We have developed an extension to Postgres tthat does *automatic* classification. A machine learning algorithm has to be trained first, using already "tagged" data. Afterwards the classifier does it automatically. We have done tests with several collections of documents and the accurracy is excellent, often as good as human.
We have not had the time (or need) to package it as an openACS package, but will hopefully be done soon.
My idea is that it could be an extension to the content repository, or maybe application specific.
I think that Timo's package is something different.