[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Recent spam increase

On 27-okt-2006, at 12:03, George Borisov wrote:

Peter Teunissen wrote:

If you're looking for a way to get rid of picture spam, try the SARE
rules (http://www.rulesemporium.com/) for spamassassin. I use these
rulesets and get very high scores on the picture spam I get. Simply add these rules to your setup using sa-update & the openprotect channel (see http://saupdates.openprotect.com/). It doesn't take the cpu cost of the
fuzzyocr plugin.

Unfortunately they are not quite good enough to get rid the new
image spam. :-(

The SARE rules in combination with a tuned Bayes database get
most of these, but a fair few still slip through.

Yep, I'm so used to bayes, I simply assumed OP would be using it. Silly of me.

OTH, I just looked at some of my recent image spam and when I substract the bayes score from the final score, most of it would still be tagged as spam (I use a threshold of 5). I'm also getting fluctuating bayes scores; to prevent bayes poisoning, I even feed them to sa-learn with --forget.

I can imagine however that if you're an isp like service, you'd leave out bayes and have a fairly high threshold to prevent false positives, you'd have to increase the scores on the image spam related SARE rules. they don't seem to hit ham so I'd think that would be safe.


Reply to: