or pressing Ctrl + J. Agent takes care of the rest. Remove a message from the Junk folder and Agent will reclassify the message as legitimate.

Trusted Contacts

Agent's junk detection is tightly integrated with Agent's address book and email filters. As a result, Agent can automatically recognize email from trusted contacts and avoid false positives by classifying the message as legitimate.

If you try to junk a message from a trusted contact, Agent will ask you for a confirmation. In this example, the user is junking a message from PayPal and Agent recognized PayPal as a trusted domain in the address book.

Since the user gets lots of phishing emails with a forged PayPal domain, they've elected to turn on junk detection for the paypal.com domain.

Trusted contacts and domains are also used for Route by Identity.