[Spambayes] Why Improve Bayesian

Tim Peters tim.one at comcast.net
Sun Feb 8 19:52:25 EST 2004


[dont bother]
> Hey heres is quick question for you guys,
> I always wonder whats the need for inventing new
> techniques when most of the bayesian filters are
> giving accuracy >99.7%?

Well, why not?  For some people it's an interesting problem.

For personal classifiers it probably doesn't matter, but for people with
truly high-volume email loads, an error rate of 0.3% is still massive; e.g.,
late last year AOL reported blocking 2 billion spam per day, and MSN 2.4
billion spam per day.  If 0.3% of them leak, that would be about 8 million
uncaught spam per day for MSN.




More information about the Spambayes mailing list