RDFS and OWL Reasoning for Linked Data - Reasoning Web - page 92

C1 Linked Data is huge , that is, it needs highly scalable or modular reasoning
techniques;
C2 Linked Data is not “pure” OWL , that is, a lot of RDF Data published as
Linked Data violates the strict syntactic corset of OWL (2) DL, and thus is
not directly interpretable under OWL Direct Semantics;
C3 Linked Data is inconsistent , that is, if you take the Web of Data in its
entirety, it is quite normal to encounter inconsistencies - not only from ac-
cidental or malicious datasets - but also because publishers may express
contradicting views;
C4 Linked Data is evolving , that is, RDF Graphs on the Web evolve, they
change, information is added and removed;
C5 Linked Data needs more than RDFS and OWL , that is, there is more implicit
data hidden in Linked Data than can be captured with the semantics of
RDFS and OWL alone.
In this lecture, we will introduce and discuss robust and scalable reasoning tech-
niques that are specifically tailored to deal with these challenges in diverse and
large-scale Linked Data settings. We first recapitulate the basic concepts of RDF,
Linked Data, RDFS, OWL and SPARQL, and, with reference to a practical real-
world scenario, we exemplify the use of query-rewriting techniques and rule-based
approaches for reasoning over Linked Data (Section 2). We then will reflect in
more detail on challenges C1-C5 above and discuss basic architectures, namely
data-warehousing and on-the-fly-traversal based approaches, to reason about
and query over Linked Data (Section 3).
While the W3C has defined two standard OWL fragments tailored for both
rule-based (OWL 2 RL) and query-rewriting (OWL 2 QL) techniques, as we
will see, standard reasoning within these fragments may not be a perfect fit for
the Linked Data use-case; we will thus discuss which OWL features are actually
predominantly used in current Linked Data and how these features relate to
various standard and non-standard OWL fragments (Section 4).
The remaining chapters will then introduce specific approaches to Linked Data
reasoning, each of which addresses some of the above-named challenges:
Section 5 introduces two rule-based approaches for Linked Data warehouses,
both of which apply a cautious form of materialisation that considers where
the axioms in question were found on the Web:
- Context-Dependent Reasoning [17], which is deployed within the Sindice
semantic web engine [54].
- Authoritative Reasoning [41,12], which is deployed within the Scalable
Authoritative OWL Reasoner (SAOR) as part of the Semantic Web
Search Engine (SWSE) [40].
Section 6 presents an alternative approach to enrich on-the-fly-traversal based
approaches for Linked Data query processing with lightweight OWL Rea-
soning [67].
Section 7 presents a technique for reasoning with “attribute equations” [10]
that models interdependencies between numerical properties expressible in
terms of simple mathematical equations, which we argue is complementary