Stu said:
> Roughly speaking, a metadata store is a collection of
> assertions that relate entities, properties and values. We
> partition those assertions in particular ways for database
> performance reasons, for functional reasons, for
> administrative reasons, for economic reasons.
>
> It seems rather important to be able to aggregate a set of
> assertions in a particular way, even though we may want to
> get query results that span many records.
My view of a "metadata record" is that it is a specific set of
assertions, created as a set. In the context of an application, it _may_
be important to record (and to present to a user) the fact that this set
of assertions, now stored along with n-million other assertions, were
made together as a unit. To do this, I'd need to store along with the
assertions some explicit indication of the fact that the specific
assertions that form my set/record are related (e.g. came from a common
source RDF/XML doc on the Web) - I think in the RDF context this is
described as "context" or "provenance".
I'm really not sure whether this is an "abstract model" consideration,
but it's certainly a consideration for "a DC metadata application".
Naomi said:
> > Please pardon my ignorance about RDF, but couldn't there be a (node)
> > called "record" with repeatable property "contains"?
Mikael said:
> That's essentially how you would express "application profiles" or
> similar in RDF, yes. And that is similar on some abstract level to the
> technique used in the experimental systems I've seen.
Yes, that's pretty close to what we did in our very tentative efforts to
model "application profiles" in the CORES project [1].
But (as I think you were suggesting) this is just grouping properties,
and knowing that a particular set of _properties_ is related as (to use
the CORES terms) a "metadata element set" or an "application profile"
doesn't help me to identify the specific set of _assertions_ made using
those properties, stored in Stu's n-million assertion metadata store,
that make up a "metadata record".
Pete
[1] CORES http://www.cores-eu.net/
|