Ontology NL Search

RDF Schema (RDFS) Reference ​

W3C Recommendation: https://www.w3.org/TR/rdf-schema/

Purpose ​

RDFS provides vocabulary for describing RDF data structure. Foundation for class hierarchies and property relationships.

Key Classes ​

rdfs:Class ​

The class of all classes. Every class is an instance of rdfs:Class.

rdfs:Resource ​

Universal class. Everything in RDF is an rdfs:Resource.

rdf:Property ​

The class of all properties.

Critical Properties ​

rdfs:domain ​

Declares which class the SUBJECT of a triple must belong to:

turtle
ex:author rdfs:domain ex:Document .

# If: <book1> ex:author "Alice"
# Then inferred: <book1> rdf:type ex:Document

Multiple domains each independently infer a type (RDFS only infers, never constrains — the subject is entailed to be an instance of every declared domain class):

turtle
ex:property rdfs:domain ex:ClassA .
ex:property rdfs:domain ex:ClassB .
<node> ex:property "value" .
# Inferred: <node> rdf:type ex:ClassA AND ex:ClassB

rdfs:range ​

Declares which class the OBJECT of a triple must belong to:

turtle
ex:author rdfs:range ex:Person .

# If: <book1> ex:author <alice>
# Then inferred: <alice> rdf:type ex:Person

rdfs:subClassOf ​

Class hierarchy:

turtle
ex:HDMap rdfs:subClassOf ex:SimulationAsset .
ex:SimulationAsset rdfs:subClassOf rdfs:Resource .

# Transitive: If A subClassOf B and B subClassOf C, then A subClassOf C

rdf:type ​

Instance-of relationship:

turtle
<asset123> rdf:type ex:HDMap .
<asset123> rdf:type ex:SimulationAsset .

RDFS vs SHACL ​

FeatureRDFSSHACL
PurposeInferenceValidation
Property → Classrdfs:domain causes type inferenceNo inference
Class → Property-sh:targetClass + sh:path validates
DirectionProperty defines its domainClass defines its properties
StandardRDF 1.1 CoreSeparate W3C spec

Critical Difference ​

RDFS says: "If property P has domain C, ANY use of P means subject IS C"
SHACL says: "If node is C, then it SHOULD have property P (validation)"

Direction is opposite!

Why We Use SHACL, Not RDFS Domain ​

If we had RDFS domains: ​

turtle
ex:roadTypes rdfs:domain ex:HDMap .
<asset> ex:roadTypes "motorway" .
# Would infer: <asset> rdf:type ex:HDMap ✅

But our ontology uses SHACL: ​

turtle
ex:HDMapShape sh:targetClass ex:HDMap ;
  sh:property [ sh:path ex:roadTypes ; ... ] .

SHACL shapes DON'T create RDFS domain assertions. They're validation constraints, not inference rules!

Query Class Hierarchies ​

Find All Subclasses (Transitive) ​

sparql
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?subclass WHERE {
  ?subclass rdfs:subClassOf+ ex:SimulationAsset .
}

Property path rdfs:subClassOf+ = one or more steps

Find Direct Subclasses Only ​

sparql
SELECT ?subclass WHERE {
  ?subclass rdfs:subClassOf ex:SimulationAsset .
  FILTER NOT EXISTS {
    ?subclass rdfs:subClassOf ?intermediate .
    ?intermediate rdfs:subClassOf ex:SimulationAsset .
    FILTER(?intermediate != ex:SimulationAsset)
  }
}

Find All Asset Domains ​

sparql
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT DISTINCT ?domain WHERE {
  ?assetClass rdfs:subClassOf ex:SimulationAsset .
  BIND(replace(str(?assetClass), ".*/([^/]+)/v[0-9]+/.*", "$1") AS ?domain)
}

Result:

domain
--------------
hdmap
scenario
ositrace
environment-model
...

Combining RDFS + SHACL ​

Use RDFS for: ​

  • ✅ Class hierarchies (ex:SimulationAsset → ex:HDMap)
  • ✅ Finding all asset types
  • ✅ Type assertions on assets

Use SHACL for: ​

  • ✅ Property-domain associations (sh:targetClass + sh:path)
  • ✅ Value constraints (sh:in, sh:pattern)
  • ✅ Cardinality constraints

Hybrid Query ​

sparql
# Find asset domains via RDFS
SELECT DISTINCT ?domain WHERE {
  ?assetClass rdfs:subClassOf ex:SimulationAsset .
  BIND(replace(str(?assetClass), ".*/([^/]+)/v[0-9]+/.*", "$1") AS ?domain)
}

# Find properties per domain via SHACL
SELECT ?domain ?property WHERE {
  ?shape sh:targetClass ?assetClass .
  ?shape sh:property [ sh:path ?property ] .
  ?assetClass rdfs:subClassOf ex:SimulationAsset .
  BIND(replace(str(?assetClass), ".*/([^/]+)/v[0-9]+/.*", "$1") AS ?domain)
}

Key Insight ​

RDFS = REASONING. SHACL = VALIDATION.

For our architecture:

  • Query RDFS: Class hierarchies (asset type discovery)
  • Query SHACL: Property-domain mappings (sh:targetClass + sh:path)

Don't expect SHACL shapes to create RDFS domains!

References ​