AW: Nested skos:Collections vs iso-thes:superGroup/subGroup

Hi Osma,

Thanks for bringing that up - I'm wrangling with that too for a fairly long time.

For STW Thesaurus for Economics, we've taken neither skos:Collections nor iso-thes:ConceptGroups, but something else (without however being fully content with it). 

To STW belongs a system of about 500 subject categories, organized in up to 4 hierarchical levels. Different from the descriptors, the subject categories are not allowed for indexing. But apart from this, the category system is very much shaped like a classification (applied to descriptors instead of documents). 

skos:Collection, aimed at display arrays and node labels, didn't seem fitting at the time, and iso-thes did not yet exist in 2008/2009, when the descision about the STW structure in SKOS had to be made. Defining the subject categories as skos:Concepts, connceted by skos:broader/narrower relationships, felt much more appropriate, and kept open a future use of the subject category system as a KOS of its own right - stand-alone or in combination with the STW descriptors. Therefore, the subject categories and the descriptors (as zbwext:Thsys and zbwext:Descriptor) were both made rdfs:subClassOf skos:concept. In itself, this is consistent for both KOS (and works well on a practical level). It also works fine for the mapping of the SDMX subject matter domains or the JEL classification and the STW subject categories, which are currently under development. (BTW, the SKOS mapping properties are rdfs:subPropertyOf skos:semanticRelation, which implies skos:Concepts as domain and range. Without being an ontology expert, I suppose that any such mapping to other KOS would make STW subject categories skos:Concepts, with the consequences described below, even if we would drop the explicit definition.)

The weak spot however is anyway the link between the subject category and its attached descriptors: We use skos:broader/narrower for that too, despite the fact that the relationship is different from inter-descriptor relationships as well as from the inter-class relationships of the subject category system. The meaning of that relationship is more of: "the descriptor belongs to a certain field of knowledge" (as expressed in the subject category). On a practical level, in SPARQL queries, a distinction between the different uses of skos:broader/narrower can be made (see examples in https://github.com/jneubert/sparql-queries/tree/master/stw), but the solution requires knowledge about the custom subclasses of skos:Concept, and therefore is not a general one.

skos:member cannot be used here instead of or in addition to skos:broader/narrower, because it implies (S31) that its domain is a skos:Collection. And according to S37, skos:Collection is disjoint with skos:Concept - which conflicts with our use of skos:Concept for the subject categories.

Unfortunately, this situation did not change with the introduction of iso-thes: If I didn't miss an essential point, iso-thes re-uses skos:member for the linking of iso-thes:ConceptGroups to skos:Concepts (which is consistent with iso-thes:ConceptGroup defined as rdfs:subClassOf skos:Collection). So applying iso-thes for connecting two kinds of concepts would result in the same conflict as described above.

I did some digging for another property without the strong onthological committment of skos:member, but did not find anything which could really fit better.

So while I would prefer iso-thes over plain skos:Collections when dealing with rather flat or "display-oriented" upper-level structures, I have no idea how such upper-level structures could be connected to concepts if they (for one reason or other) are defined as skos:Concepts themselves. Since SKOS and S31/S37 are "cast in stone", and iso-thes, I suppose, similarily so, I currently have no idea how "class" and "descriptor" types of concepts used in the same KOS could be properly be connected.

Sadly, all that doesn't help neither for implementation in Skosmos nor for possible similar situations (which hopefully can be avoided for GACS and YSO). But perhaps somebody comes up with better ideas.

Cheers, Joachim


-----Ursprüngliche Nachricht-----
Von: Osma Suominen [mailto:osma.suominen@helsinki.fi] 
Gesendet: Montag, 18. Januar 2016 09:23
An: public-esw-thes@w3.org
Betreff: Nested skos:Collections vs iso-thes:superGroup/subGroup

Hi all,

The ISO 25964 SKOS extensions define a iso-thes:ConceptGroup class (subclass of skos:Collection) and its hierarchical properties iso-thes:superGroup and its inverse iso-thes:subGroup. On the other hand, SKOS Core also allows nesting skos:Collections using the skos:member property, as its range is defined as the union of skos:Concept and skos:Collection.

A typical use case for nesting Collections or ConceptGroups is having a built-in hierarchical domain/theme-oriented classification for the concepts within a thesaurus (sometimes also called microthesauri). This kind of pattern exists for example in EuroVoc, the UNESCO Thesaurus, CAB Classified Thesaurus, GACS and probably many others.

I'm wondering when to use one or the other representation (or both).

skos:Collection and skos:member are simpler and already exist in original SKOS. But skos:Collections are really quite underspecified in original SKOS. The SKOS Primer talks mainly about array / node label use cases, and it doesn't even mention that skos:member can be used between Collections. This fact is only shown in the SKOS Reference as S32, "The rdfs:range of skos:member is the union of classes skos:Concept and skos:Collection.", without any explanation of why or when having another Collection as the object would be desirable.

The ISO 25964 extensions define the two subclasses of skos:Collection (iso-thes:ConceptGroup and iso-thes:ThesaurusArray) much better and also the ISO standard defines the relationships iso-thes:superGroup and iso-thes:subGroup, e.g. "A higher-level group of which this group is a member". So this representation is much better defined and documented.

So, if you had a classified thesaurus, or one with hierarchical microthesauri, would you use plain skos:Collections nested with skos:member, or would you represent the classes/microthesauri as iso-thes:ConceptGroups and use iso-thes:superGroup and iso-thes:subGroup for representing their hierarchy? Something else? Both?

I'm asking because the newest version of UNESCO Thesaurus seems to do both, and this is currently a bit problematic for displaying in Skosmos [2] since both kinds of hierarchy are shown to the user. I'd like to better understand which pattern to favor (if any). This also affects the modeling choices to be made in thesauri I'm involved in developing, for example YSO and GACS.

-Osma

[1] http://purl.org/iso25964/skos-thes#


[2] https://github.com/NatLibFi/Skosmos/issues/433


--
Osma Suominen
D.Sc. (Tech), Information Systems Specialist National Library of Finland P.O. Box 26 (Kaikukatu 4)
00014 HELSINGIN YLIOPISTO
Tel. +358 50 3199529
osma.suominen@helsinki.fi
http://www.nationallibrary.fi

Received on Monday, 18 January 2016 18:46:17 UTC