On 2/28/02 12:02 PM, "Gunter Ohrner" <[EMAIL PROTECTED]> wrote:
> Maybe SA needs to know some additional non-spam tests to compensate for > successful spam tests? The current scores look to me as if the GA tried its > best to avoid false positives which unfortunately contained some > spam-specific content. (Maybe parts of this list's archive.) Maybe it would > be more successful if there were other tests commonly successful with > non-spam (eg. constructs found in mailing lists) which the GA could use to > score down "regular" emails... Once 2.11 is out, I'm going to do some looking at which rules are causing the most false positives and false negatives, and I bet that will lead to some insight about how to make changes to the rules. If, for example, we find that 40% of the false positives all were happening because of one rule, then we can look at the genuine positives for that rule and see how they differ from legit email. C _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
