This paper presents and discusses the results of the KEPLER system for the 2018 edition of the Ontology Alignment Evaluation Initiative (OAEI 2018). The implemented approach is based on the exploitation of three different strategies including Information Retrieval (IR) inspired algorithm for terminological based alignment computation. For scaling up, KEPLER implements a partitioning approach, while for the management of multilingualism, KEPLER develops a well-defined strategy based on the use of a translator and structural alignment computation. This is the second year of participation and the results are encouraging.
1.1
KEPLER exploits, besides classic techniques [7], an external resource, i.e., a translator in order to deal with multilingualism. 1.2
The main idea of KEPLER is to exploit the expressiveness of the OWL language to detect and compute the similarity between entities of two given ontologies through six complementary modules as presented in Figure 1.
Entities are described using OWL primitives with their semantics. An ontology is seen as a semantic graph where entities are nodes connected by links (the predicates). These links have specified semantics. The alignment workflow is detailed as follows.
Parsing and pretreatment: this module extracts the ontological entities initially represented by a primary form of lists. In other words, at the parsing stage, the main goal is to transform an OWL ontology in a well defined structure that preserves and highlight all the information contained in processed ontology. It has a significant impact on the results of the similarity computation thereafter. The result is a set of entities names and their associated descriptions.
Partitioning: KEPLER follows a divide and conquer strategy. Therefore, this module aims at splitting ontologies into smaller parts to support the alignment task [8]. Consequently, partitioning a set B(C) is to find subsets B1, B2,..., Bn, encompassing semantically close elements bound by a relevant set of relationships, i.e., O = SfB1; B2; :::; Bng, where Bi is an ontological block, and n is the resulting number of extracted blocks. Hence, we can define an ontological portion as a reduced ontology that could be extracted from another larger one by splitting up the latter according to its constituents : structures and semantics. One way to obtain such a partitioning is to maximize the relationships inside a block while minimizing the relationship between the blocks themselves. The resulting partitioning quality can be evaluated using different criteria: – The size of the generated blocks: that must have a reasonable size, i.e., a number of elements that could be handled by an alignment tool; – The number of the generated blocks: this number should be as small as possible to limit the number of block pairs to be aligned latter; – The compactness degree of a block: a block is said to be substantially compact if relations (lexical and structural ones) are stronger inside the block and lower outside it.
Translation : in order to deal with multilingualism, two alternatives are followed: i) either considering one of the languages of the input ontologies as a pivot, therefore translating the second one to this chosen pivot; ii) choosing a pivot language and translate the inputs ontologies to this pivot. Further to these alternatives, an external resource, i.e., WordNet3 is used. Therefore the pivot language used by KEPLER is the English language. The translation process is performed usinig the Microsoft Bing 4 translator.
Indexing : one of the issue in Ontology Alignment is the cost of computing the similarity between all the entities of the input ontologies. To deal with this issue, the indexing strategy is one of the novelties of our approach. It consists in reducing the search space through the use of techniques borrowed from the IR domain. An effective search strategy is implemented on top of the built indexes of the two input ontologies. To enable faster searching, the driving idea that was previously used in the ServOMap system [9] is to perform the analysis of the ontologies in advance and store it in an optimized format for the search.
Candidate Mappings Identification : the role of this module is to find the entities in common between the indexes. Once the indexes are set up, the querying step is activated. To do so, the querying strategy implemented satisfies both the terminology search and semantic aspects at once. Indeed, the task is querying documents in a vector space that contains a set of ontological entities and their synonyms obtained via WordNet for each Ontology. It is worthy to mention that indexes querying is done in both senses (each ontology plays successively the role of querying component).
Filtering and Recovery: the filtering module consists of two complementary submodules, each one is responsible of a specific task in order to refine the set of primarily identified candidates mappings. At this stage, once the list of candidates is ready, the alignment method uses a first filter. This filter eliminates the redundancy between these candidates by eliminating possible duplicates. In addition, there is always the concern about false positives. The second filter eliminates false positives candidates. This filter is applied to what is called partially redundant entities. An entity is considered as partially redundant if it belongs to two different mappings. Being given three ontological 3 https://wordnet.princeton.edu/ 4 https://www.bing.com/translator entities e1, e2 and e3, if on the one hand, e1 is aligned to e2, and secondly, e1 is aligned to e3, this last alignment is qualified as doubtful. As the KEPLER system generates (1 : 1) mappings, an entity cannot belongs to several mappings. Therefore, given the topology of two suspicious entities (e3 neighbors with e1 neighbors, e2 neighbors with e1 neighbors ) with respect to the redundant entity e1, the idea is to retain the couple having the highest topological proximity value. All candidates are subject to this filter before to generate the final alignment.
Alignment Generation : The result of the alignment process provides a set of mappings, which are serialized in the RDF format. 2
2.1
In this section, we present the results obtained by KEPLER system for the OAEI 2018 edition.
This track consists in two real world ontologies to be matched, the source ontology describing the Adult Mouse Anatomy (with 2744 classes) and the target ontology is the NCI Thesaurus describing the Human Anatomy (with 3304 classes). For this track, KEPLER succeeded to extract 74% of correct mappings with a precision of 95% and recall of 74%. KEPLER handled easyly the input ontologies of this track thanks to the partionning module Ontopart [10, 8] 2.2
The conference track consists of 15 ontologies from the conference organization domain and each ontology must be matched against every other ontologies. The dataset describes the domain of organizing conferences from different perspectives. Precision values for to evaluation settings are respectively 76% and 58%. Recall values are 48% and 68%. 2.3
The Multifram dataset is composed of a subset of the Conference track, translated in nine different languages (i.e., Chinese, Czech, Dutch, French, German, Portuguese, Russian, Spanish and Arabic). With a special focus on multilingualism, it is possible to evaluate and compare the performance of alignment approaches through these test cases. Based on several previous contributions [11–16], the designed main goal of the MultiFarm track is to evaluate the ability of the alignment systems to deal with multilingual ontologies. It serves the purpose of evaluating the strength and weakness of a given system across languages. In the different ontologies setting, KEPLER is ranked second with a recall value of 0:21 and a precision value of 0:40. Whereas in the same ontologies setting, it lasted at the first place with a recall value of 0:36, and a precision value of 0:85. 2.4
KEPLER succeeds in the best case, to obtain 27% of recall and a precision of 100%. 2.5
In the scalability register, this track consists in finding alignments between the
Foundational Model of Anatomy (FMA), SNOMED CT, and the National Cancer Institute
Thesaurus (NCI). These ontologies are semantically rich and contain tens of thousands
of classes. The Large BioMed Track consists of three matching problems, i.e., (
3
In the Phenotype track, the system succeeded in processing only the DOID-ORDO subcase by identifying 1824 matches for 1237 expected ones, [Precision : 0.86 / Recall : 0.59].
In this paper, we briefly described the alignment system KEPLER with comments of
the results obtained according to the OAEI 2018 tracks, corresponding to the SEALS
platform evaluation modality. Several observations regarding these results were
highlighted, in particular the impact of the elimination of any ontological resource on the
similarity values. KEPLER is an ongoing work which borrows its idea from two
previous systems, CLONA [15] and SERVOMAP [9]. It showed promising results for this
second participation. As future work, the idea is to support the instance based ontology
alignment in a wider range and contexts [17]. We have dealt with this issue before [18,
19], but the test base update imposes other challenges in terms of the used ontological
languages and the evolutive semantic description formalisms.
4. Kachroudi, M., Zghal, S., Ben Yahia, S.: Alignement d’ontologies en utilisant des ressources
externes : Linked data. In: Actes des 4èmes Journées Francophones sur les Ontologies
(JFO’2011), Montréal, Canada (2011) 259–264
5. Diallo, G.: Efficient building of local repository of distributed ontologies. In: Proceedings
of the Seventh International Conference on Signal-Image Technology and Internet-Based
Systems, SITIS 2011, Dijon, France, November 28 - December 1, 2011. (2011) 159–166
6. Dramé, K., Diallo, G., Delva, F., Dartigues, J., Mouillet, E., Salamon, R., Mougin, F.: Reuse
of termino-ontological resources and text corpora for building a multilingual domain
ontology: An application to alzheimer’s disease. Journal of Biomedical Informatics 48 (2014)
171–182
7. Euzenat, J., Mocan, A., Scharffe, F.: Ontology alignments: an ontology management
perspective. Ontology management: semantic web, semantic web services, and business
applications (2008)
8. Kachroudi, M., Zghal, S., Ben Yahia, S.: Ontopart: at the cross-roads of ontology partitioning
and scalable ontology alignment systems. International Journal of Metadata, Semantics and
Ontologies 8(
Semantics 5(44) doi:10.1186/2041-1480-5-44 (2014)
10. Kachroudi, M., Hassen, W., Zghal, S., Ben Yahia, S.: Large ontologies partitioning for
alignment techniques scaling. In: Proceedings of the 9th International Conference on Web
Information Systems and Technologies (WEBIST), 8-10 May, Aachen, Germany (2013) 165–168
11. Kachroudi, M., Ben Yahia, S., Zghal, S.: Damo - direct alignment for multilingual
ontologies. In: Proceedings of the 3rd International Conference on Knowledge Engineering and
Ontology Development (KEOD), 26-29 October, Paris, France (2011) 110–117
12. Kachroudi, M., Zghal, S., Ben Yahia, S.: When external linguistic resource supports
crosslingual ontology alignment. In: Proceedings of the 5th International Conference on Web and
Information Technologies (ICWIT 2013), 9-12, May, Hammamet, Tunisia (2013) 327–336
13. Kachroudi, M., Zghal, S., Ben Yahia, S.: Using linguistic resource for cross-lingual ontology
alignment. International Journal of Recent Contributions from Engineering 1(
In: Actes de la 6èmes Edition des Journées Francophones sur les Ontologies (JFO’2016), Bordeaux, France (2016) 69–74