From: <[EMAIL PROTECTED]>
Kristopher Austin wrote:
RANK RULE NAME COUNT %OFRULES %OFMAIL %OFSPAM
%OFHAM
------------------------------------------------------------
1 HTML_MESSAGE 45870 5.13 27.72 70.37
55.36
Wait... so 27% of all mail is HTML, 70% of spam is HTML, and 55% of ham is HTML?
<<jdow>>
So what's the problem? (He's not running Bayes or it's badly broken, though.)
TOP SPAM RULES FIRED
------------------------------------------------------------
RANK RULE NAME COUNT %OFRULES %OFMAIL %OFSPAM %OFHAM
------------------------------------------------------------
1 BAYES_99 962 4.81 32.97 93.04 0.11
2 RCVD_IN_XBL 574 2.87 19.67 55.51 0.05
3 HTML_MESSAGE 571 2.86 19.57 55.22 7.91
4 URIBL_BLACKB 563 2.82 19.29 54.45 0.05
5 URIBL_JP_SURBL 484 2.42 16.59 46.81 0.00
6 URIBL_SC_SURBL 479 2.40 16.42 46.32 0.00
7 RCVD_IN_BL_SPAMCOP_NET 440 2.20 15.08 42.55 3.34
8 URIBL_OB_SURBL 409 2.05 14.02 39.56 0.00
9 URIBL_WS_SURBL 403 2.02 13.81 38.97 0.00
10 URIBL_SBL 397 1.99 13.61 38.39 0.05
11 URIBL_AB_SURBL 368 1.84 12.61 35.59 0.00
12 JD_TO_EARTHLINK 357 1.79 12.23 34.53 2.71
13 RCVD_IN_SORBS_DUL 270 1.35 9.25 26.11 0.53
14 RCVD_IN_DSBL 253 1.27 8.67 24.47 0.00
15 URIBL_XS_SURBL 241 1.21 8.26 23.31 0.00
16 LW_MULT_RECIP3 237 1.19 8.12 22.92 2.60
17 JD_MY_NAME 231 1.16 7.92 22.34 2.71
18 DNS_FROM_RFC_POST 194 0.97 6.65 18.76 0.11
19 MIME_HTML_ONLY 192 0.96 6.58 18.57 0.80
20 JD_TO_EARTHLINKCOM 189 0.95 6.48 18.28 0.00
------------------------------------------------------------
TOP HAM RULES FIRED
------------------------------------------------------------
RANK RULE NAME COUNT %OFRULES %OFMAIL %OFSPAM %OFHAM
------------------------------------------------------------
1 BAYES_00 1654 20.19 56.68 0.39 87.79
2 JD_LKML_RELAY 787 9.61 26.97 0.77 41.77
3 JD_PATCH_SUBJ 316 3.86 10.83 0.00 16.77
4 RATWR10a_MESSID 287 3.50 9.84 2.71 15.23
5 JD_CHICKENPOX 247 3.02 8.46 11.90 13.11
6 NOT_TO_ME 231 2.82 7.92 16.54 12.26
7 RCVD_BY_IP 183 2.23 6.27 11.70 9.71
8 HTML_MESSAGE 149 1.82 5.11 55.22 7.91
9 UHS_BCW 135 1.65 4.63 0.10 7.17
10 SARE_MSGID_LONG40 120 1.46 4.11 0.19 6.37
11 JD_MANGY_MORTGAGES 118 1.44 4.04 11.61 6.26
12 USER_IN_WHITELIST 111 1.35 3.80 0.00 5.89
13 JD_GENERIC 90 1.10 3.08 0.87 4.78
14 BAYES_50 78 0.95 2.67 0.97 4.14
15 HELO_EQ_LT4_SA 76 0.93 2.60 4.16 4.03
16 BAYES_20 63 0.77 2.16 0.10 3.34
17 RCVD_IN_BL_SPAMCOP_NET 63 0.77 2.16 42.55 3.34
18 FM_MULTI_ODD2 61 0.74 2.09 6.09 3.24
19 WHITELIST_NTDEV 61 0.74 2.09 0.00 3.24
20 JD_MANGY_MORTGAGE 60 0.73 2.06 0.48 3.18
------------------------------------------------------------
==========8<-------------
Note that there are a lot of rules in there which I intentionally score at
the 0.01 point level so I see them explicitly. I use them in meta rules to
create some interesting special cases that are rather effective. (And as for
the HTML - consider that I am on a lot of mailing lists that are basically
text only and have high volume.
Note that the results are a little "skewed" with reality. I live with a
fairly high false positive rate for the LKML nonsense. So that will affect
the false positive rate on some of the BLs, for example.
{^_-}
