Re: Soft ontology reuse (was: Is LOD-cloud alive?)

Hi Stian,

Excellent post.

Yes, I suspect that many people use some variant of your soft reuse, probably very soft, and encouragement to think about making it less soft, and your suggestions for doing so are valuable.

Just a comment on your second example, in case people take it literally:
I would consider a triple such as
   <http://example.com/person1> ex:companyName "Example inc." .
leaps out at me as less than best practice.
Whenever I see a string with a predicate that isn't rdfs:label or similar, I question whether this is the right way of modelling.
In this case (as I'm sure you well know) you have a person, who has some relationship with a string "Example inc.".
(It could be they work for it, or own it, or hate it or something else, I guess.)
Whatever it is that the string might be thought of as identifying (be it a company or a film or whatever), you can't make any statements about, such as whether some other person has the same relationship with that thing, what its type is, etc., etc.
Hence the FOAF and schema.org structures, of course.

Best
Hugh

> On 19 Jan 2017, at 10:56, Stian Soiland-Reyes <soiland-reyes@manchester.ac.uk> wrote:
> 
> On Wed, 18 Jan 2017 12:01:22 -0800, Krzysztof Janowicz <janowicz@ucsb.edu> wrote:
>> No, they are open. They are just not 'linked' to other ontologies. In 
>> fact, many people would say that the way some of us think about 'reuse' 
>> of ontologies is highly problematic but that is an entirely different 
>> story. Keep in mind that the discussion here is a reflection on 
>> Alberto's email, not about whether LOV is useful or not (and I clearly 
>> believe it is).
> 
> Yes, directly reusing ontologies (e.g. owl:imports) can come with many
> technical challenges such as inconsitencies at reasoning level (e.g. OWL)
> or network dependencies (purl.org still fresh in mind)
> 
> So I understand many don't want to do such strong reuse - indeed I have myself
> moved instead to a model of a "soft reuse", with extensions of existing
> ontologies through "citations" instead.
> 
> 
> ## Soft ontology reuse
> 
> Copy only the URI (and type), then (important!) cite back to the original
> ontology with rdfs:isDefinedBy.  If that ontology does not already have an
> authorative rdfs:isDefinedBy or owl:versionIRI you can use, cry for a bit, and
> then use a retrievable IRI to the ontology - which might differ from the
> namespace.
> 
> 
> For instance (adapted from http://purl.org/pav/):
> 
> 	@prefix pav:  <http://purl.org/pav/> .
> 	@prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#> .
> 	@prefix dcterms:  <http://purl.org/dc/terms/> .
> 
> 
> 	pav:authoredBy  rdf:type    owl:ObjectProperty ;
> 	  rdfs:subPropertyOf  dcterms:creator .
> 
> 	dcterms:creator  rdfs:isDefinedBy  <http://purl.org/dc/terms/> .
> 
> Luckily here the namespace is already what is is declared as rdfs:isDefinedBy
> upstream and we don't need to cry :)
> 
> 
> (BTW, copying anything more than the URI would mean you are making a derived
> work and it would no longer be "fair use", and so have to fully comply with
> (and propagate) the original license - if that is even stated)
> 
> 
> With soft ontology reuse, inferences like dcterms:creator being a subproperty
> of dcterms:contributor would not be visible unless you also loaded the full
> dcterms ontology - a kind of optional inference if you like.
> 
> I think this would be my prefered way of soft ontology reuse.
> 
> 
> ## Even softer
> 
> 
> In some cases the above can get fragile - e.g. the vocabulary you cite is not
> directly compatible ontology-wise (e.g. too strong rdfs:range). In that case I
> would strongly encourage to still have a rdfs:seeAlso or SKOS mapping to
> existing ontologies where it seems obvious.
> 
> 
> For example, let's say you have an ontology of workers at companies, but it is
> for whatever reason not directly compatible with the model of FOAF; let's say
> you have a simple property ex:companyName:
> 
>    <http://example.com/person1> ex:companyName "Example inc." .
> 
> In FOAF this would instead be modelled indirectly with a resource for the work
> place and an inverse link foaf:member to the person:
> 
>    <http://example.com/> a foaf:Organization ;
>        foaf:name "Example inc." ;
>        foaf:member <http://example.com/person1> .
> 
> And in schema.org using schema:worksFor from the person:
> 
>   <http://example.com/person1> schema:worksFor <http://example.com> .
> 
>   <http://example.com> a schema:Organization;
>      schema:name "Example inc." .
> 
> 
> And so we can't use rdfs:isSubPropertyOf here to map ex:companyName to FOAF
> or schema.org, as that would imply things like person1 being an organization
> and having the name "Example inc.".
> 
> But let's still keep some kind of link:
> 
> 
> 	@prefix rdfs:   <http://www.w3.org/2000/01/rdf-schema#> .
> 	@prefix owl:    <http://www.w3.org/2002/07/owl#>.
> 	@prefix foaf:   <http://xmlns.com/foaf/0.1/>.
> 	@prefix skos:   <http://www.w3.org/2004/02/skos/core#>.
> 	@prefix schema: <http://schema.org/>.
> 	@prefix ex:     <http://example.org/>.
> 
> 	ex:companyName a owl:DatatypeProperty ;
> 	  rdfs:seeAlso foaf:name, foaf:member, schema:worksFor, schema:name ;
>          skos:closeMatch schema:worksFor ;
> 	  skos:relatedMatch foaf:name, foaf:member, schema:name .
> 
> 	foaf:name rdfs:isDefinedBy <http://xmlns.com/foaf/0.1/> .
> 	foaf:member rdfs:isDefinedBy <http://xmlns.com/foaf/0.1/> .
>        schema:name rdfs:isDefinedBy schema:name .
>        schema:worksFor rdfs:isDefinedBy schema:worksFor .
> 
> The various SKOS mapping properties can be used when something is more specific
> or broader: 
> https://www.w3.org/TR/skos-reference/#mapping
> 
> 
> This would mean your ontology can be not just Open, but also Linked - and thus
> resources with statements using just your ontology could be seen as part of the
> mythical "LOD cloud" rather than an isolated "semantic island".
> 
> Obviously in this second approach you can't do much reasoning and infer
> SKOS/FOAF statements, but at least you can follow the links rather than 
> be an isolated ontology. 
> 
> An isolated ontology/namespace with no links would in my view would just be
> slightly better than a arbitrary custom XML or JSON properties.  (Luckily using
> URIs still allows others to do such a mapping at a later point)
> 
> 
> 
> -- 
> Stian Soiland-Reyes
> University of Manchester
> http://www.esciencelab.org.uk/
> http://orcid.org/0000-0001-9842-9718
> 
>

Received on Thursday, 19 January 2017 18:38:29 UTC