bayes learning

[Ronan McGlue <r....@qub.ac.uk>]

am i right in thinking that messages which are spam but have attracted 
low baysian scores should be sa-learn't appropriately. and messages 
which arent spam but have attracted (this is where i start to get a 
little confused) high baysian scores or just high SA scores (ie more 
than the spam threshold)

I have a setup at the minute in the university where i have 2 spamd 
servers with 2 imap folders on one of them. this is used by a select 
group(the email technical group) to train bayes by copying *not 
forwarding* appropriate mails to it. questions pertinent to this setup 
follow...

how should the bayes be taught?? should it be fed all spam mails i can 
get...? what about normal personal mails should it be fed those also... 
or is it only in the cases when they have been learned incorrectly 
previously....


is it ok to learn a message that has already been scanned by 
spamassassin... ie with the full SA headers etc....
also what about learning spam gotten through a mailing list.. ie 
recently i got a lot of the german spam through a couple of lists i sub 
to... should i learn them as ham or just leave them be...???

all these questions, are they frequently asked??

thanks

ronan


-- 
========

Regards

Ronan McGlue
Info. Services
QUB