=Paper=
{{Paper
|id=None
|storemode=property
|title=MEDLEY results for OAEI 2012
|pdfUrl=https://ceur-ws.org/Vol-946/oaei12_paper8.pdf
|volume=Vol-946
|dblpUrl=https://dblp.org/rec/conf/semweb/Hassen12
}}
==MEDLEY results for OAEI 2012
==
https://ceur-ws.org/Vol-946/oaei12_paper8.pdf
M EDLEY Results for OAEI 2012
Walid Hassen
University of Tunis El Manar - Faculty of Sciences of Tunis
Computer Science Department - LIPAH
Campus Universitaire, 1060 Tunis, Tunisia
walidhassenlipah@gmail.com
Abstract. M EDLEY is an alignment method based on lexical and structural treat-
ments. This method includes a specific technique to deal with multilingual on-
tologies. This paper introduces M EDLEY and summarizes the results for OAEI
2012.
1 Presentation of the system
M EDLEY can be presented as an OWL ontology alignment method that relies on simple
similarity metrics. Each ontology pair, can be transformed into graphs structures. This
means that links are OWL primitives and nodes are classes, properties, and individuals.
The algorithm includes a lexical, structural treatment. Each node can be matched with
few ones, then M EDLEY select pairs that maximize the global similarity value.
1.1 State, purpose, general statement
M EDLEY generates alignments between OWL-DL ontologies based on simple lexical
metrics and structures matching between links of each node (class, property, instance).
Specific treatment is applied for multilinguality issue, using a dictionnary to find equiv-
alence between concepts labelled in diffrent natural languages.
1.2 Specific techniques used
Each entity in the first ontology is aligned each entity in the second, in a primary step,
in lexical metrics, then in structural treatment. The algorithm reiterate this process for
all ontologies’s concepts.
– Lexical treatment : q-gram [1] and levenshtein [2] measures were used to calculate
the similarity measures between nodes. In addition, treatments and tokenization
stemmatisation were conducted.
– Structural treatment :If an entity belongs to a given ontology has a neighbor that is
already part of the alignment set, then the node that neighbor is aligned to must be
a neighbor of any prospective match for this entity.
1.3 Adaptations made for the evaluation
The M EDLEY method deals with three test suites used in the Ontology Alignment Eval-
uation Initiative (OAEI 2012). The method was wrapped in a certain folder structure to
be evaluated locally after being integrated in the SEALS platform. The package con-
tains all the libs files required by the method and a zipped .jar file that acts as a bridge.
1.4 Link to the system and parameters file
The release of the M EDLEY method and the parameter file used for OAEI 2012 are
located at https://github.com/medley.
2 Results
In this section, we present the results obtained by M EDLEY in the OAEI 2012.
2.1 Benchmark
The benchmark tests sets can be divided into eight groups: 101, 20x, 22x, 23x, 24x,
25x, 26x and 30x. For each group the mean values of precision and recall are computed.
Table 1 shows the values of the evaluation metrics. Tables 1, 2 and 3 recapitulate the
obtained values for this track.
Table 1. Results on Biblio
Test group Precision Recall F-Measure
101 0.72 1.0 0.84
20x 0.43 0.4 0.408
22x 0.716 1.0 .988
23x 0.781 1.0 0.853
24x 0.633 0.572 0.571
25x 0.51 0.4 0.421
26x 0.322 0.357 0.31
2.2 Conference
In scenario 1, M EDLEY have 0.54 of precision and 0.50, with 0.52 as recall an f-
measure about 0.52. In scenario 2, M EDLEY performs 0.59 of precision, 0.42 recall
and 0.49 of f-measure.
Table 2. Results on Benchmark 2
Test group Precision Recall F-Measure
101 1.00 1.00 1.00
20x 0.697 0.4 0.493
22x 0.998 1.0 1.0
23x 0.995 1.0 1.0
24x 0.787 0.57 0.63
25x 0.757 0.439 035
26x 0.611 0.354 0.435
Table 3. Results on Benchmark 3
Test group Precision Recall F-Measure
101 0.79 1.00 0.88
20x 0.568 0.4 0.454
22x 0.805 1.0 0.888
23x 0.88 1.0 0.93
24x 0.715 0.571 0.609
25x 0.642 0.4 0.463
26x 0.695 0.352 0.398
Fig. 1. M EDLEY components
2.3 Multifarm
For treating multilingual ontologies, our method uses an external resource as sketched
by figure 1 for the translation stage1 . Tables 4, 5, 6, 7, 8, 9 and 10 summarize the results.
1
http://www.freelang.com/dictionnaire/index.php
Table 4. Group (cz) as source ontology
Test group Precision Recall F-Measure
cz-de 0.51 0.07 0.13
cz-en 0.33 0.09 0.14
cz-es 0.43 0.07 0.12
cz-fr 0.33 0.05 0.09
cz-nl 0.33 0.06 0.10
cz-pt 0.46 0.08 0.13
cz-ru 0.00 0.00 NaN
Table 5. Group (de) as source ontology
Test group Precision Recall F-Measure
de-en 0.40 0.10 0.15
de-es 0.43 0.09 0.15
de-fr 0.40 0.09 0.14
de-nl 0.38 0.09 0.15
de-pt 0.43 0.09 0.15
de-ru 0.00 0.00 NaN
Table 6. Group (en) as source ontology
Test group Precision Recall F-Measure
en-es 0.54 0.48 0.51
en-fr 0.62 0.61 0.61
en-nl 0.56 0.42 0.48
en-pt 0.57 0.51 0.54
en-ru 0.05 0.00 0.00
Table 7. Group (es) as source ontology
Test group Precision Recall F-Measure
es-fr 0.31 0.04 0.08
es-nl 0.21 0.03 0.05
es-pt 0.50 0.11 0.18
es-ru 0.02 0.00 0.00
3 General comments
We participate this year for the first time in OAEI and see the result obtained by our
method. The evaluation and comparison of ontology alignment and schema matching
components as OAEI is very useful for the development of such
Table 8. Group (fr) as source ontology
Test group Precision Recall F-Measure
fr-nl 0.45 0.10 0.16
fr-pt 0.35 0.08 0.14
fr-ru 0.00 0.00 NaN
Table 9. Group (nl) as source ontology
Test group Precision Recall F-Measure
nl-pt 0.31 0.07 0.11
nl-ru 0.03 0.00 0.00
Table 10. Group (pt) as source ontology
Test group Precision Recall F-Measure
pt-ru 0.03 0.00 0.00
3.1 Discussions on the way to improve the proposed system
M EDLEY is still a primary work that needs to be adressed on few levels, notably, to deal
with greater ontologies.
4 Conclusion
In this paper, we presented M EDLEY as an alignment method. The new proposed method
M EDLEY, shows a special focus on multilinguality. The alignment process is based on
examining the structures and the informative wealth on each ontlogy pair to align.
References
1. Ukkonen, E.: Approximate string-matching with Q - GRAMS and maximal matchs. Theoretical
Computer Science 92(1) (1992) 191–211
2. Levenshtein, I.V.: Binary codes capables of corrections, deletions, insertions and reversals.
Soviet Physics-Doklady 10(8) (1966) 707–710