Databases Reference
In-Depth Information
Ta b l e 5 .
Statistics about test ontologies
Ontology
DL expressivity
SC Ontology
33
AL
D
20081
28
8
5
3542
(
)
Adhesome
34
12043
40
33
37 2032
ALCHN
(
D
)
GeoSkills
35
ALCHOIN
D
14966
613
23
21 2620
(
)
Eukariotic
36
38
11
1
0
11
ALCON
Breast Cancer
37
878
196
22
3
113
ALCROF
(
D
)
Economy
38
1625
339
45
8
482
ALCH
(
D
)
Resist
39
239
349
134 38
75
ALUF
(
D
)
Finance
40
16014
323
247 74 2466
ALCROIQ
(
D
)
Earthrealm
41
931
2364 215 36
171
ALCHO
(
D
)
requirement” in the screenshot. For instance, a knowledge engineer could decide to
import additional information available as Linked Data and run the CELOE algorithm
again to see whether di
ff
erent suggestions are provided with additional background
knowledge.
Evaluation.
To evaluate the suggestions made by our learning algorithm, we tested it
on a variety of real-world ontologies of di
erent sizes and domains. Please note that we
intentionally do not perform an evaluation of the machine learning technique as such on
existing benchmarks, since we build on the base algorithm already evaluated in detail in
[88]. It was shown that this algorithm is superior to other supervised learning algorithms
for OWL and at least competitive with the state of the art in ILP. Instead, we focus on
its use within the ontology engineering scenario. The goals of the evaluation are to
1. determine the influence of reasoning and heuristics on suggestions, 2. to evaluate
whether the method is su
ff
cient to work on large real-world ontologies.
To perform the evaluation, we wrote a dedicated plugin for the Protégé ontology
editor. This allows the evaluators to browse the ontology while deciding whether the
suggestions made are reasonable. The plugin works as follows: First, all classes with
at least 5 inferred instances are determined. For each such class, we run CELOE with
di
ciently e
ff
erent settings to generate suggestions for definitions. Specifically, we tested two
