Bayasian/Learning Spam Filter
How does it work?
-
Provided the anti-spam filtering is active upon your domain, (see the next question), each message is filtered via a Bayesian process.
If the mail is determined to be spam it is either:
- Rejected for delivery, and saved in your domain quarantine area.
- Delivered to you with a subject prefix added.
- Delivered to you with a special header added.
If the mail isn't determined to be spam, or if the classification cannot be determined it is delivered to you as normal - with one of the following headers added:
X-MS.com-spam: yes The message was spam. X-MS.com-spam: no The message was not spam. X-MS.com-spam: unsure The message type couldn't be determined - training required. How do I enable it?
-
You may enable the spam filtering via the Domain-Management page, linked to upon your domain list.
How do I correct mistakes?
-
If you receive a mail to your account which should have been rejected as spam then you should train it. This will let the system know more about what you consider to be spam.
(An alternative for mail which is sent to users which are invalid is to setup spam traps, or configure your list of valid recipients.)
To train a message as either "spam" or "not-spam" you should simply bounce it to one of two special addresses. For example if you have the domain example.org managed for you then bounce the messages which are spam to:
example.org@spam.mail-scanning.com
To train a message as non-spam instead use the address;
example.org@notspam.mail-scanning.com
Note: You should "bounce" not "forward" the messages to get optimum results.
How do I train lots of mistakes?
If you have built up a large collection of mistakenly classified messages that you'd like to be trained that is possible via the mboxtraining page.
Uploaded mailboxes will be processed within the hour.