Craig R Hughes wrote: >Olivier Nicole wrote: > >>It would be best to avoid ruining the slowly building good reputation >>of SA (attending Apricot yesterday, SA was cited as the best anti-spam >>product one could choose -- Apricot is a yearly international >>conference in Asia-Pacific). >> > >I agree that baseless discrimination is bad; however the goal here is not to >punish evil country, or the people who live in them. In fact, the score of 3.0 >for ROUND_THE_WORLD means that even if you happen to have one of those TLDs, you >still need to be sending something spammy-looking to trigger the rule. As well >as not be in someone's autowhitelist. If the default rules find you triggering >too often on legitimate mail from those TLDs, then it's easier than simple to >set that rule's score lower, or even to 0.0 disabling it altogether. > >It's great to hear that SA is increasingly being viewed as the #1 anti spam >product. In some part I think that's probably due to both its effectiveness, >and its flexibility. The simple fact is that in the corpus, there are 687 >pieces of spam which triggered ROUND_THE_WORLD, and only 40 pieces of nonspam. >Of the nonspam, only 2/40 triggered additional rules putting them over the >threshold, both of which were postings to bugtraq (daniel, could you check >them?) > >Y 5 /home/daniel/corpus/nonspam/security.bugtraq/7377 >GAPPY_TEXT,ROUND_THE_WORLD >Y 5 /home/daniel/corpus/nonspam/security.bugtraq/6911 >GAPPY_TEXT,ROUND_THE_WORLD > >The other 38 were identified correctly as nonspam. > >Of the spam, 687/687 were correctly identified as spam. > Craig, I'd be curious to see this corpus -- where can I find it? I'd like to know, once and for all, how badly this kills the non-spam. Also, is there a testbed suite for checking the results against an arbitrary corpus?
-- http://www.pricegrabber.com | Dog is my co-pilot. _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk