Bret Miller wrote:
I have a bayes question I am hoping someone may be able to
answer for me. Since implementing bayes it has been doing a
very good job except for one thing.
One particular spam email is not getting tagged as spam. My
rules are scoring the email high enough to be tagged as spam,
but it is also hitting the BAYES_00 rule, which is deducting
4.9 point, thus causing the email to not be tagged as spam.
I am very new to bayes so some of my terms may be incorrect.
But it would appear that bayes has "learned" something incorrectly.
I am not sure if something got autolearned as ham, etc. But,
my question is how do I go about finding out exactly what
within bayes is causing this email to be scored as BAYES_00?
And more importantly, how do I "undo" it?
Bayes tokenizes the e-mail, so it's hard to point at exactly what might
make it think it's spam. The best way to combat this is to sa-learn
--spam the message when it comes it. That way, if it was autolearned as
ham, it's reversed. If tokens appeared in several ham messages, then you
might have to repeat this a few times before the scores get reversed
enough that it hits bayes_99 intead.
Bret
I don't know how your box is set up, but if you are using the mbox
format for keeping mail, then you may want to single out this one
message and run:
* spamassassin --test-mode /file/containing/spam_message
and get whatever the scoring is. Then run:
* sa-learn --spam --spam [--mbox?] /file/containing/spam_message
Then re-run:
* spamassassin --test-mode /file/containing/spam_message
You should notice that the score is now being scored by BAYES_99. After
you do that run:
* spamc -c < /file/containing/spam_message
And make sure that you are getting the same score from spamc. You could
have a configuration issue. Not sure if this would help you but I had
the same problem about 2 weeks ago, and this helped me.
-=Aubrey=-
