Ontology matching is an e ective strategy to nd the correspondences among di erent ontologies in a scalable and a heterogeneous semantic web. In order to nd these correspondences, a matching system should be built aiming to ensure the interoperability between ontologies. POMap (Pairwise Ontology Mapping) is an automated ontology matching system dealing with the three main types of heterogeneity: syntactic, semantic and structural. During our rst participation in the OAEI campaign, POMap succeeded to be one of the top three performing systems in the Anatomy track. In the remaining of this paper, we brie y introduce POMap and discuss its OAEI 2017 results according to four tracks: Anatomy, Conference, Large Biomedical Ontologies, Disease and Phenotype.
1.1
An ontology can model a particular domain as well as the semantic
relationships between its entities in order to ensure its reuse by di erent stakeholders.
Several ontologies describing the similar domain can be generated and used by
various parties de ned by di erent 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
various ontologies. Indeed, many researchers has been proposing and developing
many automated ontology matching systems. Ontology matching is the process
of nding 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 gure 1, POMap is
pursuing 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 nd 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 [
The POMap work ow for our rst participation on the OAEI comprises three main steps, as agged by the gure 1: Ontology indexing and loading, ontology matching and output alignment generation.
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 relationship 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.
The rst 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 rst
participation 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
participation. In order to ensure the semantic matching, we employed Uberon [
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 rst 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 nal alignment.
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.
For the structural matcher based on siblings, we follow the intuition of: if two
entities match, then their sibling should somehow similar [
Concerning the structural matcher based on subclasses, we pursue the
intuition that if two classes are similar, then their subclasses should be similar
[
As a nal step, we generate an RDF le, which contains the alignment based on the resulted mappings resulted by all the employed matchers. 1.3
The SEALS wrapped version of POMap for the OAEI 2017 is available at: https://goo.gl/mZ4PzR 1.4
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.
The Anatomy track consists of nding the alignments between the Adult Mouse Anatomy and the NCI Thesaurus describing the human anatomy. The evaluation 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 ve 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 e ectivenesses of the overall employed matchers, the use of all the names of the input ontologies and applying an e cient 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. The purpose of the conference track is to nd the correspondences within a collection 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 modularities 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 rst modularity and partially for the third modularity. Therefore, we plan for our next participation in the OAEI to include the property matching in order to make a more comprehensive evaluation of this track. This tracks aims to nd the alignment between three large ontologies: Foundational Model of Anatomy (FMA), SNOMED CT, and the National Cancer Institute Thesaurus (NCI). Among six matching tasks between these three ontologies, POMap succeeded to perform the matching between FMA-NCI (small fragments) and FMA-SNOMED (small fragments) with an F-Measure respectively of 86.1% and 41.6%. For the other tasks of the large biomedical track, POMap exceeded the de ned timeout. As a future work, we are planning to cope with the matching process of the larger ontologies in a shorter time. This track is based on a real use case in order to nd alignments between disease and phenotype ontologies. Speci cally, 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 Ontology(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 fth highest F-measure according to the 2-vote silver standard, 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
The rst 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 matching without any propagation syntactic similarity score or computation of a structural 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 con guration.