For the RDF export of HubMed's Tag Storage, I've tried to use Richard Newman's tag ontology [RDF/XML, N3] to describe each 'tagging'. This is mixed in with RSS 1.0 (RDF/XML) to describe each 'tagged item'.
Hopefully I've got everything right - it validates anyway and seems to make sense.
What I could do with now is a way to encode statements, in the form "[subject] blah blah blah [object]" into usable RDF, but I think this kind of natural language processing might just be a bit too complex.