> -----Original Message----- > From: Matus UHLAR - fantomas [mailto:uh...@fantomas.sk] > Sent: Wednesday, May 20, 2009 4:36 PM > > > > From: Matus UHLAR - fantomas [mailto:uh...@fantomas.sk] > > > > > > looks to me like collaborative system containg functionalities like > > > those in > > > SA already (dcc,razor.pyzor,blacklists) and bayes > > On 20.05.09 15:26, Giampaolo Tomassoni wrote: > > It looks to me that it goes a bit further by allowing your own SNF > > "training" and, maybe, rules to be globally available. > > What's the difference between SA autolearn combined with manual > training? :)
The main difference seems to be that statistical data about the effectiveness of, say, factory rules are automatically collected among users and centrally used to further increase rules effectiveness. I was about to propose something like that some time ago, but on Bayes tokens. In SA probably it would be difficult to collaboratively apply rule scores based on automatically reported effectiveness, since there are too many versions of too many SA rules around, and too many way to configure (and misconfigure) SA which may impair any attempt of metering a rule effectiveness. Instead, it is years now the basic Bayes engine didn't change. It is also difficult to misconfigure (it works or it doesn't). We may attempt to collaboratively share data on tokens it gathers... Giampaolo
