On Thu, Feb 27, 2003 at 10:06:58AM -0800, Christopher Eykamp wrote: > I am wondering why the following message was autolearned. The message > scored heavily negative, yet it was learned as spam (which, in fact, it > was, but an error in my whitelist let it slip through). I would not expect > any message from a whitelisted sender would be learned as spam. > > Note that the tests with mixed capitalization are my own. > > ----------------- > > From: Blair.com <[EMAIL PROTECTED]> > Subject: Shop today at Blair.com - No charge for standard shipping > To: [EMAIL PROTECTED] > Date: Wed, 26 Feb 2003 18:38:44 -0800 > X-Spam-Status: No, hits=-76.3 required=6.0 > tests=BAYES_90,CLICK_BELOW,CoOptNetwork,EXCUSE_11,FREE, > HTML_40_50,HTML_FONT_COLOR_RED,HTML_LINK_CLICK_HERE, > HTML_TABLE_THICK_BORDER,HTML_WEB_BUGS,NO_COST,PercentOff, > SpamToken1,SpamToken2,SpamWordsInURL,SpecialOffer, > TimesNewRoman,USER_IN_WHITELIST > autolearn=spam version=2.50 > X-Spam-Level: > X-Spam-Checker-Version: SpamAssassin 2.50 (1.173-2003-02-20-exp) > > ------------------
This is because whitelist/blacklist scores are ignored when determining how to autolearn. In addition, Bayes scores are also ignored. This means that with only autolearning, and no corrective actions, SA will eventually move Bayes unlearn itself as it acts more like the general SA ruleset. You would then only need to manually train a few messages here and there to keep the Bayes rules in check. IMO, ideally once you train SA, you should be able to turn on autolearn and SA will keep modifying it's scoring based on bayes such that. In other words, I think that SA should autolearn based on the total hits value and not ignore bayes scores. I'm sure that there's reasons to not do this, but I think the benefits outweigh the drawbacks. -Dave ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
