Sunday, February 13, 2005, 1:40:18 AM, [EMAIL PROTECTED] wrote: > http://bugzilla.spamassassin.org/show_bug.cgi?id=4135 > Summary: Suggestion for Improved Bayesian Filtering > ReportedBy: [EMAIL PROTECTED]
> 2) By analyzing boundaries between text and HTML, develop heuristics > to discard leading and trailing plain text "red herring" blocks of > text that are added simply to undermine conventional Bayesian > analysis. Tom, Do you find that these "red herring" blocks of text actually cause any problems? I find that their very use of randomized text, or literary text, provides fodder for Bayes because of their significant difference from conversational email, technical email, and newsletter email (none of which bears enough of a relationship with the red herring sections to cause Bayes any confusion here). Bob Menschel
