W3C home > Mailing lists > Public > semantic-web@w3.org > June 2013

Re: The need for RDF in Linked Data

From: Eric Prud'hommeaux <eric@w3.org>
Date: Tue, 18 Jun 2013 13:44:23 -0400
To: "Peter F. Patel-Schneider" <pfpschneider@gmail.com>
Cc: David Booth <david@dbooth.org>, Phil Archer <phila@w3.org>, semantic-web@w3.org
Message-ID: <20130618174421.GC11963@w3.org>
* Peter F. Patel-Schneider <pfpschneider@gmail.com> [2013-06-17 22:22-0700]
> 
> On 06/17/2013 09:47 PM, David Booth wrote:
> >On 06/17/2013 04:41 PM, Phil Archer wrote:
> >>On 17/06/2013 06:26, David Booth wrote:
> >>[..]
> >>>For example, suppose the client application dereferences a URI and
> >>>obtains a comma-separated-values (CSV) document.  unless the client
> >>>application knew how to interpret that file, it would not be able to
> >>>make meaningful use of that data.
> >>
> >>Subject to W3C Member approval and other bits of process, we hope to
> >>launch a WG to define exactly that in the near future (say Sept/Oct).
> >>i.e. define a metadata format and association mechanisms with CSV so
> >>that you can express row and column headings, data types and, I hope,
> >>basic templating rules for turning string values into URIs but I'm not
> >>sure that's in the charter (I'm thinking something like GREL
> >>http://code.google.com/p/google-refine/wiki/GRELFunctions). That same
> >>data could be used to generate other formats besides RDF (OData etc.)
> >
> >Excellent!  If it allows CSV documents to be standards-based
> >interpreted as RDF then it would be a perfect example of the fact
> >that data does not have to *look* (overtly) like RDF to *be* RDF.
> >
> >David
> >
> >
> Yeah, if the metadata is something like:
> 1/ a column can either be the id for the node of the row (in which
> case it must be an IRI column or a blank node id column) or be
> related to the node by a given property, if there is no id column
> then the nodes are blank nodes with no id;
> 2/ each column is typed, either as an IRI, in which case a prefix
> may be given to turn the string in the column into an IRI, or as a
> blank node id, in which case the string is a blank node id local to
> the table or local to the document depending on a flag, or as a
> typed value, in which case the datatype IRI and optionally language
> tag is given; and
> 3/ each column says whether empty cells produce a value (or IRI or
> blank node id) or not;
> then CSV documents can be thought of as RDF.

If that were written as "CREATE TABLE ...", we could rely on the RDB
Direct Mapping <http://www.w3.org/TR/rdb-direct-mapping/> to say that
the RDF representation of:

  Birthday
    name  date
    Bob   1990-04-27
    Sue   1997-04-20

is:

    <Birthday/name=Bob> <Birthday/name> "Bob" ;
                        <Birthday/date> "1990-04-27"^^xsd:date .
    <Birthday/name=Sue> <Birthday/name> "Sue" ;
                        <Birthday/date> "1997-04-20"^^xsd:date .


> If a CSV document can include multiple CSV tables, then CSV
> documents can even be reasonably efficient carriers of RDF.

I've never seen a proposal for this, but mime/multipart might work
well.


> peter
> 
> PS:  What's left for the WG to do?

declare victory?
-- 
-ericP
Received on Tuesday, 18 June 2013 17:44:52 UTC

This archive was generated by hypermail 2.4.0 : Thursday, 24 March 2022 20:41:38 UTC