Re: vocabs, metadata set, datasets

On Fri, Jan 21, 2011 at 09:46:07AM +0000, Gordon Dunsire wrote:
> Librarians are used to seeing this record with Z substituted by an appropriate
> literal (usually something akin to the value of the corresponding
> skos:prefLabel):
>  
> :R :p1 "text".
> :R :p2 <:Z skos:prefLabel> "label". // this touches on the data packaging issue
> (btw, is there a syntax to show a triple chain like this?)

One easy way:

    :R :p2 :Z 
    :Z skos:prefLabel "label"

> So there are different points-of-view, and it is probably always possible to
> constrain a "record" to triples with one, and only one, subject. Does RDF let us
> have the cake, and eat it?

One could locally decide that all "records" produced must
be limited to statements about one resource.  In terms of
the DCMI Abstract Model (DCAM) [1], a "record" would not
encompass an entire Description Set, but only one Description.
I invoke DCAM here precisely because its design after 2003 was
a response to the limitations of "flat" (i.e. single-resource)
Dublin Core descriptions -- i.e., by the modeling implications
of needs as simple as putting authors' contact information
into the description of a conference paper.

As I see it, records are a packaging mechanism useful for
managing data and tracking provenance.  A particular RDF graph
can consist of multiple interconnected subjects, and there may
be good reasons to record that multi-entity graph in a record.

Yes, one could impose a rule that the graph recorded in a
record consist entirely of statements with the same subject,
but practical needs for shoehorning additional information
into records could tempt people to come up with workarounds
that are messy from a modeling point of view.

Tom

[1] http://www.dublincore.org/documents/abstract-model/

-- 
Tom Baker <tbaker@tbaker.de>

Received on Friday, 21 January 2011 13:49:24 UTC