[Spambayes] message rules

Seth Goodman sethg at goodmanassociates.com
Thu Oct 12 16:34:51 CEST 2006


Andrew Russell wrote on Thursday, September 28, 2006 2:28 AM -0500:

> I am having difficulty with just one aspect of my spambayes setup,
> that being where I have filtering for message rules. It appears that
> any message that meets the requirements of my message rule avoids the
> attention on spambayes.

This is by design, due to limitations of the Outlook programming
interface.  I didn't write the code, but I am aware that Outlook makes
it difficult for add-in applications to get reliable notification of
incoming mail events at a specific time relative to processing Outlook
rules.  Sometimes you get notification at the end of rules processing,
while other times it occurs while the rules are still running.  To get
consistent behavior, SpamBayes has a timer that delays evaluating a
message for a couple of seconds after its arrival, to insure that
Outlook rules finish first.

Though the delay can be annoying, if you have accumulated a large amount
if incoming mail, it actually works in your favor.  Most of the mail you
identify in rules is either from mailing lists or has other criteria
that make it unlikely to be spam in the first place.  There is not much
benefit in training SpamBayes to recognize this mail, as simple Outlook
rules can do it reliably.  By reducing the amount of mail SpamBayes has
to classify, it can do a better job.

For example, most of my incoming traffic is from mailing lists.  Most of
the lists already have spam filters (not this one, since it includes
spam samples).  If my Outlook rules remove all the list messages, the
mail flow left for SpamBayes is personal correspondence.  By not
including these list messages in SpamBayes' training set, there is a
wide variety of language that SpamBayes doesn't have to learn is ham.
It can "concentrate" on learning the language style of my personal
correspondents, and can thus more easily discern spam.

--
Seth Goodman



More information about the SpamBayes mailing list