MailScanner implementation

MW Mike Weiner (5028) MWeiner at AG.COM
Thu May 29 00:53:26 IST 2003

> I think this is a mistake. SA is not foolproof. You will get FPs, and
> the score appropriate
> for your site to flag for quarantine or deletion could take weeks to
> determine. Moreover, you'll
> need a spam/ham feedback look from your users to train the bayesian db.
> If you test properly and train , you'll be able to determine a
> high-score for quarantine or deletion without much risk of FPs,
> but there will be a grey area best left to MUA filters. Frankly, it is
> also good PR for you for your
> users to see emails tagged as spam. Your users will get involved, and
> frankly in my experience
> I got great feedback.

Excellent feedback to me, thank you, and frankly i DO agree with you. I am
not, as an already swamped sa, a "mail" cop to coin a term. And i really
dont want to be the admin responsible for mistakeningly removing the CEO's
critical email he was waiting for from a prospective partner, that happened
to attach or write the email in such a way as SA would trash it. Agreed, i
like the tagging, and i started the implementation that way privately, but
opening it up now to more users, i now have 2 sets; 1 wanting the tagged
email simply deleted before getting to the exchange box (admins and the
like), and the other that i want to send the tagged email to (normal users).
Guess i will have to see how to integrate procmail or similar to
post-process the MailScanner output before sending to their final

> The bayesian filters will help, but they need to be continually
> trained. Some of that is automated by MS, but what is spam is a
> semantic question, which only humans are really good at answering at
> this stage in AI development.

Here again i couldnt agree more, and am spending some time looking at
getting a better handle on the bayesian filters.

THANKS for all your input, it was greatly appreciated.

Michael Weiner

More information about the MailScanner mailing list