Databases Reference
In-Depth Information
an annotated logic program framework to rank assertions under a PageRank
model, where the marginal assertion in a contradiction is defeated [12].
With respect to C4 (dynamic Linked Data), Context-Dependent reasoning
allows entailments to be updated on a context-by-context basis, where changes
to the ontology base can also be eciently supported (see [17]); Authoritative
reasoning does not directly support incremental updates, where truth mainte-
nance techniques would be required. (The following section presents an approach
that better handles reasoning and querying over Linked Data in highly dynamic
scenarios.)
With respect to C5 (more than RDFS/OWL required), both approaches gen-
eralise to the application of arbitrary rule-based reasoning, where the Context-
Dependent framework - a means to manage contexts - generalises further to any
form of deductive (or even inductive) reasoning process, as required.
Comparison of Both Approaches.
In terms of the differences between both ap-
proaches, the Context-Dependent approach is designed to run over small con-
texts, typically involving one “assertional” document and its recursive ontology
imports. Although the framework can handle aggregate contexts, the larger these
aggregate contexts become, the closer Context-Dependent reasoning resembles
the naıve case of standard reasoning over a monolithic graph. Thus, Context-
Dependent reasoning is not well-suited to deriving entailments across assertional
documents. The T-Box generated during Authoritative reasoning can be used to
cautiously derive entailments across assertional documents (effectively reflecting
a common consensus for a T-Box across all contexts); however, in practice, to
achieve scalability, the A-Linear profile disables such inferences.
Conversely, Context-Dependent reasoning trusts all axioms in a local context,
whereas Authoritative reasoning does not. In other words, Context-Dependent
reasoning allows non-authoritative reasoning within contexts, which Authorita-
tive reasoning never allows. With reference to Example 9, if a document imports
the DBpedia ontology involved, Context-Dependent reasoning will permit trans-
reasoning will not.
Support for same-as?
A primary limitation common to both approaches is the
inability to effectively reason over
owl:sameAs
relations. Context-Dependent rea-
soning can only process such relations with a single context, which will miss the
bulk of equivalence relations between assertional documents. In theory, Author-
itative reasoning can support
owl:sameAs
inferences, but for scalability reasons,
rules with A-Box joins are disabled in the SAOR implementation. However, in
other more focussed works, we have looked at specialised methods for authori-
tative reasoning of
owl:sameAs
relations in a Linked Data setting [42].
Indeed,
owl:sameAs
can produce huge volumes of inferences: in previous
work [42], we found 33,052 equivalent terms within a single (correct)
owl:sameAs
clique, which would require 33
,
052
2
=1
,
092
,
434
,
704 triples just to materi-
alise the pair-wise and reflexive
owl:sameAs
relations between terms in this one
group, even before applying any of the
eq-rep-*
rules for replacement. Given the
importance of
owl:sameAs
reasoning for aligning entities in Linked Data, the
