Am 24.02.2015 um 22:56 schrieb Yves Goergen:
Last but not least, get your Bayes setup running and it will give you the extra edge.I once had Bayes enabled, but since it's an unattended server system, it can only learn from itself. And that had worked really bad in the past. So I disabled it completely last time I set it up. How should Bayes work if nobody gives feedback about the messages from their Thunderbird clients? And I've tried creating rules for those Polish words, but it's different words all the time. I wonder whether they actually mean something. And it's only very few words per messages, many even with corrupt encoding including HTML entities. Again, how could Bayes help here?
starting with 200 spam and 200 ham samples bayes helps *a lot* and running any contentfilter without bayes is pure nonsense
a spamfilter si not about "that is spam because this" - it#s about scoring and bayes is a important part of the scoring
just set it up with a central bayes-db and feed it only with the samples *you* got and anybody would benefit
with sensible RBL scoring, SpamAssassin including bayes and unofficial ClamAV signatures you block around 95% of all junk without invest any second of work after setup and combined with remote-hashing services into the score mix you get up to 98% with nearly zero false positives
what you have is a dumb contentfilter without bayes and useable clamav signatures with no RBL scroing before the contentfilter - that can't work at all
signature.asc
Description: OpenPGP digital signature
