bayesian spam filtering and IRC

[Andy Smith]

Hello,

I'm familiar with the very basic concept of Bayesian filtering, but
have little practical experience with it.  I was wondering
potentially how useful it could be when applied to IRC, i.e. built
into an IRC server?

Assume the IRC server keeps the last few hundred messages sent by
any given user, and when that user quits from the network normally
their messages are considered to be examples of good IRC behaviour.

If, however, the user should be disconnected by an IRC operator in
some specific manner that indicates that the user was spamming or
engaging in other antisocial activity, all of their output will be
considered bad IRC behaviour.  (I'm really thinking mass ads)

Since Bayesian filtering has been around for a while and I've not
seen this suggested I am guessing that it wouldn't actually work.
If so, could you give me some brief explanation why, or pointers to
more info?

Thanks,
Andy