Hi Erik,
in addition to your specific named graphs, you also need to include the default
context (with a null URI) when
executing your query. This ensures that the inferred triples can also be read.
Unfortunately, at the moment it is not possible to retrieve only those triples
which are inferred from a
certain combination of named graphs. I also think that it is difficult, if not
impossible, to implement this
on top of a forward chaining reasoner since it would require to store tons of
provenance information.
A backward chaining reasoner may be better suited for this usage scenario.
Virtuoso provides a mechanism called rule sets. These rule sets are associated
with a URI (e.g., <urn:ruleset1>) and
are based on RDFS/OWL descriptions retrieved from a combination of named graphs
(e.g. <urn:graph1>, <urn:graph2>, ...).
When executing a SPARQL query a rule set can be selected with a Virtuoso
specific SPARQL extension:
define input:inference '<urn:ruleset1>'
select ?s where { ?s ?p ?o }
Afaik, Virtuoso's rule sets are implemented by using query rewriting.
Maybe, OWLIM can implement something comparable for its forward chaining
approach?
For example, the user may specify multiple rule sets and associated named
graphs for the inferred triples:
<urn:graph1>, <urn:graph2>, ... => <urn:graph-inferred-with-ruleset1>
Now, OWLIM's reasoner is executed for each rule set (and combination of named
graphs) instead of using
only one global reasoner.
I don't know if this approach is scalable since it would produce many
duplicated inferred triples and
also may lead to many reasoner runs if a named graph is changed that takes part
in multiple rule sets.
Another approach can be that OWLIM uses some sort of filtering to exclude those
inferred triples
where not at least one asserted triple within the queried named graphs exists
that mentions its subject,
predicate or object.
I've simulated this approach by using SPARQL filters but it tends to be rather
slow with OWLIM SE:
select ?o where {
?s ?p ?o
filter (!bound(?o) || isLiteral(?o) || exists {
graph ?g { { ?o ?p1 ?o1 } union { ?s2 ?o ?o2 } union { ?s3 ?o3 ?o } }
}
}
Note: "!bound(?o)" is not required in this query - I've just included it to
demonstrate the generic case.
Best regards,
Ken
From: owlim-discussion-boun...@ontotext.com
[mailto:owlim-discussion-boun...@ontotext.com] On Behalf Of Erik Boye
Sent: Thursday, January 03, 2013 5:16 PM
To: owlim-discussion@ontotext.com
Subject: [Owlim-discussion] Inference with named graph (context)
Hi,
We are in the process of establishing an infrastructure for storing and
querying power system network models based on OWLIM and Sesame. We need to
track different versions of the power system network model (at the same time),
and thought could be achieved using named graphs: One named graph per version
of the power system network model.
The queries that are performed on the power system network model heavily
depends on OWLIM inference. With only one named graph in the repository, all
works fine. But when there are more than one named graph in the repository, and
queries need to be restricted to a specific named graph, our queries fail:
To restrict the queries to a specific named graph the following Sesame API
method is used:
ContextAwareConnection#setReadContexts(URI[])
The problem seems to be related to the fact that statements inferred by OWLIM
are stored in the unnamed graph (default context), and therefore not included
when restricting the queries to a specific named graph. Is this the expected
behavior? If yes, why? And does anyone have a suggestion for a workaround,
except from the obvious one: separation on the repository level?
Best regards,
Erik
_______________________________________________
Owlim-discussion mailing list
Owlim-discussion@ontotext.com
http://ontomail.semdata.org/cgi-bin/mailman/listinfo/owlim-discussion
