At 11:54 PM 10/31/02, Jack Coates wrote:
I've been experimenting with Eric Raymond's bayesian spam filter (bogofilter - http://sourceforge.net/projects/bogofilter) since August and have been pleased with it.On Thu, 2002-10-31 at 20:37, Vox wrote: > > This time Jack Coates <[EMAIL PROTECTED]> > becomes daring and writes: > > > I'd also suggest preparing management for a new box -- classic pentiums > > are fine for C apps but real dogs when it comes to Perl. > > There's a C implementation of spamassassin which comes in the > package, which is the recommended one to use when you are using it > in a mailserver. > > Vox >spamc and spamd are C hooks, but the base code is still Perl. spamd basically acts like mod_cgi by loading spamassassin once instead of every time a message comes through.
As I understand SpamAssassin, it uses a set of patterns (rules) generated by people with expert knowledge of rule writing.
With bogofilter you "train" it by giving it a corpus of messages that _you_ have identified as spam or as non-spam. bogofilter builds two databases, i.e. word lists, from these messages. Having these word lists, bogofilter can then look at incoming messages, compare their words to the word lists, and compute a probability that the new message is spam (or non-spam).
I've been using bogofilter to process the email to my domain for the past month and have been very pleased with its ability to discriminate between spam and non-spam.
By the way, bogofilter was written in C (for speed) and it _is_ fast.
David
Want to buy your Pack or Services from MandrakeSoft? Go to http://www.mandrakestore.com
