[unisog] Barracuda effectiveness (vs Puremessage)

rick pim rick at post.queensu.ca
Mon Jul 16 19:17:56 GMT 2007


 > We have the Barracuda set to do the things their SE recommends. And  
 > now they are recommending manual tuning involving keyword lists,  
 > extra blacklists (beyond their own and xbl/sbl from spamhaus), bayes,  
 > etc.

manually tag enough messages as "spam" and "not spam" to enable their
bayesian filtering and they get significantly better. (i think it's
200 of each). last time i did it it took a couple of days since it was
just a couple of us doing it and we were just using our own mail to do
the tagging. (actual 'face time' was fairly modest.) we've got some
users trained to send us new examples that (a) are spam and (b) aren't
flagged by the bayesian filters.

their 'intent analysis' tends to be -- in my experience -- overly
aggressive.

rp




More information about the unisog mailing list