Is it possible to implement blank node skolemization in SPARQL without iteration? It seems to me that iteration is required to skolemize chains of blank nodes, such as:
@prefix : <http://example.com/> .
[ a :A ;
:p1 [
a :B
]
] .
A SPARQL Update operation for skolemization can start from the blank nodes that appear as subjects only in triples without blank node objects:
DELETE {
?b1 ?outP ?outO .
?inS ?inP ?b1 .
}
INSERT {
?iri ?outP ?outO .
?inS ?inP ?iri .
}
WHERE {
{
SELECT ?b1 (uuid() AS ?iri)
WHERE {
{
SELECT DISTINCT ?b1
WHERE {
?b1 ?p1 [] .
FILTER isBlank(?b1)
FILTER NOT EXISTS {
?b1 ?p2 ?b2 .
FILTER isBlank(?b2)
}
}
}
}
}
?b1 ?outP ?outO .
OPTIONAL {
?inS ?inP ?b1 .
}
}
This operation can be repeated until no blank nodes are found in the data:
ASK {
?bnode ?p [] .
FILTER isBlank(?bnode)
}
Is it possible to avoid the iteration and implement the blank node skolemization in a single SPARQL Update operation?
(Also, this approach assumes there are no "orphan" blank nodes (i.e. blank nodes that appear only as objects).)
I found a two-step solution skolemising subjects and objects separately and storing the blank node aliases (links between blank nodes and IRIs via
owl:sameAs
) as intermediate data: