Article Preview
Top1. Introduction
The semantic web community relies on ontologies to overcome the crucial problem of semantic heterogeneity. However, these ontologies are themselves heterogeneous. This heterogeneity may occur at syntactical, terminological, conceptual and semiotic levels (Euzénat and Shvaiko, 2013). Ontology alignment, defined as the process of identification of semantic correspondences between entities of different ontologies to be aligned (Euzénat and Shvaiko, 2013), is prerequisite and necessary for the problem of semantic interoperability between different sources of distributed information.
Due to the size and the number of ontologies, ontology alignment cannot be done manually beyond a certain complexity. Automatic techniques or at least semi-automatic should be developed to reduce the burden of manual creation and maintenance of ontology alignment. However the automatic identification of correspondences between ontologies is very difficult due to their conceptual divergence (Ehrig, 2007). Several methods have been elaborated to semantically align their entities. These methods are generally based on the calculation of the similarity of their names, relationships and instances of concepts. These methods can be terminological, structural or extensional; and most of the systems combine these approaches (Euzénat and Shvaiko, 2013) (Ehrig, 2007).
Furthermore, ontology-alignment approaches can be split globally into two main categories (Euzénat and Shvaiko, 2013): reasoning-based approaches and the approaches based on the calculation of similarities.
In fact, when ontologies contain a lot of instances there is a very good opportunity to use the extensions of concepts in order to align them using instance-based techniques. In practice, two situations may arise in terms of sharing instances between ontologies to align (Rahm, 2011). When they have common instances, it is recommended to use metrics, such as Jaccard metric (Jaccard, 1901), which evaluates the overlap of instances between concepts, and these concepts are considered similar when their overlap is important (Euzénat and Shvaiko, 2013). In contrast, when ontologies do not share or share few instances, one generally proceeds to create a common artificial corpus of instances between concepts to align (Schopman et al., 2012).
The objective of this paper is to present a new approach for ontology alignment in order to find new semantic correspondences between entities of the two ontologies to align. The approach that we describe in this paper fits into the category of instance-based ontology-alignment methods when the ontologies to align do not have many common instances i.e. do not have a common corpus of instances, which makes the task of identifying the semantic relation between concepts very difficult. More precisely, we propose a new technique for creating such corpus which consists of combining reasoning-based and instance-based approaches.
The intuition behind our approach is to create a shared inferred taxonomy by the two ontologies to align, which serves as support for the creation of a common corpus of instances. The principle is to first create this common corpus of instances, then apply the instance-based metrics for calculating similarities.
The creation of the corpus occurs in three steps. The first step consists of merging the two ontologies to be aligned, by exploiting an initial alignment obtained by a domain expert or by existing ontology-alignment methods. The second step consists of generating the taxonomy inferred from the ontology resulting from the previous step. The last step consists of exploiting the new subsumption relations identified to operate a mutual transfer of instances between the concepts of two ontologies. Finally, the similarities between the concepts are calculated by instance-based metric and correspondences identified.
The rest of the paper is organized as follows. First, an illustrative example is presented in Section 2, and related work is presented in Section 3 on existing approaches for reasoning-based and instance-based ontology alignment. Section 4 describes our contribution and Section 5 presents the aspects related to the implementation of our system. We provide the evaluation results with a detailed analysis in Section 6, in order to show the efficiency of our approach. Section 7 contains concluding remarks and sets directions for future work.