Loading ...
Sorry, an error occurred while loading the content.
 

FW: Statements about statements -- Help needed!

Expand Messages
  • Misha Wolf
    [forwarded] ... From: Russell Duhon [mailto:fugu13@mac.com] Sent: 22 December 2005 18:55 To: Misha Wolf Cc: Semantic Web; newsml-2@yahoogroups.com Subject: Re:
    Message 1 of 3 , Dec 23, 2005
      [forwarded]

      -----Original Message-----
      From: Russell Duhon [mailto:fugu13@...]
      Sent: 22 December 2005 18:55
      To: Misha Wolf
      Cc: Semantic Web; newsml-2@yahoogroups.com
      Subject: Re: Statements about statements -- Help needed!

      The reification approach would likely be a good bet.

      I've worked up a quick example, working from content in your model
      draft ( http://www.iptc.org/NAR/1.0/specification/DRAFT-NAR_1.0-spec-
      Model_12.pdf ) and an example in the RDF Primer.

      Here's how it might look (fragmentary, of course):

      random:triple1 rdf:type rdf:Statement .
      random:triple1 rdf:subject random:item42 .
      random:triple1 rdf:predicate newsml:provider .
      random:triple1 rdf:object "iptc.org"^^xsd:string .
      random:triple1 newsml:confidence "0.4"^^xsd:float .

      You could create subtypes of Statement that took various properties
      (such as newsml:confidence) if some statements had one set of
      properties and some statements another.

      I haven't worked too much with reification in RDF, but this seems a
      fairly straightforward application of the notion, and its a fairly
      elegant way to handle this sort of thing. I could see some
      interesting applications that filtered your RDF based on properties
      of Statements . . . actually, that's a really neat idea. I'm going to
      have to think about how that might be applied to give organizations
      fine-grained control over the statements they make available through
      (perhaps) SPARQL endpoints.

      Russell





      On Dec 22, 2005, at 12:27 PM, Misha Wolf wrote:

      >
      > In working on the XSLT transform to convert NewsML 2 [1] metadata
      > to RDF, we're having problems with statements about statements. We
      > would appreciate advice, especially if it is couched in terms we can
      > understand.
      >
      > A NewsML metadata element can take as a value one of the following:
      >
      > - a concept, represented by a CURIE [2]
      >
      > - a string
      >
      > - a structured element, eg one describing a person or an
      > organisation
      >
      > - an rdf:bag containing elements whose values are taken from the
      > above list
      >
      > The metadata element typicaly relates the value to a news item, eg:
      >
      > - a creator|contributor|subject|genre|etc of news item X is Y
      >
      > NewsML 2 allows one to make a number of additional statements. Some
      > of these are about the value itself; others are about the assignment
      > of the value.
      >
      > The first category hasn't caused us problems. Examples of such
      > statements are:
      >
      > - concept X is of type Y
      >
      > - concept X is the same concept as concept Y
      >
      > - concept X is the child of concept Y
      >
      > - the title (aka label) of concept X is the string "Y"
      >
      > It is the second category that is causing us problems. It includes
      > statements such as:
      >
      > - this statement was created by A
      >
      > - this statement was created on date/time B
      >
      > - this statement was made with confidence level C
      >
      > - the relevance level of this statement to the news item is D
      >
      > - this statement is based on information derived by method E
      >
      > We've studied the RDF Primer [3], especially section 4.3 RDF
      > Reification [4] and the description given there highlights a number
      > of problems. For example, there does not appear to be any way,
      > within an RDF graph, to identify a specific instance of a triple.
      > So it seems that we couldn't express:
      >
      > - Fred said on 22 Dec 2005, with a confidence of 40%, that
      > a subject of this news item is concept X.
      >
      > The section seems to offer two semi-viable options:
      >
      > - Use rdf:ID to identify a statement and then make statements about
      > the identified statement.
      >
      > - Use some other, application-specific, way to identify statements
      > but then don't expect anyone else to understand it.
      >
      > We're not clear how either of these would be represented in the
      > various triples notations and we're not sure which path to follow.
      >
      > Any advice would be gratefully received, especially if reading it
      > does not require the consumption of large quantities of headache
      > remedy.
      >
      > [1] http://www.iptc.org/dev/
      > [2] http://www.w3.org/2001/sw/BestPractices/HTML/2006-01-15-rdfa-
      > primer
      > [3] http://www.w3.org/TR/2004/REC-rdf-primer-20040210/
      > [4] http://www.w3.org/TR/2004/REC-rdf-primer-20040210/#reification
      >
      > Misha Wolf
      > News Standards Manager, Reuters, www.reuters.com
      > Chair, News Metadata Framework WG, IPTC, www.iptc.org/dev
      > Vice-Chair, NewsML 2 Architecture WP, IPTC, www.iptc.org/dev
      >
      >
      > To find out more about Reuters visit www.about.reuters.com
      >
      > Any views expressed in this message are those of the individual
      > sender, except where the sender specifically states them to be the
      > views of Reuters Ltd.
      >
      >



      To find out more about Reuters visit www.about.reuters.com

      Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Reuters Ltd.
    • Misha Wolf
      [forwarded] ________________________________ From: Russell Duhon [mailto:fugu13@mac.com] Sent: 22 December 2005 20:20 To: Charles McCathieNevile Cc: Misha
      Message 2 of 3 , Dec 23, 2005

        [forwarded]


        From: Russell Duhon [mailto:fugu13@...]
        Sent: 22 December 2005 20:20
        To: Charles McCathieNevile
        Cc: Misha Wolf; Semantic Web; newsml-2@yahoogroups.com
        Subject: Re: Statements about statements -- Help needed!

        Interesting; while I certainly see problems with writing it all out by hand, I'd hope nobody's writing large amounts of RDF out by hand, and generating the triples is very straightforward. Also, talking about a set of statements is an application-level issue. At the RDF level you're just throwing more triples into the pot. Could you elaborate on the issues you had?

        Also, looking over your examples, things seem a bit odd. First, rdf:nodeID is file-internal, and can't be used across different files. And you don't use rdf:resouce to specify an instance with an rdf:nodeID as an object, you use rdf:nodeID again. Additionally, your approach does not seem to be tackling the same need -- the "results" of the evaluation/report described by the EARL statement are very closely tied to the metadata about the report -- they're really all parts/properties of the same "thing", an evaluation or report.

        With NewsML, however, it will likely be important to extract just the statements, and work with those, or to have more than one set of metadata about the same statements. The statement metadata isn't really about the same thing as the item metadata. By using the Statement approach, the plain triples are part of the RDF to be manipulated, and I can ask (sorry for the pidgin) things like: ?x newsml:subject ho:trains instead of having to ask ?y newsml:concerning ?x, ?y newsml:concerningSubject ho:trains . Particularly, the first captures the meaning far more accurately, which I feel is Very Important to good RDF.

        Though actually, I've just thought of a weakness of the statement approach, where you might have more than one person making different confidence evaluations of the same statement, or similar.

        Therefore I advocate the following basic approach:


        random:triple1 rdf:type rdf:Statement .
        random:triple1 rdf:subject random:item42 .
        random:triple1 rdf:predicate newsml:provider .
        random:triple1 rdf:object "iptc.org"^^xsd:string .
        random:triple1 newsml:meta random:metadata5 .
        random:metadata5 rdf:type newsml:StatementMetadata .
        random:metadata5 newsml:evaluator random:person1.
        random:metadata5 newsml:confidence "0.5"^^xsd:float .


        After all, nearly anything can be solved with an additional degree of indirection ;-) .

        Russell


        On Dec 22, 2005, at 2:35 PM, Charles McCathieNevile wrote:


        On Thu, 22 Dec 2005 19:54:48 +0100, Russell Duhon <fugu13@...> wrote:

        The reification approach would likely be a good bet.

         I've worked up a quick example, working from content in your model draft ( http://www.iptc.org/
        NAR/1.0/specification/DRAFT-NAR_1.0-spec-Model_12.pdf ) and an example in the RDF Primer.

         Here's how it might look (fragmentary, of course):
         random:triple1 rdf:type rdf:Statement .
        random:triple1 rdf:subject random:item42 .
        random:triple1 rdf:predicate newsml:provider .
        random:triple1 rdf:object "iptc.org"^^xsd:string .
        random:triple1 newsml:confidence "0.4"^^xsd:float .

        Earlier drafts of EARL actually took this approach. While it is functionally pretty similar, we moved away from it because it seemed just a tiny bit more complex to explain, and it really got quite complicated if you wanted to talk about a set of statements.

        Cheers

        Chaals

        -- 
        Charles McCathieNevile                     chaals@...
          hablo español  -  je parle français  -  jeg lærer norsk
             Peek into the kitchen: http://snapshot.opera.com/




        To find out more about Reuters visit www.about.reuters.com

        Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Reuters Ltd.
      Your message has been successfully submitted and would be delivered to recipients shortly.