RDF representation of union data types

[jechterhoff]

Description

A <<union>> class represents a choice between a set of different properties and their values. Unions are used in INSPIRE application schemas and need to be represented in INSPIRE RDF schemas as well.

The schema conversion rule for union data types in ISO 19150-2 is insufficient as it does not handle cases where values are a mix of object or datatypes, or the same value type is used by more than one option. The conversion rule focuses on the value types used by the properties of a union, completely ignoring the fact that the properties themselves can carry meaning and therefore must not be discarded.

An approach is needed to represent union data types in INSPIRE RDF schemas.

[jechterhoff]

A <<union>> can be represented in OWL using a logical combination of unqualified cardinality constraints. The according class expression ensures that values are only given for one of the union properties.

:opt1 rdf:type owl:DatatypeProperty ; rdfs:range xsd:string .

:opt2 rdf:type owl:DatatypeProperty ; rdfs:range xsd:integer .

:opt3 rdf:type owl:DatatypeProperty ; rdfs:range xsd:string .

:Union1 rdf:type owl:Class ;

rdfs:subClassOf [ rdf:type owl:Class ; owl:unionOf ( [ owl:intersectionOf ( [ rdf:type owl:Restriction ; owl:onProperty :opt1 ; owl:minCardinality "1"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt2 ; owl:cardinality "0"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt3 ; owl:cardinality "0"^^xsd:nonNegativeInteger]) ; rdf:type owl:Class] [ owl:intersectionOf ( [ rdf:type owl:Restriction ; owl:onProperty :opt1 ; owl:cardinality "0"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt2 ; owl:cardinality "0"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt3 ; owl:minCardinality "1"^^xsd:nonNegativeInteger]) ; rdf:type owl:Class] [ owl:intersectionOf ( [ rdf:type owl:Restriction ; owl:onProperty :opt1 ; owl:cardinality "0"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt2 ; owl:minCardinality "1"^^xsd:nonNegativeInteger] [ rdf:type owl:Restriction ; owl:onProperty :opt3 ; owl:cardinality "0"^^xsd:nonNegativeInteger]) ; rdf:type owl:Class] ) ] .

[jechterhoff]

However, if the <<union>> itself is represented as an OWL class, then encoding of actual data would require the creation of individuals that just represent the "choice". A more natural approach would be to encode the "choice" where it is actually of interest, i.e. where the <<union>> is used as value type.

[jechterhoff]

Therefore, don't convert the <<union>> to an OWL class. Instead, transform it as follows:

• Each UML property (P) that has the <<union>> as value type is replaced by new properties, one for each option (O) of the <<union>>. The name of a new property shall represent the specific combination of P and O, and shall be unique among the names of the properties that belong to the class that P belongs to. The value type of the new property will be the value type of O. The multiplicity is set as follows: the lower bound is always 0, while the upper bound is the product of the maximum multiplicity of P and O - or unbounded if one of them is unbounded. The documentation (natural / human readable name, definition, description) of P and O is merged.
• For each UML property (P) that is replaced by new properties as described before AND that has maximum multiplicity of 1, create a logical combination of class expressions and add it to the class that P belongs to:
  • For each new property (N):
  • Create a set of unqualified cardinality restrictions, to express that the minimum cardinality of N is 0 (so there can be any number of values for this property, it can also be omitted completely), and the cardinality of the other new properties is restricted to exactly 0.
  • Combine the cardinality restrictions within an intersection class expression. The meaning of this expression is that if a value is given for N, then no values may be provided for the other new properties.
  • Combine the intersection class expressions for all new properties within a single union class expression. This essentially represents the meaning of the <<union>>: if a value is given for one of the options, then there must not be values for the other options.
  • NOTE: If the maximum multiplicity of P is greater than one, then it can have values from more than one option. In that case, it would not be correct to add a class expression that restricts values to only one of the new properties.

[jechterhoff]

The described approach has been documented in the guidelines.