I understand the theory of not wanting to incorrectly learn messages. But my question is about that statements practicality in every situation. It makes sense with a diverse user base, say an ISP, but what about a small company with few users that are all competent to bot train and that will keep up with it. Also that has email practices that make it not feasible to just scan their inboxs for ham. Only auto learning the extremes as SA is setup requires users to train large amounts of messages even if they where correctly tagged if keeping SA up to date is desired. Where as if SA has the option to auto learn every message based on how it is tagged, users would only have to retrain the messages in error, which should be significantly less the the previous way. The Idea is to only have to deal with messages that are incorrectly tagged. If one can guarantee that the user base will keep up with the retraining, assuming that is true, how would this effect the overall accuracy. The few messages that are incorrectly learned would be timely relearned correctly. -- View this message in context: http://www.nabble.com/autolearning-tf3334541.html#a9272701 Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
