=Paper=
{{Paper
|id=Vol-2032/oaei17_paper10
|storemode=property
|title=POMap results for OAEI 2017
|pdfUrl=https://ceur-ws.org/Vol-2032/oaei17_paper10.pdf
|volume=Vol-2032
|authors=Amir Laadhar,Faiza Ghozzi,Imen Megdiche,Franck Ravat,Olivier Teste,Faiez Gargouri
|dblpUrl=https://dblp.org/rec/conf/semweb/LaadharGMRTG17
}}
==POMap results for OAEI 2017==
https://ceur-ws.org/Vol-2032/oaei17_paper10.pdf
POMap results for OAEI 2017
Amir Laadhar1 , Faiza Ghozzi2 , Imen Megdiche1 , Franck Ravat1 , Olivier
Teste1 , and Faiez Gargouri2
1
Paul Sabatier University, IRIT (CNRS/UMR 5505) 118 Route de Narbonne 31062
Toulouse, France
{amir.laadhar,imen.megdiche,franck.ravat,olivier.teste}@irit.fr,
2
University of Sfax, MIRACL Sakiet Ezzit 3021, Tunisie
{faiza.ghozzi,faiez.gargouri}@isims.usf.tn
Abstract. Ontology matching is an effective strategy to find the corre-
spondences among different ontologies in a scalable and a heterogeneous
semantic web. In order to find these correspondences, a matching system
should be built aiming to ensure the interoperability between ontologies.
POMap (Pairwise Ontology Mapping) is an automated ontology match-
ing system dealing with the three main types of heterogeneity: syntactic,
semantic and structural. During our first participation in the OAEI cam-
paign, POMap succeeded to be one of the top three performing systems
in the Anatomy track. In the remaining of this paper, we briefly intro-
duce POMap and discuss its OAEI 2017 results according to four tracks:
Anatomy, Conference, Large Biomedical Ontologies, Disease and Pheno-
type.
Keywords: Semantic web, ontology matching, semantic matching, syn-
tactic matching, structural matching
1 Presentation of the system
1.1 State, purpose, general statement
An ontology can model a particular domain as well as the semantic relation-
ships between its entities in order to ensure its reuse by different stakeholders.
Several ontologies describing the similar domain can be generated and used by
various parties defined by different terminologies. Despite the standardization
of the ontology representation, the heterogeneity problem emerges. Therefore,
it is important to overcome this heterogeneity to ensure the reusability of var-
ious ontologies. Indeed, many researchers has been proposing and developing
many automated ontology matching systems. Ontology matching is the process
of finding a set of correspondences between the entities of two or more ontologies
representing a similar domain. Therefore, these systems are using a variety of
strategies relying on the combination of several techniques such as: Syntactic,
semantic and structural based strategies. As depicted in figure 1, POMap is pur-
suing a sequential composition during the mentioned three matching techniques.
POMap is exploring all these three techniques in order to ensure a high quality
matching. Only dealing with the anatomy track, we employ a semantic matcher.
Then, for all the other OAEI tracks, we used a syntactic matcher, which follows
an all-against-all strategy. Next, our structural matcher takes as an input the
generated mappings from the semantic matcher and the syntactic matcher in
order to find new correspondences. The adopted sequential composition aims to
prune the search space used by the structural matcher. This structural matcher
is composed of two structural sub-matchers: siblings and subclasses. A broader
explanation of POMap could be found in [1]. In the next subsection, we will
briefly describe each component of our system as well as the used techniques.
1.2 Specific techniques used
The POMap workflow for our first participation on the OAEI comprises three
main steps, as flagged by the figure 1: Ontology indexing and loading, ontology
matching and output alignment generation.
Fig. 1. The architecture of POMap.
Step 1: Ontlogy indexing and loading
The initial step of POMap is the extraction of all the annotations within the
two input ontologies. In terms of lexical indexing, POMap builts a multimap
data structure that contains the triplet: the set of entities, their annotations
as well as the property type of each annotation. For the structural indexing,
all relationships between the extracted entities are stored in a multimap data
structure. Every record of this multimap contains two entities and the relation-
ship property between them. After accomplishing the lexical and the structural
indexing, we perform several preprocessing strategies, such as: the removal of
non-alphanumeric characters, the removal of stopwords, the stemming process
and the lowercasing.
Step 2: Ontology matching
Step 2.1: The semantic Matcher
The first step in the matching process is performing the semantic matcher.
We argue this choice by the high precision of the adopted semantic matcher.
Therefore, we will be based on it to enrich the resulted mappings by new ones
through the use of syntactic and structural strategies. During this first partici-
pation in the OAEI campaign, we adopted the semantic matching only for the
Anatomy track. We plan to expand the use of this matcher in our future partici-
pation. In order to ensure the semantic matching, we employed Uberon [3] as an
external biomedical knowledge source for the alignment of the Anatomy track.
Uberon is an integrated cross-species ontology covering anatomical structures
and includes relationships to taxon-specific anatomical ontologies. Indeed, we
explored the property ”hasDbXref”, which is mentioned in almost every class of
Uberon. This property references the classes’ URI of some external ontologies
such as the human and mouse of the Anatomy track. Consequently, we align
every two entities of the Anatomy track in case if they are both referenced in a
single class of Uberon.
Step 2.2: The syntactic Matcher
After performing the semantic matching process, we are able now to apply the
syntactic matcher. This syntactic matcher computes the similarity score between
every two names of the two input ontologies using a string similarity measure.
The variety of the existing state of the art similarity measure arises the problem
of choosing the right one associated with its optimal threshold. Therefore, we
tested the available syntactic similarity measure (https://goo.gl/1kUgkH) while
variating the associated threshold value. Hence, we selected ISUB combined with
a threshold of 0.9. Only the couple of entities having a similarity score above
0.9 are considered as new mappings candidates. As we are performing a pairwise
(1:1) matching process, for every single entity from the first ontology, we select
only one entity with the maximum similarity score. In case of two candidate
mappings have the exactly same similarity score, we consider randomly one of
them as the final alignment.
Step 2.3: The structural Matcher
For the set of available correspondences derived from the semantic and the
syntactic matcher, we are able to enrich them by a set of new correspondences
through the use of the structural matching. This structural matcher is composed
of two sub-matchers based on siblings and subclasses.
Step 2.3.1: The structural Matcher based on siblings
For the structural matcher based on siblings, we follow the intuition of: if two
entities match, then their sibling should somehow similar [2]. Therefore, if two
entities are aligned using the syntactic matcher, we compute the similarity score
between their siblings. Then, following an alignment multiplicity of 1:1, we match
the siblings having a similarity score between ISUB 0.9 (syntactic threshold) and
ISUB 0.8. The resulted mappings from the structural matcher based on siblings
are added to the already discovered correspondences by the two earlier matchers.
Step 2.3.2: The structural Matcher based on subclasses
Concerning the structural matcher based on subclasses, we pursue the in-
tuition that if two classes are similar, then their subclasses should be similar
[2]. This intuition should be straightforward applied if two classes are having a
very small number of subclasses. Nonetheless, this will be complicated in case of
there are many descendants. Therefore, as a first step, we remove all the com-
mon tokens between an already aligned entity and its descendants. We argue
that there is a syntactic inheritance between an entity an their descendants.
Therefore, the removal of these similar tokens, will permits to better capture the
similarity between two entities. Then, we compute the similarity score among
all the descendants of two already aligned entities while applying the similar-
ity measure of Monge Elkan 0.85 [4]. Unlike ISUB, we argue the use of Monge
Elkan due to its particularity in capturing the dissimilarity between two textual
sequences containing numerical values. However, this similarity measure is not
recommended for a heavy matching process, due to its time consuming.
Step 3: Output alignment generation
As a final step, we generate an RDF file, which contains the alignment based
on the resulted mappings resulted by all the employed matchers.
1.3 Link to the system and parameters file
The SEALS wrapped version of POMap for the OAEI 2017 is available at:
https://goo.gl/mZ4PzR
1.4 Link to the set of provided alignments
The resulted alignments by POMap as well as the results for each track during
our participation in OAEI 2017 are available at: https://goo.gl/mZ4PzR.
2 Results
2.1 Anatomy
The Anatomy track consists of finding the alignments between the Adult Mouse
Anatomy and the NCI Thesaurus describing the human anatomy. The evalua-
tion was run on a server coupled with 3.46 GHz (6 cores) and 8GB of RAM.
Table 1 draws the performance of POMap compared to the five top matching
systems. Our matching system achieved the third best result for this dataset
with an F-measure of 93.3%, which is very close to the top results. We argue
the importance of the obtained results by the effectivenesses of the overall em-
ployed matchers, the use of all the names of the input ontologies and applying
an efficient preprocessing process. The remaining challenge is to speed up the
execution time by applying more optimizations. We also target the improvement
of precision value for our next participation in the OAEI.
Table 1. POMap results in the anatomy track compared to the OAEI 2017 systems.
System Precision Recall F-Measure Runtime
AML 0.95 0.936 .943 47
YAM-BIO 0.948 0.922 0.935 70
POMap 0.94 0.925 0.933 808
LogMapBio 0.889 0.899 0.894 820
XMap 0.926 .836 .893 37
2.2 Conference
The purpose of the conference track is to find the correspondences within a col-
lection of ontologies describing the domain of organizing conferences. Matching
systems are evaluated according to the combination of three reference alignments
along with three evaluation modalities (M1,M2 and M3). These evaluation mod-
ularities are containing respectively: only classes, properties as well as classes
and properties. Since we did not focus on the matching of properties, the table
2 draws the obtained results by POMap results only for the first modularity
and partially for the third modularity. Therefore, we plan for our next partici-
pation in the OAEI to include the property matching in order to make a more
comprehensive evaluation of this track.
2.3 Large biomedical ontologies
This tracks aims to find the alignment between three large ontologies: Foun-
dational Model of Anatomy (FMA), SNOMED CT, and the National Cancer
Institute Thesaurus (NCI). Among six matching tasks between these three on-
tologies, POMap succeeded to perform the matching between FMA-NCI (small
Table 2. POMap results for the conference track
Precision Recall F1-Measure
Ra1-M1 0.88 0.47 .61
Ra1-M3 0.73 0.4 0.52
Ra2-M1 0.83 0.43 0.57
Ra2-M3 0.67 .37 .48
Ra2-M1 0.889 0.899 0.894
Ra2-M3 0.69 0.38 0.49
fragments) and FMA-SNOMED (small fragments) with an F-Measure respec-
tively of 86.1% and 41.6%. For the other tasks of the large biomedical track,
POMap exceeded the defined timeout. As a future work, we are planning to
cope with the matching process of the larger ontologies in a shorter time.
2.4 Disease and Phenotype
This track is based on a real use case in order to find alignments between disease
and phenotype ontologies. Specifically, the selected ontologies are the Human
Phenotype Ontology (HPO), the Mammalian Phenotype Ontology (MP), the
Human Disease Ontology (DOID) and the Orphanet and Rare Diseases Ontol-
ogy(ORDO). The evaluation was run on an Ubuntu Laptop with an Intel Core
i7-4600U CPU @ 2.10GHz x 4 coupled with 15Gb RAM. Due to the timeout
limit, POMap succeeded to complete tow tasks (HP-MP and DOID-ORDO) out
the four tasks of this track. POMap produced 2024 mappings in the HP-MP
task associated with 402 unique mappings. Among twelve matching systems,
POMap achieved the fifth highest F-measure according to the 2-vote silver stan-
dard, with an F-Measure of 73.2%. In the DOID-ORDO task, POMap generated
3222 mappings with 666 unique ones. According to the 2-vote silver standard, it
scored an F-Measure of 80.5%.
3 Conclusion
The first version of POMap ontology matching system as well as its obtained
results in the OAEI campaign were presented in this paper. We proposed three
matchers: semantic, syntactic and structural. We performed the structural match-
ing without any propagation syntactic similarity score or computation of a struc-
tural similarity score. We are guided only by the syntactic treatment of both
subclasses and siblings. The obtained results are promising especially for disease
and phenotype as well as the anatomy track in which we ranked as the third top
performing matching system. However, we did not opt to match larger ontologies
in the given runtime threshold. Consequently, we are planning to optimize our
matching system for larger biomedical tasks while taking into consideration the
automatic tuning of the matching configuration.
References
1. A. Laadhar, F. Ghozzi, I. Megdiche, F. Ravat, O. Teste, F. Gargouri POMap:
An Effective Pairwise Ontology Matching System 9th International Joint Confer-
ence on Knowledge Discovery, Knowledge Engineering and Knowledge Management
(KEOD’17), Funchal (Madeira, Portugal) 2017
2. Shvaiko, P., Euzenat, J. (2013). Ontology matching: state of the art and future
challenges. IEEE Transactions on knowledge and data engineering, 25(1),
3. Mungall, Christopher J., et al. ”Uberon, an integrative multi-species anatomy on-
tology.” Genome biology 13.1 (2012): R5.
4. Monge, Alvaro E., and Charles Elkan. ”The Field Matching Problem: Algorithms
and Applications.” KDD. 1996.