[Spambayes] Comments and Kudos
avi-j at pacbell.net
Sun Dec 28 20:18:53 EST 2003
>> I am still getting a few messages moved to the Junk Suspects
>> folder (with scores in the 70% range) which I think Spambayes
>> should clearly be identifying as Spam by now on the basis of
>> the messages it's trained on.
>Have you looked at the 'clues' for these messages? It's usually quite
>obvious from looking at them why messages score as they do. Note also that
>it's much better to have a roughly equal number of trained ham and spam.
It's easy to do this, but even if I can see why some spam is getting
relatively low scores (as I said, in the 70s), what can I do to change that?
As for training with a roughly equal number of trained ham and spam, I
unfortunately trained with about 20 times more kosher ham ;) than spam: I
now have a corpus of about 180 pieces of spam that I can use to retrain. At
this point, with 2,073 good and only 97 spam in the database, would it be a
good idea to retrain with an equal number? Or should I just leave things as
Thanks again for your advice and for an excellent program!
More information about the Spambayes