To SKOS or NOT to SKOS, part 2

To SKOS or NOT to SKOS, part 1

In the existing KOS we have upper nodes with labels such as "Family, Youth and Houshold".
These are to me rather classification aids than pure concepts.
How can we interpret those upper nodes?

Facets

Can we consider those as facets?
According to the Willpower Info glossary on thesaurus terms, a facet serves as grouping mechanism for concepts of the same inherent category.
These categories are normally chosen, so that facets are mutually exclusive; a concept then cannot occur in more than one facet.
E.g. the AAT thesaurus (Art & Architecture Thesaurus ®) has following facets:

• Objects,
• Materials,
• Activities,
• Agents,
• Styles and Periods,
• Physical Attributes,
• Associated Concepts.

SKOS-wise these facets can be translated into subclasses of skos:Concept.
For our case this would give:


And the concepts being an instance of one of those subclasses.


Two remarks related to this approach:

1. I'm not convinced that "Education" and "Family, Youth and Houshold" are really facets, since I don't think these are mutually exclusive.
2. our ConceptScheme (aggregation) still needs some toplevel concepts; meaning that for every subclass we need to create a topconcept, which resembles the subclass, a topconcept which isn't really a concept, cf. our initial statement.
   In fact we are reconfronted with the intial issue.

Categories

Another approach then.
We could consider the upper nodes as categories or domains, but not being exclusive this time.
These types of use of categories have been mapped in the past onto skos:Collection(s).
Cf. the conversion of the Dutch GCAA thesaurus
With this approach we create for every category a skos:Collection.


And the individual concepts become then skos:members of the respective skos:Collection.


Issues with this approach:

1. What do we do now for the topconcept? Just one node which represents the whole ConceptScheme? Doesn't seem to be a real concept to me.
2. We are using Collection now in a situation where the classification is done only based on one property.
   Normally skos:Collection comes in the picture when we are able to classify on multiple properties.

The normal use of skos/Collection

We explain this based on an example using wines.

Wine (with narrower concepts)

• italian wine
• french wine
• cheap wine
• red wine
• white wine
• expensive wine

The lowest level concepts could be grouped meaningfully on certain properties:

Wine

• wine by region (skos:Collection)
• french wine
• italian wine
• wine by price (skos:Collection)
• cheap wine
• expensive wine
• wine by colour (skos:Collection)
• red wine
• white wine

In our use case following the same logic this would give:

topconcept (cf. issue 1)

• concept by category (skos:Collection)
• education concept
• elementary education
• ....
• family, youth and houshold concept
• - childcare

while what GCAA did is

topconcept (cf. issue 1)

• education concept (skos:Collection)
• elementary education
• ....
• family, youth and houshold concept (skos:Collection)
• childcare

which is defining the Collections based on the different values of one property,


while the previous ones are based on the different properties themselves.


I do not know what to think about it.

Coordinated concepts

Coordination is an activity in which concepts from a KOS are combined, e.g. "Car" and "Leasing" to "Car leasing".

In SKOS however, the Semantic Web Deployment Group decided to postpone the whole issue of coordination.

Conclusion

What my conclusion for the moment is, that SKOS is easily to be used when you follow the traditional thesauri semantics,
but when you are confronted with dirtier classifications and navigation schemes, it never feels right, whatever you try.

Or am I overlooking something?

My next move is to break outside SKOS and start modeling in OWL, using the SKOS properties only when applicable.

Stay tuned.