W3C home > Mailing lists > Public > public-esw-thes@w3.org > February 2005

RE: [PORT] progress update

From: Miles, AJ \(Alistair\) <A.J.Miles@rl.ac.uk>
Date: Fri, 4 Feb 2005 15:37:58 -0000
Message-ID: <F5839D944C66C049BDB45F4C1E3DF89D18DB72@exchange31.fed.cclrc.ac.uk>
To: "Dan Brickley" <danbri@w3.org>
Cc: <public-esw-thes@w3.org>

Hi Dan,

Cheers for the text below, will mine it for goodies.

> Tricky re terminology. People may well ask why we didn't simply use
> RDF's existing structures. My advise is simply to note this 
> as an issue 
> and get the spec out as a Working Draft for proper review. Can always 
> change later.

Could you give me a couple of bullet points on what's contentious about rdf:Bag and rdf:Seq?  Or a link to a discussion of bag and seq?
> The rules stuff is interesting btw. I was wondering if skos:narrower
> etc semantics over collections could be done simply with OWL and 
> transitivity etc., if Collection were a subclass of Concept, so
> collections were concepts (are they?).

Everyone I've heard from so far agrees that skos:Collections should *not* be a sub-class of skos:Concept.  I.e. in the BS8723 language, an 'array of concepts' is *not* a concept in its own right.  So e.g. 'people by age' should *not* be treated as a label of a concept.  And this rules out the simple transitivity mechanism I think.  

> Yes, array seems very software-oriented terminology. Is 
> ordering important?
> The current collections don't look ordered. How is order 
> preserved? You
> say "sometimes the ordering is meaningful" but I don't see 
> any order in
> the graph...

The way it's done currently - 

(1) to represent a 'Collection' where the ordering is *not* meaningful:

[RDF/XML snip]
    <rdfs:label>milk by source animal</rdfs:label>
    <skos:member rdf:resource="http://www.example.com/buffalomilk#concept"/>
    <skos:member rdf:resource="http://www.example.com/cowmilk#concept"/>
    <skos:member rdf:resource="http://www.example.com/goatmilk#concept"/>
    <skos:member rdf:resource="http://www.example.com/sheepmilk#concept"/>

(2) to represent a 'Collection' where the ordering *is* meaningful:

[RDF/XML snip]
    <rdfs:label>people by age</rdfs:label>
    <skos:memberList rdf:parseType="Collection">
      <skos:Concept rdf:about="http://www.example.com/infants#concept"/>
      <skos:Concept rdf:about="http://www.example.com/children#concept"/>
      <skos:Concept rdf:about="http://www.example.com/adults#concept"/>

... with the implication that a collection should be assumed to be *unordered* unless explicitly typed as a skos:OrderedCollection.  (As another issue, is it OK to operate on an assumption like this, given that RDF is open-world?  We could of course add another class 'skos:UnorderedCollection' but I didn't want to weigh SKOS Core down with too much vocab).

Does this still look OK to you?

> How about... just a sentence or so recording this as an open 
> issue, one that is
> orthogonal to the SKOS vocab design but important for 
> implementors, and
> that is related to practical deployment questions. Cite TAG
> http-range-14 issue and WebArch REC, at least.

Sounds good to me.



> s/SKOS family of standards/SKOS approach/   (we're not a standard yet,
> amongst other things).
> s/SKOS Core is a model/The SKOS Core provides a model/
> I think we need a sentence or two on the relationship with RDF here.
> Do you have that elsewhere we could borrow from? The relationship is 
> pretty subtle, given that RDFS/OWL also capture conceptual schemes...
> [[
> SKOS Core is an extension of the RDF model [ref], which is in turn an
> extension of the graph data model [ref]. For an explanation of these
> concepts, please refer to [ref]. The reader of this guide 
> should have at
> least a basic understanding of the RDF model and the graph data model.
> ]]
> This needs a little rewrite. The term 'RDF model' is unfashionable, 
> post-RDFCore. It used to mean, roughtly, 'graph data model', but 
> now evokes Model Theory etc. SKOS isn't a 'semantic extension' to the 
> RDF model theory; this wording could give the impression it is. 
> How about:
> [[
> The SKOS Core is an application of the Resource Description Framework
> (RDF). RDF provides a simple data formalism for talking about things, 
> their inter-relationships, categories ("classes"), and properties. 
> See [RDF Concepts] for an overview of RDF, [RDF Semantics] 
> for its formal mathematical 
> basis, and [RDF Syntax] for details of the RDF/XML document format 
> used to exchange descriptions of things in RDF. It can be useful to
> understand the subtle, layered relationship between SKOS and RDF, 
> particularly when building applications that combine SKOS data with
> other information modeled using RDF.
> In purely technical terms, SKOS is an RDF vocabulary 
> (specified using RDFS/OWL);
> a vocabulary that specialises in describing abstractions it calls 
> 'Concepts' and various of their properties and relationships. 
> In more social terms, SKOS provides an encoding for data 
> structures that 
> are in the <em>Thesaurus</em> tradition. There are two main 
> ways in which 
> thesaurus-like data can be represented using RDF. The 
> (non-SKOS) approach which
> makes the most use of RDF and OWL features is to create a new 
> RDFS/OWL vocabulary that captures the contents of the 
> thesaurus. Since 
> RDF is based on a "classes, instances and properties" modeling style, 
> the translation of a thesaurus into an RDF vocabulary can be a major 
> undertaking, particularly for large and/or informal thesauri, 
> or those  
> whose concept structures don't map clearly into "native RDF". An 
> alternative approach is to use RDF, but to use RDF to 
> describe the thesaurus 
> itself. This is the SKOS approach. Technically, it creates an extra 
> layer of indirection, so that from the RDF point of view we are 
> describing things such as 'The concept of Economic 
> integration', rather 
> than <em>Economic integration<em> itself. @@a more 
> instance-oriented eg
> would be useful here@@        A SKOS representation of a thesaurus 
> maps fairly directly onto the data original structures, and can 
> often be created without expensive re-modeling and analysis.
> As an application of RDF, SKOS concept descriptions share RDF's 
> standard XML representation, and can be mixed, merged and queried 
> with any other RDF data. However, because SKOS introduces its own 
> notion of categorization hierarchies, represented in terms of 
> relationships 
> amongst named SKOS concepts, SKOS does not fully exploit all the 
> representational facilities of RDF, RDFS and OWL. [brief OWL 
> example here?] 
> SKOS is intended to provide both a stable encoding of thesaurus data 
> within the RDF graph formalism, as well as a migration path 
> for exploring 
> the costs and benefits of moving from thesaurus-like to 
> RDF-like modeling 
> formalisms.
> ]]
> OK that's kinda rambling and too long for that bit of the doc, but 
> maybe you can butcher it, or use elsewhere? 
> I think it's important to say something like that up front...
> I see you took out "SKOS Core is not an XML syntax for concept
> schemes.", which makes sense; I guess what I was trying to do above
> was give people some tools for thinking about what SKOS is, as 
> well as what it is not. But it's v hard to do :(
> cheers,
> Dan
Received on Friday, 4 February 2005 15:38:31 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 2 March 2016 13:32:05 UTC