>In short, you extrapolate your data semantics to an ontology layer
>instead of directly in the data / metadata itself, which is a huge
>advantage. But it does require you to understand ontology work, and
>you're probably more screwed if you don't.
I don't disagree with this, the key piece that is probably missing is what
can be delivered with the current crop of software, and whether there is
enough to justify the effort on a large scale. There has been some
interesting work done on patent retrieval, the IR issues that arise when a
pharmaceutical company does something like patent an amino acid sequence
for example, and semantic web tools like jena seem to have some traction
in this space. Of course, you could still leave the operational state of
your metadata in whatever system currently sustains it, and mix and match
in other applications during the research stage. That would limit the
impact of rouge URIs in namespaces and inconsistent metadata schemes at
least, maybe Gary Larson could even be coaxed out of retirement to do a
"when ontologies go bad" cartoon. The SIGIR folks almost always haul data
out of operational systems to try the variations that eventually feed back
to the source, and maybe one of the challenges is to figure out how to
foster the interesting skunkworks and other experimental systems that give
direction to the path ahead.
art
Received on Wed May 23 2007 - 18:01:45 EDT