Best practice on using the Bayesian filter

  1. Home
  2. Knowledge Base
  3. GMS
  4. Best practice on using the Bayesian filter


GMS Anti-Spam now can be integrated with a Bayesian filter on a per user basis, allowing each user to manage their own spam using the statistical analysis of incoming email, and a user maintained spam dictionary.

In most cases, following some simple guidelines can result in huge reductions in junk email of up to or even over 99%, at the same time avoiding most false positives.


Firstly, when the Bayesian filter is activated on the system, it is not active for a user until they logon to GMS Webmail and select the filter from the configuration tree.

Having activated the filter, you then need to "train" it to recognise junk email. This is a very simple process, as the filter has tools for importing email.

For best results, take a sample of equal numbers of "good&quot mail and "bad" mail, and place them into two new folders in GMS Webmail.

Now, using the tools supplied on the Bayesian Filter Tools page, import the "good" mail as Mail, and the "bad" mail as Junk. By using equal numbers of "good" and "bad", you ensure that there is a balance in the dictionary created, so that it is more effective.

Now your Bayesian Filter is protecting you against junk email. You can see the filter in action by checking your quarantine folder for the mail being blocked. If the filter is working correctly, it can be left as is, as there is no need to further train it. However there might be occasions when it gets things wrong – either missing a junk mail so that it gets into your inbox, or otherwise quarantining a "good" email, which it should not have done.

In these cases, you can further train the filter with these messages. For example, if the filter allows a piece of junk mail to hit your Inbox, you can train it, by pressing the "Junk" icon, which is a red envelope. Likewise, if you find "good" mail in your quarantine folder, you can use the "Mail" icon to indicate that this is "good" mail.

Also worth considering is whether you want the filter to act at all on certain emails. For example, if an email address is in one of your address books other than your block list, then you might consider all mail from this address to be "good". If this is the case, you can implement the white list filter. Select the white list filter from the list, and activate it, setting it to "Transfer directly to InBox" if a match is found. Now, select Filters at the root of the configuration tree, and move the white list filter above the Bayesian in the list. Now, any mail in any of your address books will bypass the Bayesian filter, to help avoid false positives.

By combining the filters in GMS Webmail and GMS Anti-Spam, a user can manage their own junk email problem with ease, removing the burden from the administrator.

See Also:

Keywords:Bayesian spam anti spam white list junk filter

Was this article helpful?

Related Articles