RE: Concept Scheme Versioning [VM]

Hi Andy,

Thanks alot for this, I'm ccing SWBP-WG vocab management (VM) task force, as this is very much a VM issue aswell.

I do think the proposal in the current WD SKOS Core Guide for versioning of concept schemes [1] needs more discussion, and doesn't go far enough.  Hopefully as we collect more requirements we can make it better - all notes you have on versioning requirement would be very useful.

My own thinking re versioning of RDF vocabularies has moved on a little since I drafted that section for the guide.  I wrote a page on the ESW wiki about versioning recently, you might find it interesting, see [2].

In a nutshell, my current position is that it's best to version the *description* of the concept (i.e. the RDF statements about it) but don't 'version' the concept itself.  I.e. each 'version' of a concept scheme is a *named graph*.  You then use *provenance* information to distinguish between differing descriptions of a concept from different scheme versions.

SPARQL query language allows you to specify the provenance of statements (i.e. 'I want statements about concept X from named graph Y'), see [3].  This overcomes the problems outlined below wrt merging of graphs.

You can of course chose to issue distinct URIs for each version of e.g. a DDC category.  However, the dublin core experience (moving from 1.0 to 1.1) was that versioning the URIs for each property (e.g. dc:title) was both confusing and expensive to implement.  Dan Brickley and Tom Baker could say more about that than me, they were directly involved.  DCMI currently publishes historical versions of the DC schemas, but does not issue a new URI whenever a change is made to the description of a property.  I believe these considerations will be similar whether you are managing a small set of properties (e.g. the DC elements) or whether you are managing a larger set of concepts (e.g. the DDC).

That's all I have for the moment.  Cheers,

Al.





[1] http://www.w3.org/TR/2005/WD-swbp-skos-core-guide-20050510/#secschemeversioning

[2] http://esw.w3.org/topic/ConfigurationManagement
[3] http://www.w3.org/TR/rdf-sparql-query/#rdfDataset


> -----Original Message-----
> From: public-esw-thes-request@w3.org
> [mailto:public-esw-thes-request@w3.org]On Behalf Of Houghton,Andrew
> Sent: 15 July 2005 16:06
> To: public-esw-thes@w3.org
> Subject: Concept Scheme Versioning
> 
> 
> 
> I'm having a problem with the SKOS Core Guide recommendation 
> for Concept Scheme Versioning [1].  The Guide recommends 
> (maybe that's a little strong since this section is under the 
> Open Issues heading) that each expression (version) of the 
> controlled vocabulary have a skos:ConceptScheme and each 
> concept in the controlled vocabulary use a skos:inScheme 
> property to relate which skos:ConceptScheme it belongs to.  
> The URI for each skos:ConceptScheme must be unique and the 
> example indicates that each skos:Concept URI must be the same 
> across expressions of the controlled vocabulary.
> 
> Where I'm having a problem, when I try to apply SKOS to 
> controlled vocabularies, is in multiple areas: RDF and URI's. 
>  I'm not having a problem with the skos:ConceptScheme part, 
> but the use of skos:inScheme for a skos:Concept and the 
> concept's URI.  Each term in a controlled vocabulary is 
> defined by its scope notes, hierarchal relationships.  Lets 
> say that I have a thesauri that is issued yearly for the 
> years 2003, 2004, and 2005.  As part of my SKOS conversion, I 
> decide to put each version in a separate document instance. 
> In each of the document instances, e.g., 2003.skos, 
> 2004.skos, 2005.skos, I create a skos:ConceptScheme element 
> with a unique URI across all document instances.  This is 
> straight forward.
> 
> However, when it comes to defining the concepts in each of 
> the document instances, this is where I start to have some 
> concerns.  The Guide indicates that a common URI be used for 
> concepts across all document instances.  My understanding of 
> RDF is that when each of my document instances is placed in 
> an RDF triple store, RDF will merge the skos:Concept elements 
> that have the same URI.  My dilemma is using a non-versioned 
> URI for the skos:Concept elements will merge the concepts.  
> This may not be what you want.
> 
> (1) Lets say that in the 2003 version of the thesauri you 
> define a concept with a certain scope, but latter in the 2005 
> version you restrict the scope.  When RDF merges the concepts 
> from the 2003 and 2005 document instances, you will have 
> conflicting scope notes.  What happens when someone in 2003 
> assigns this concept to a resource and someone else processes 
> it in 2005 against the 2005 version?  Since the scope 
> changed, but the URI is the same in the 2003 and 2005 
> versions how does the person processing it against the 2005 
> version know whether the resource is in scope or out of scope?
> 
> (2) Lets say that in the 2005 version of the thesauri you 
> define a concept with a certain scope, but latter in the 2005 
> version the concept is removed.  When RDF merges the concepts 
> from the 2003 and 2005 document instances, you *may* have 
> conflicting properties depending upon how you indicated the 
> removal of the concept in the 2005 version.
> 
> (3) Lets say that in the 2003 version of the thesauri you 
> define a concept with a certain scope, but latter in the 2005 
> version you change the hierarchal relationships.  When RDF 
> merges the concepts from the 2003 and 2005 document 
> instances, you have a mixture of scope notes and relationship 
> properties, e.g., NT, BT, RT, etc.  How do you know what the 
> true definition of the concept is?
> 
> It seems to me that you really want each concept, in each 
> expression (version) of the thesauri, to have a version 
> specific URI.  Granted there will be common concepts between 
> expressions of a thesauri that are defined with the same 
> scope and relationships, but there should be some mechanism, 
> possible OWL, to indicate which ones are the same or 
> different and exactly how they are different.
> 
> I would appreciate peoples comments on whether I have 
> misunderstood something and whether my concerns are valid or 
> not and why not.
> 
> 
> Thanks, Andy.
> 
> 
[1] 
http://www.w3.org/TR/2005/WD-swbp-skos-core-guide-20050510/#secschemeversioning

Received on Monday, 25 July 2005 14:58:53 UTC