Re: I guess it's a stupid questions.

On Mon, 15 Nov 2004 09:42:37 +0100, Eric Jain <eric.jain@isb-sib.ch> wrote:
> Danny Ayers wrote:
> > I've often had doubts, but haven't yet really encountered any
> > situation for which the lack of RDF contexts/quads has been a killer.
> 
> Here is my use case:
> 
> We have a lot of documents each of which consists of several hundred
> statements. Every document has some metadata such as when it was last
> revised. This information can easily be indicated when such a document
> is stored in a single file, using rdf:about="". 

If the metadata is about the document, then you don't have a problem,
changing rdf:about="" to rdf:about="http://uri-of-the-document" will
preserve the semantics.

The other solution of
> course would be to reify all statements, which is definitely not
> practical (which is not to say that reification isn't useful for making
> assertions about individual statements).

Assuming you want something like this, why isn't it practical? (I'm
sure I've seen discussion somewhere around implementing reification
efficiently, to minimise the overhead).

> The important point is that I can no longer make use of this metadata
> after loading the data into an RDF database (e.g. retrieve a set of
> statements or search only statements that are available under a license
> that allows non-commercial use), unless the database supports some kind
> of context.

Does the source data say that the individual statements are licensed,
or just the document in which they are found? Ok, dodging the question
a little. But it does sound like you're already making super-RDF
inferences about the data:

"'a something b' in documentX"
"documentX modified 2004-11-15" 
=> "'a something b'  modified 2004-11-15"

In which case, what's the harm in expressing the information slightly
differently, such as in one of the patterns in the 'N-ary Relations'
doc [1]?

> I would be quite surprised if I were the only person on this planet with
> this problem...

Yep, sure. But apart from periodic flurries, this list isn't jammed
with comments on the problem, which suggests it isn't the show-stopper
it first seems. I wonder if anyone from the SW Best Practices &D WG
has anything to say on the matter...

> > context can be done in a way that is RDF-friendly and useful without
> > needing quads though - check the good Mr. Beckett's approach in
> > Redland:
> 
> For all I can tell he *is*, in principal, using quads, isn't he?
> 
>    graph.add(triple, identifier)

I really will dodge answering that ;-) 
But it does raise the question - if RDF stores can implement contexts,
does it matter that the RDF model doesn't as such support them? (I
dunno, like I said named graphs sounds appealing).

Cheers,
Danny.


[1] http://www.w3.org/TR/swbp-n-aryRelations/

-- 

http://dannyayers.com

Received on Monday, 15 November 2004 12:01:21 UTC