[Spambayes] spam on the spambayes list
Meyer, Tony
T.A.Meyer at massey.ac.nz
Mon Jul 14 14:07:56 EDT 2003
> I'm not sure if you all saw the posting, with subject
> starting "I bet you spent your while life"...
> My SpamBayes did not catch it as spam,
Nor did the python.org spambayes system (if I read the headers rightly,
it scored 60.1%).
> and I'm in somewhat of
> a quandary; if I mark it as spam, will that not increase the
> probability that any spambayes list traffic might be so marked?
Yes it will increase the probability, but it's unlikely that it would be
significant enough to make any difference in the end score. For
example, I trained it as spam, and these are the spambayes-related
tokens that were changed (new data shown):
'subject:Spambayes' 0.0045347 83 1
'spambayes' 0.00586835 64 1
'email name:spambayes' 0.00780016 48 1
'url:spambayes' 0.00780016 48 1
'to:addr:spambayes' 0.00911255 41 1
'email addr:python.org' 0.0109189 76 4
'sender:addr:spambayes-bounces+t.a.meyer=massey.ac.nz' 0.0676502
5 1
'sender:addr:python.org' 0.0931363 80 52
'to:addr:python.org' 0.112153 65 52
These are all still strong ham clues - given the ham counts for these
clues, I'd have to train at least 50 or so more like this to just get a
neutral score for these tokens. Your scores may differ, of course,
depending on what you have fed spambayes.
It's a reasonable rule to just train on everything that is
misclassified.
=Tony Meyer
More information about the Spambayes
mailing list