=Paper=
{{Paper
|id=Vol-2788/om2020_poster4
|storemode=property
|title=Towards matching of domain ontologies to cross-domain ontology: evaluation perspective
|pdfUrl=https://ceur-ws.org/Vol-2788/om2020_poster4.pdf
|volume=Vol-2788
|authors=Martin Šatra,Ondřej Zamazal
|dblpUrl=https://dblp.org/rec/conf/semweb/SatraZ20
}}
==Towards matching of domain ontologies to cross-domain ontology: evaluation perspective==
https://ceur-ws.org/Vol-2788/om2020_poster4.pdf
Towards Matching of Domain Ontologies to
Cross-Domain Ontology: Evaluation Perspective
Martin Šatra and Ondřej Zamazal
Department of Information and Knowledge Engineering,
University of Economics, W. Churchill Sq.4, 130 67 Prague 3, Czech Republic,
{satm03|ondrej.zamazal}@vse.cz
1 Introduction
Ontology matching, as a process of matching two or more ontologies, is usually
aimed at matching of domain ontologies. However, there are also other kinds of
ontologies which make sense to align (and particularly with domain ontologies).
Cross-domain (general) ontologies cover more domains. For example, the DB-
pedia ontology is a cross-domain ontology. It contains concepts, such as Agent,
Device, Food, Place, from diverse domains. In comparison, domain ontologies
focus on concepts from one area. For instance, the confof ontology from Onto-
Farm1 contains concepts such as Contribution, Event, Person dealing with the
conference organization.
While motivation use cases (such as information integration and information
sharing, e.g. in [1]) for matching of domain ontologies to a cross-domain ontology
are to a large degree similar as for matching of domain ontologies, there are
different challenges with regard to matching. We claim that matching to cross-
domain ontology is more difficult for traditional ontology matching systems since
a cross-domain ontology contains concepts from various areas and it is more
difficult to recognize proper concepts to align. Next a cross-domain ontology is
usually larger. In all, we can expect a higher amount of false positives (lowering
precision) since string-based matching techniques will be more often confused.
There has not yet been much work done on this kind of matching. Authors in [3]
focused on matching enhanced with knowledge of the domain and they evaluated
their approach on matching two domain ontologies to the DBpedia ontology.
Further there is a close effort of matching of foundational ontologies [2].
2 Reference Alignment and Evaluation
For building of reference alignments (RA) we merely focused on entities of DB-
pedia ontology 2 from DBpedia namespace and three ontologies from OntoFarm:
0
Copyright c 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
1
https://owl.vse.cz/ontofarm/
2
http://downloads.dbpedia.org/2016-10/dbpedia_2016-10.owl
confof, ekaw, sigkdd. The process of constructing RA was supported by basic
ontology matching techniques available from the Alignment API.3 Further, a
thorough manual matching was applied. Based on these input a tentative RA
were prepared.4 Finally, the RA were reconciled with the existing RA for the con-
ference track of OAEI (Ontology Alignment Evaluation Initiative)5 consisting of
correspondences between OntoFarm domain ontologies. The resulted RA contain
both equivalence and subsumption correspondences with 1:1 cardinality.6
For evaluation (merely equivalence correspondences) we employed several
matching systems from OAEI 2019: AML, DOME, LogMap and LogMapLt.7
According to the results in Table 1 AML, DOME and LogMap have very sim-
lar results in terms of F1 -measure. While LogMap is better in precision, AML
and DOME are better in recall. The system based only on string technique,
LogMapLt, has the lowest F1 -measure. As expected evaluation metrics are rather
low (e.g. 0.42 vs. 0.70 in terms of comparing F1 -measures with regard to the re-
sult of matching of domain ontologies in the conference track of OAEI 2019).
Table 1. Precision, F1 -measure and Recall for systems (micro-average).
System Prec. F1 -m. Rec.
AML 0.30 0.42 0.67
DOME 0.32 0.42 0.60
LogMap 0.37 0.41 0.47
LogMapLt 0.33 0.36 0.40
3 Conclusions and Future Work
Low scores of measures show that the corresponding test cases are difficult for
traditional ontology matching systems since they mainly focus on matching of
domain ontologies. In future we plan to engage more systems and we also plan
to extend the RA. We envisage to employ the RA within the conference track
of the OAEI 2020 as a new challenge for matching systems.
Ondřej Zamazal is supported by the CSF grant no. 18-23964S.
References
1. G. Kobilarov, T. Scott, Y. Raimond, S. Oliver, C. Sizemore, M. Smethurst, C. Bizer,
and R. Lee. Media meets semantic web–how the bbc uses dbpedia and linked data
to make connections. In ESWC. Springer, 2009.
2. D. Schmidt, A. Pease, C. Trojahn, and R. Vieira. Aligning conference ontologies
with SUMO: A report on manual alignment via wordnet. In Proc. of the Joint
Ontology Workshops, CEUR, 2019.
3. K. Slabbekoorn, L. Hollink, and G.-J. Houben. Domain-aware ontology matching.
In International Semantic Web Conference, pages 542–558. Springer, 2012.
3
http://alignapi.gforge.inria.fr/
4
RA were done by one evaluator and eventually one referee confirmed the resulted
RA during a discussion.
5
http://oaei.ontologymatching.org/
6
Available on the OntoFarm web, https://owl.vse.cz/ontofarm/#ra-to-dbpedia.
7
System papers are available at http://om2019.ontologymatching.org/#ap