Databases Reference
In-Depth Information
The adapted algorithm for solving the learning problems, which occur in the ontology
engineering process, is called
CELOE (Class Expression Learning for Ontology En-
gineering)
. It was implemented within the open-source framework DL-Learner.
28
DL-
Learner [79,80] leverages a modular architecture, which allows to define di
erent types
of components: knowledge sources (e.g. OWL files), reasoners (e.g. DIG
29
or OWL
API based), learning problems, and learning algorithms. In this overview, we focus on
the latter two component types, i.e. we define the class expression learning problem in
ontology engineering and provide an algorithm for solving it.
ff
Learning Problem.
The process of learning in logics, i.e. trying to find high-level ex-
planations for given data, is also called
inductive reasoning
as opposed to
inference
or
deductive reasoning
. The main di
erence is that in deductive reasoning it is formally
shown whether a statement follows from a knowledge base, whereas in inductive learn-
ing new statements are invented. Learning problems, which are similar to the one we
will analyse, have been investigated in
Inductive Logic Programming
[121] and, in fact,
the method presented here can be used to solve a variety of machine learning tasks apart
from ontology engineering.
In the ontology learning problem we consider, we want to learn a formal description
of a class
A
, which has (inferred or asserted) instances in the considered ontology. In the
case that
A
is already described by a class expression
C
via axioms of the form
A
ff
C
or
A
generalised, or relearned from
scratch by the learning algorithm. To define the class learning problem, we need the
notion of a
retrieval
reasoner operation
R
K
(
C
).
R
K
(
C
) returns the set of all instances of
C
in a knowledge base
≡
C
, those can be either refined, i.e. specialised
/
is clear from the context, the subscript can be omitted.
Definition 6 (class learning problem).
Let an existing named class A in a knowl-
edge base
K
.If
K
K
be given. The
class learning problem
is to find an expression C such that
R
K
(
C
)
R
K
(
A
)
.
Clearly, the learned expression
C
is a description of (the instances of)
A
. Such an ex-
pression is a candidate for adding an axiom of the form
A
=
≡
C
or
A
C
to the
knowledge base
. If a solution of the learning problem exists, then the used base
learning algorithm (as presented in the following subsection) is complete, i.e. guar-
anteed to find a correct solution if one exists in the target language and there are
no time and memory constraints (see [87,88] for the proof). In most cases, we will
not find a solution to the learning problem, but rather an approximation. This is nat-
ural, since a knowledge base may contain false class assignments or some objects
in the knowledge base are described at di
K
erent levels of detail. For instance, in
Example 1, the city “Apia” might be typed as “Capital” in a knowledge base, but
not related to the country “Samoa”. However, if most of the other cities are related
to countries via a role isCapitalOf, then the learning algorithm may still suggest
City and isCapitalOf at least one Country since this describes the majority
of capitals in the knowledge base well. If the knowledge engineer agrees with such a
definition, then a tool can assist him in completing missing information about some
capitals.
ff
