Which messages to feed to Bayes?
drew at THEMARSHALLS.CO.UK
Fri Feb 27 10:02:17 GMT 2004
Jan-Peter Koopmann said:
> HI Matt,
>> No.. SA requires a complete email to be trained, and it
>> learns from both the headers and the body at the same time.
> is that really true? Of course it would be perfect to get the original
> message. But a changed header does not automatically mean that learning
> the rest of the message does not make send, does it? The database will
> still learn from the body, that the words are considered to be spammy.
> Exchange 2000 users simply have no way to feed back the original message.
> Even the public folder approach (which we use for about a year now) is not
> perfect since Exchange 2000 and 2003 will always return the e-mail in HTML
> format, even if it was plain text in the first place.
Exchange users can forward original mail, complete with headers if they
forward the original mail as an attachment. The easiest way to do this is
reduce the window size of Outlook and the new message window. Then simply
drag and drop the offending message(s) into the new message window and
send it on. Job done. The attached message retains it's headers and
> Still I see that my so trained bayes database is getting more and more
The only challenge that I have not yet mastered is how to either merge 2
bayes databases (Not import from an old format) or allow two different
users to use the same database. When I changed the file group permissions
bayes learning stopped altogether (So I assume SA like to only have rw
user permissions only on the bayes files). I have a natty Postfix script
that will feed mail sent to a particular address to sa-learn
automatically. the problem is that Postfix is piping the mail to the
script and insists on doing so as any user except it's self (Or root, not
that I really want it doing so due to the security issues). The challenge
is that MS runs as Postfix user, the SA bayes files therefore are owned by
Postfix and only can have user rw permissions. Any ideas any one. My head
In line with our policy, this message has
been scanned for viruses and dangerous
content by MailScanner, and is believed to be clean.
More information about the MailScanner