Very cool! I was very impressed when they showed me the automated testing systems they've got running for the SA project. However, the test results do not agree always with my own tests (due to corpus data not the the algorithm). I'm going to try and work with Theo sometime very soon to get my R&D server running the nightly checks because I want to incorporate my corpus of ham/spam as well. I'm also hoping that by doing this when I refine a rule, I can visually see the email that had a FP. The hardest part I'm having right now is using the data to do refinement but the privacy issues are obviously paramount.
Regards, KAM
----- Original Message ----- From: "Jeff Chan" jeffc@surbl.org To: "SURBL Discuss" discuss@lists.surbl.org Sent: Friday, May 12, 2006 11:56 AM Subject: [SURBL-Discuss] Recent SpamAssassin mass check results
FWIW Here are last Saturday's SA mass check results, courtesy of Theo:
http://www.surbl.org/news.html
MSECS SPAM% HAM% S/O RANK SCORE NAME 0 181939 52229 0.777 0.00 0.00 (all messages) 0.00000 77.6959 22.3041 0.777 0.00 0.00 (all messages as %) 22.377 28.8009 0.0000 1.000 1.00 0.00 URIBL_SC_SURBL 26.604 34.2378 0.0134 1.000 1.00 0.00 URIBL_WS_SURBL 24.854 31.9854 0.0115 1.000 1.00 0.00 URIBL_JP_SURBL 12.423 15.9889 0.0000 1.000 0.98 0.00 URIBL_AB_SURBL 23.278 29.9463 0.0479 0.998 0.96 0.00 URIBL_OB_SURBL 0.236 0.3028 0.0038 0.988 0.67 0.00 URIBL_PH_SURBL 15.377 19.7803 0.0383 0.998 0.95 0.00 URIBL_SBL 29.707 38.1606 0.2585 0.993 0.85 0.00 URIBL_BLACK 0.020 0.0264 0.0000 1.000 0.50 0.00 URIBL_RED 0.515 0.4353 0.7946 0.354 0.45 0.00 URIBL_GREY
Of particular relevance are the low false positives of some of the SURBL lists such as SC, AB and PH as shown in the low HAM% numbers. (Note that PH is important to use and score highly in order to detect phishes. It doesn't detect a large percentage of spams, but it likely detects many phishes.) The last three are presumably uribl.com lists.
FPs on OB remain too high IMO, but we're continually working to try to improve both the FN and FP rates.
Jeff C.
Don't harm innocent bystanders.
Discuss mailing list Discuss@lists.surbl.org http://lists.surbl.org/mailman/listinfo/discuss