info.mailtraq.com > Anti-Spam > Mailtraq's Intelligent Anti-spam Protection Mailtraq's Intelligent Anti-spam Protection Mailtraq Professional includes built-in Bayesian intelligent spam detection, with several advanced features:
- Powerful - uses latest Bayesian techniques to recognise spam never seen before.
- Cost-effective - adaptive system recognises new spammer techniques - no need for subscriptions or constant signature updates.
- Risk-free mail rejection - senders get notification and temporary bypass address.
- Quick set-up - no need to install and plug-in third-party software.
- Easy administration - both central and user-controlled models supported.
- Flexible response - mail that's identified as spam can either be rejected, or moved into a 'spam' folder for periodic checking.
Read more...
Configuration Tips
Mailtraq provides two mechanisms for spam analysis: 'On Receipt' and 'On delivery to Mailbox'.
You can only use 'On Receipt' if you are receiving messages directly by SMTP, but you can use 'On Delivery' in either case.
The Tutorials refer to 'On delivery to Mailbox' and you will find detailed configuration assistance from the 'Help' in the Mailtraq dialog boxes.
The position of the spam score 'Slider' in the dialog box is preset at the optimum point. It is seldom necessary to move it; small movements can have a big effect.
Identifying Messages It is strongly recommended that 'message' option is enabled in both Receipt-mode and Mailbox-mode. This message modifies the Subject line of redirected messages and helps make clear which of Mailtraq's anti-spam engines caused a rejection.
Training
Read how in - Training the System
Be careful with your spam training - if you only train on spam, and never tell it about good emails you can get the system to think all mail is spam!
So use your mailbox to train it on some good messages - or the Administrator can do a bunch at a time in the Console - and keep an eye on the stats page:
Options | Anti-spam | More

The messages trained as spam and non-spam should be roughly the same within about ±10% - if not, train up some more of the one that is low.
In the above example, 99 messages have been flagged as non-spam, and 83 as spam, so some more spam training would increase efficiency.
When selecting messages for training it increases efficiency if you select different types of messages that are spam and non-spam - rather than just picking on, for example, 10 examples from a mailing list.
Subject=[spam:$score] $header(subject)
|