W3C home > Mailing lists > Public > public-lod@w3.org > January 2009

Re: Granular dereferencing ( prop by prop ) using REST + LinkedData; Ideas?

From: Bernhard Schandl <bernhard.schandl@univie.ac.at>
Date: Mon, 5 Jan 2009 13:34:16 +0100
Cc: "Richard Cyganiak" <richard@cyganiak.de>, "Aldo Bucchi" <aldo.bucchi@gmail.com>, "public-lod@w3.org" <public-lod@w3.org>
Message-Id: <E503F54C-6899-40BE-B498-061E06673B73@univie.ac.at>
To: Yves Raimond <yves.raimond@gmail.com>

On Jan 5, 2009, at 12:27 , Yves Raimond wrote:

>> Also what is the point of providing explicit examples instead of  
>> just ASKing
>> the endpoint if it returns useful data?
>
> Well, this is about auto-discovery - this is about finding the
> endpoint that could answer that query, or finding a RDF document
> holding this information.

I think this is more the task of search engines and crawlers. My  
concern is that it may be difficult to define the actual meaning of an  
example. Consider the statement you gave above:

> <http://example.org/persons_nyc.rdf> void:example { :al_pacino
> :birthPlace :New_York }.

How should a client interpret the example? Does this mean the dataset  
holds information about :al_pacino (for which his birth place is an  
example), about the :birthPlace property (which information?), or  
about :New_York? Or does it hold data about things having  
a :birthPlace :New_York? Or does it describe the many :birthPlaces  
of :al_pacino? Maybe it describes a large number of relationships  
between :al_pacino and :New_York, of which :birthPlace is just one.

>> I think it might be sufficient to just publish which vocabularies  
>> are used
>> by a certain endpoint. Even dbpedia uses a restricted set of  
>> vocabularies,
>> so if a client would know in advance which vocabularies are used,  
>> it could
>> decide if the data returned from this endpoint is useful. This  
>> could be even
>> more restricted to publishing "application profiles" of  
>> vocabularies; i.e.,
>> subsets of the vocabularies that are actually used within a dataset.
>
> Hmm. Won't this end-up defining almost one application profile per
> dataset or RDF documents?

Exactly.

> Are there really canonical application profiles?

No. Vocabularies are canonical; an application profile is specific for  
one application, i.e., one data set or one client application.

> For DBTune, for example, it wouldn't be useful to say that
> each dataset use the Music Ontology, FOAF, etc. You need a finer grain
> to state that Jamendo deals with tags, artists, and Creative Commons
> records, Musicbrainz with editorial information, the BBC John Peel
> sessions with musical performances, etc.

That is was I meant by saying "subsets of the vocabularies". I imagine  
application profiles not on the level of entire vocabularies, but on  
the element level. For instance, an application profile for your  
example could point to the :birthPlaces property, stating that this  
property is used in this dataset. The creation of such profiles could  
be automated entirely, IMHO.

Best regards, Bernhard
Received on Monday, 5 January 2009 12:34:53 UTC

This archive was generated by hypermail 2.3.1 : Sunday, 31 March 2013 14:24:19 UTC