On Sat, 2004-05-08 at 10:08 +1000, Angus Lees wrote: > At Fri, 7 May 2004 11:04:55 +1000, Mary Gardiner wrote: > > I train it on all my spam and non-spam, and I train it every week on > > mail received during that week. (With a cronjob, I just need to make > > sure false negatives and positives are moved into an appropriate > > folder.) I don't delete the existing token database ever. > > .. so with all that manual spam/ham classification/archiving, is there > actually any point running an "automatic" spam filter anymore? > > >From what I can see any spam filter that needs training is missing the > point - but I've never actually run any of the Bayesian filters so its > purely ignorant prejudice ;)
I'll bite. The Bayesian filters have the advantage that when they make mistakes you can teach them to do better next time. >From my perspective, I used to use SpamAssassin way back. It was relatively effective, but didn't react to changes in "Spam-Zeitgeist" if you will. Eventually I ended up shuffling a heap of mail around anyway (it started letting spam slip through on a daily basis). Bayesian filters let you "harness the power" of that shuffling to improve the filter. There is a bootstrap period of course, and that's the sort of role that I think rule-based tools like SpamAssassin fill well. The mozilla mail client's interface also offers a nifty approach to bootstrapping, which doesn't require much work from the user at all. James. -- James Gregory <[EMAIL PROTECTED]> -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
