VOConcepts paper

[Doug Tody]

Some more details on the simple use case I mentioned.

Suppose we have a catalog of spectra, containing spectra from
many types of objects, including stellar spectra.  If the
information is available we would like to characterize the type
of stellar spectra as precisely as we can, as the UCD proposal
makes possible.  Hence we have "stars;class.carbon", "stars.pulsar",
"stars.variable;class.variation.flare", and so forth.  Each object is
classified with such a object-type UCD, or possibly several such UCDs.

When it comes time to query the spectral catalog we may want to refine
the query to the level of an individual UCD, but more generally we will
want to query for a class of objects.  Hence we issue a query with a
parameter such as "targetclass=star".

We don't want to query for a list of UCDs, because there might be hundreds
of different types of object in the overall class "star".  This would be
too awkward a query mechanism, plus it would require too much knowledge
on the part of the client.

Hence, to support query by targetclass, the service needs to support
mapping of common object class names or acronyms (star, galaxy, AGN, QSO,
etc.) to fully-qualified names (UCD-type names).

To do this in a reliable fashion we need to standardize the common names
or acronyms, and define standard mappings to the fully-qualified names.
We don't want to build this vocabulary (which will evolve constantly) into
each service implementation, so it needs to be possible to download these
mappings via the network, and cache the translation table in the service.
They need to be in a standard form which the service code can deal with,
probably something based on XML.  The mechanism should be simple enough
to support arbitrary implementations without requiring using some complex
inference framework.

I think this is more or less the problem we need to solve.   In general we
want to specify object types as precisely as we can at the level of the
object metadata, but the query mechanism needs to be powerful enough to
deal with broader classes of objects.  I doubt if it matters much whether
we are dealing with events or datasets.     - Doug