Daniel Quinlan wrote: DQ> >>> score X_NOT_PRESENT -1.920 DQ> >> DQ> >> This one is on my hitlist as well. Didn't work out very well. DQ> DQ> Craig R Hughes <[EMAIL PROTECTED]> writes: DQ> DQ> > But it actually turns out to be great at clawing back false DQ> > positives. I think we should leave it in with the low score. DQ> DQ> Hrmm, it seems responsible for a lot of missed spam, though. Maybe we DQ> could write a rule that uses it and other criteria to detect that sort DQ> of local email. They rarely seem to have any MIME and have relatively DQ> few Received: headers.
[craig@belphegore masses]$ fgrep X_NOT_PRESE analysis 0.002 12 -1.9200 X_NOT_PRESENT 0.076 469 -1.9200 X_NOT_PRESENT So it's in 7.6% of all false negatives in the corpus, and 0.2% of false positives. The ratio of FN:FP should be about 12:1 according the the way the GA weights things, so the actual 35:1 ratio is way, way off. This should certainly be fixed, possibly by just dropping the rule altogether. DQ> I also have some good headers ready to check into HEAD for "clawing DQ> back" functionality. Here they are: DQ> DQ> header X_MAILING_LIST exists:X-Mailing-List DQ> header RESENT_TO exists:Resent-To DQ> header USER_AGENT exists:User-Agent DQ> header X_LOOP exists:X-Loop DQ> DQ> score RESENT_TO -1.000 DQ> score USER_AGENT -1.000 DQ> score X_LOOP -1.000 DQ> score X_MAILING_LIST -1.000 Stick em in head, we'll see how they perform. C _______________________________________________________________ Don't miss the 2002 Sprint PCS Application Developer's Conference August 25-28 in Las Vegas - http://devcon.sprintpcs.com/adp/index.cfm?source=osdntextlink _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
