Raw

Although there's definitely a place for up-front schema design, while coding it's often convenient to leave such stuff open, subject to change as the code informs. RDF instance data still makes a lot of (semantic) sense without having a schema/ontology, so this can be deferred indefinitely...



Sometime soon I'm going to need a RESTful web service description language and an access control/authentication language (note to self: remember to check what KnoBot does for these - is there an ESW page for the latter yet?)

Of the WebDescriptionProposals, Ian Davis' WAIF looks the best fit for what I need, not least because it's already RDF-friendly. From his write-up there's everything needed to put manually put together a schema/ontology (hmm, the parameters stuff looks like it might be expressable pretty richly with OWL cardinality). But I think there's enough info in his instance examples to automatically build the skeleton of a schema - which is all I need in the near future (to use with Jena's schemagen, that autogenerates Java source corresponding to classes & properties).

Has anyone set something up for this yet? For Jena? If not I'll add it to my to-do, shouldn't take long at all (and could actually make a handy online service, giving me another use case...)

Here's how I imagine it working:

1. RDF model is loaded with all available instance data
2. RDFS/OWL inference applied 
3. Statements matching  {?c rdf:type rdfs:Class}, {?p rdf:type rdf:Property}, {?p rdfs:domain ?d} etc extracted (and/or corresponding OWL terms)
4. Statements annotating the classes and properties extracted
5. Skeleton schema serialised out 

I'm not 100% sure about 4 (which in practice would be part of 3), I think {?c ?p o?}, {?p ?p 2 o?} might do the trick. For those cases, and in general, it would be necessary to filter out any statements involving a bnode (assuming there were some inferred). 

The extraction could be done in the main code, but could be sweeter done as a SPARQL query. 

The resulting schema would be missing things like special (individual) resource, and I doubt much OWLishness could be reliably extracted. But still it could be a good timesaver. Heh, could probably have coded it in the time it taken to write this, but I'm meant to be working...