Re: Definition of

I think it might be possible to reach agreement on some basic first steps
and terminology? As some starting suggestions picking up on what people have
said:

1. Leonard's basic definition of 'facet' as mutually exclusive
groupings with internal hierarchies of concepts is something widely agreed
that could be essentially captured in the core SKOS.
> "Just a word to clarify this. My understanding of this is that
> facets are mutually exclusive (a concept cannot belong to more than one)
and that
> the concepts within a facet can be arranged into arrays, each array
> having a specified "characteristic of division" (e.g. "vehicles _by_
> number of wheels")."

I think this is what is currently intended in the SKOS scheme? - we were
essentially querying details of the RDF encoding in our initial feedback.

2. We can make the distinction between 'facets' in this sense and 'arrays'
by characteristics of division as above (avoiding use of the term 'facet'
for arrays, although some people have used it this way). Arrays by
characteristics of division could be an option to include in SKOS Core but I
think they are less crucial in a core scheme.

3. I take Stella's point about sometimes mutual exclusivity breaking down in
real world knowledge systems. My reading was that this is a separate issue
from the 13 CRG categories that could (occasionally) occur within any mutual
exclusive knowledge structure that is poly-hierarchical. It would be
minimised by a good facet organisation but might occur in contexts like
Stella's examples. I don't see any way round this (and I don't think it
should hold us back), except that we might have another optional property
for a concept to express the notion that, although occuring in a faceted
system, this was an exception as regards exclusivity?

4. We can make a distinction between
a) the general notion of mutually exclusive (hierarchical) facet structures
(as in 1 above) - without detailing any specific top level names

b) a possible set of useful, commonly occuring 'fundamental
categories/facets' - eg
the 13 CRG categories Aida described - (thing - kind - part - property -
material - process - operation - patient - product - by-product - agent -
space - time) - or Ranganathan's smaller PMEST set. These can be mapped to
particular facets in particular thesauri and have possibilities for acting
as a super-structure bridging between different KOS. They also allow us to
express some idea of role within a string as Leonard says.

I think that it might be easier to start in the basic SKOS with just the
notion (a) above and leave (b) for higher level frameworks.

The (later) higher level framework could allow the definition of how concept
members of facets can be combined togther in strings in a particular KOS or
particular KOS application (the synthesis rules). Also it might allow the
specification of a particular set of fundamental catgeories/facets.

Eg perhaps the SKOS RDF could define that there is something called a
facet and that a corresponding OWL or RDFS would capture the mutual
exclusivity - this may have been (roughly) what Dan was intending?
> ".. facets defined as RDF properties annotated in OWL,
> versus in a more traditional library-style thesaurus? And then mappings
> specified, in prose or RDF...?"

5. I do think its important to have the basic idea of mutually exclusive
facets in the core KOS representation. It would allow different applications
to make use of the notion in different ways, drawing on a common
representation.
However I'm not sure that making it a sub-property of the standard 'broader'
property is quite right - is another type of property more appropriate? The
current scheme intends I think that the facet be represented  by the concept
at the top of the hierarchy? The issue might boil down to whether it is
better to introduce another entity, 'facet', in addition to 'concept' - or
just to make do with 'concepts' only? -- perhaps Al can correct me if I have
misunderstood the intentions of the current scheme. Our thought was that
introducing a second entity for facet might be more flexible for future
developments.

This may be a fairly fine detail in the general SKOS scheme which does seem
to be getting close to the basic notion of facet if  the mutual exclusivity
can be covered in Owl.

Doug

Received on Friday, 27 February 2004 07:19:37 UTC