Mmm... that's interesting because our Bayesian analysis filter is
working wonderfully. Sure, the odd one or two messages get through
every now and then but training seems to work well. Don't forget you
should train it with good messages as well as bad.
There *are* patterns which the filter can detect. Every spam has a
payload of some sort and more often than not the url of the payload
remains fairly constant no matter how they try to disguise the lure text
with images etc.
On the other hand my ISP has just implemented Greylisting
(http://projects.puremagic.com/greylisting/whitepaper.html) and although
it is probably just a short term measure (maybe lasting a year or two
until the spammers catch up) it has been dramatic in cutting down spam
and has the benefit of needing no constant training.
not much hope on these with Bayesian Analysis IMO. There aren't any
patterns which such a filter can detect.
Would love to hear some ideas on how to detect them reliably, too.
I'm running james 2.3 with the BayesianAnalysis filter. It is an
awesome
combination except there are two fairly specific spam messages that
seem to
get through no matter how many times I feed them to the engine,
Regards,
- David Legg
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
