Clay Davis wrote: > Has anyone thrown this to the SA wolves... I mean group, to get their > opnion? (get ready to duck!) > Clay Disclaimer: I'm just a community member, and really don't care about Guinevere or GWAVA, nor do I know much about either.
Their statements about accuracy make me laugh. Really, it sounds like they're dropping the SA based product to increase sales of their in-house engines. Reading the article, their "spam 2.0" solution sounds like a re-arrangement of how SA works with bayes. Of course the article isn't technical enough to know for sure, but it sounds like they're using a SA style rule-based autolearner to train a bayes system, but when it comes time to score mail they use the bayes only. How this is any kind of radical departure from SA's existing autolearning ability is beyond me. Unless they've found some form of learning categorizer that works better than bayes, I think they'll eventually find there's a good reason SA uses both static rules and bayes. Bayes alone causes lag in adapting to new spam trends.
