Re: Multiple objects vs. intersectionOf when defining subclasses

On 3 Jul 2019, at 11:28, Antoine Zimmermann <antoine.zimmermann@emse.fr> wrote:
> 
> From an OWL inference point of view, the two design patterns give identical results. However, the two structures may lead to different behaviours:
> 1) different optimisations can apply to different structures, even if they are equivalent

In this case, probably not. Maybe if the conjunction appears elsewhere negated (so normalisation can kick in in tableau reasoners). Otherwise, it almost certainly doesn’t make a difference (e.g., probably handled in parsing).

> 2) for data processing other than OWL reasoning (query answering,

First, query answering *is* a kind of reasoning and this shouldn’t interfere. Querying the axiomatic structure of the Tbox…yes, it could be annoying. But it could be helpful.

> indexing,

I don’t know what that is.

> graph analysis,

Ditto.

> etc), the shape of the graph is important. In general, the structure with owl:intersectionOf is more cumbersome and should be avoided in the right-hand-side of a subClassOf axiom.

It’s true that they are equivalent, but  if you ever want to strengthen your axiom to an equivalence, it will make a big difference.

Also, from a “graph shape” perspective, you can use the conjunction to cluster “related” subsumers. You need to recognise that this is a very weak signal for other people, but hey, it’s a signal.

It will bump your ontology out of RDFS, which can be a bit annoying. It will also, for many measurers, boost your “inferred subsumption” count, which is a tad annoying.

My rule of thumb is:

1. Use the conjunction if it feels more natural from a modelling perspective and especially if you ever want to consider it for a definition.
2. If 1 but you want to use an RDFS reasoner (or other subsumption graph traverser) then set up a normalisation step in your publishing process that basically splits all those conjunctions.

If you always are happy to use an e.g., EL reasoner, don’t worry about 2.

Cheers,
Bijan.

Received on Wednesday, 3 July 2019 14:10:11 UTC