Re: Learning curves and such (was Re: pgFoundry)

On Fri, May 20, 2005 at 11:59:00PM +0900, Michael Glaesemann wrote:

> >Right, if you classify the information coming in, you can set controls
> >over who sees it. What we don't do now is any kind of classification.
>
> This may be a bit off-the-wall, but I recall Joel Spolsky recently
> writing about using Bayesian filtering to classify mail into groups
> other than spam/ham. I wonder if there's any use for something like
> that in this case.
>
> http://www.joelonsoftware.com/articles/FogBugzII.html

No, definitely not. Pseudo-bayesian classification as used by the more
optimistic spam-filtering folks is pretty crappy at the best of times,
and it's really unusuable for more than 3-4 categories.

There are natural language analysis techniques that'll do this sort of
thing, but they're in the realms of research projects, not canned
tools.

Cheers,
Steve