RE : ISSUE 77 and postcoordination [and ISSUE-40!]

Hi Jakob,

Your comment is a very valid one.
Actually there is a closely related issue, ISSUE-40 ConceptCoordination
Initially it is about pre-coordination, but I do believe that the same representation mechanisms can be used for both problems (at least in an RDF-oriented view)

About the content of your proposal: I find the pattern quite meaningful, but am really not sure that using Collections is optimal. What  I'm afraid of is that in their current uses, Collections are rather interpreted as 'unions' of concepts, while some coordination cases makes me really think of 'intersection'(at least from a boolean query perspective)
But it may be worthwile to investigate this further...

Best,

Antoine

[1] http://www.w3.org/2006/07/SWD/track/issues/40


-------- Message d'origine--------
De: public-esw-thes-request@w3.org de la part de Jakob Voss
Date: lun. 10/03/2008 04:22
À: public-esw-thes@w3.org
Objet : ISSUE 77 and postcoordination
 

Hi!

I must raise another issue related to ISSUE 77 (skos:subject) about
collections of concepts. How do you encode postcoordination? After
dealing with the encoding of classifications and authority files in SKOS
I am working on a paper on encoding social tagging information with
SKOS. So I stumbled upon the skos:subject property and encoding of
subject indexing.

I was somehow suprised to see skos:subject missing in the current
working draft (chapter 11.2, issue 77). To my point of view skos:subject
is one of the pillars of SKOS (together with skos:Concept,
skow:prefLabel and skos:broader/narrower). It might be enough to use
dc:subject but then the SKOS recommendation should clearly state the
semantics it implies with using dc:subject.

In particular I found two related gaps in the current draft. First is
how to encode postcoordination of concepts and second is how to map to
coordinated concepts. Let me give an example:

Given one Concept Scheme with two concepts labeled "holdiay" and "2008":

 x:holiday a skos:Concept; skos:prefLabel "Holiday" .
 x:y2k8 a skos:Concept; skos:prefLabel "2008" .

How do you encode that fact that a resource '#R' was indexed with both
together in a specific context (person, date, etc.)? You somehow have to
connect two statements:

 #R skos:subject x:holiday .
 #R skos:subject x:y2k8 .

Reification might be a solution but reification in RDF is where the real
problems start, so better avoid it. The second use case is how to map a
concept in one vocabulary to a union of two terms in another vocabulary.
The early mapping spec [1] contained the classed AND, OR, and NOT but
these seem to have faded away (?). OR is not a problem as far as I can
see and NOT could be dropped because of complexity, but how do you
encode an AND? Given a second Concept Scheme with a concept labeled
"holiday2008":

 x:h2008 a skos:Concept; skos:prefLabel "2008" .

how do you encode the mapping between x:h2008 and x:holiday together
with x:y2k8 ? 

The solution I found, seems to answer both questions. First you have to
broaden the rdfs:range of skos:subject, skos:exactMatch,
skos:broadMatch, skos:narrowMatch and skos:relatedMatch from
skos:Concept to skos:Concept, skos:Collection and
skos:OrderedCollection. Second specify the semantics:

<A> skos:subject <C> .
<C> a skos:Collection ; skos:member <X> , <Y> .

entails

<A> skos:subject <X> ; skos:subject <Y>

And the same with mapping relations instead of skos:subject and with
skos:OrderedCollection instead of skos:Collection.

What does this mean? You can now

1. Map between a concept and and a set of coordinated concepts:

x:y2k8 skos:exactMatch [ 
  a skos:Collection; 
  skos:member x:holiday ;
  skos:member x:y2k8 
]

2. Coordinate Concepts into a (sorted) collection and index resources
with this coordinated collection.

#R skos:subject [
  a skos:Collection; 
  skos:member x:holiday ;
  skos:member x:y2k8 
]

Why is support of postcoordination needed in SKOS? Because without you
cannot specify the set and order of concepts that was used to index a
resource! How would you say person <P> indexed resource <R> with
concepts <C1> and <C2> at time <T>? With the proposed enhancement to the
current draft you can say it without additional classes and properties
or even reification:

<R> skos:subject _:x .
_:x a skos:Collection ;
  dc:creator <P> ;
  skos:member <C1> ;
  skos:member <C1> .

Actually the statement says "Ressource <R> is indexed with a set of
concepts <C1> and <C2> that was created at time <T> by person <P>" - but
in practise it's the same.

Greetings,
Jakob

[1] http://www.w3.org/2004/02/skos/mapping/spec/


-- 
Verbundzentrale des GBV (VZG)
Digitale Bibliothek - Jakob Voß
Platz der Goettinger Sieben 1
37073 Goettingen - Germany
+49 (0)551 39-10242
http://www.gbv.de
jakob.voss@gbv.de