Since different sensor ontologies are developed independently and for different requirements, a concept in one sensor ontology could be described with different terminologies or in different context in another sensor ontology, which leads to the ontology heterogeneity problem. To bridge the semantic gap between the sensor ontologies, authors propose a semi-automatic sensor ontology matching technique based on an Interactive MOEA (IMOEA), which can utilize the user's knowledge to direct MOEA's search direction. In particular, authors construct a new multi-objective optimal model for the sensor ontology matching problem, and design an IMOEA with t-dominance rule to solve the sensor ontology matching problem. In experiments, the benchmark track and anatomy track from the Ontology Alignment Evaluation Initiative (OAEI) and two pairs of real sensor ontologies are used to test performance of the authors' proposal. The experimental results show the effectiveness of the approach.
Top1. Introduction
A sensor ontology provides a formal specification on the sensor concepts and their relationships, which is a state-of-the-art technique for addressing the data heterogeneity issue in the semantic sensor web (Xue et al., 2019c). In particular, a sensor ontology can be defined as 3-tuple (C,P,R), where C, P, and R are respectively the set of classes, properties and relationships, which can describe the sensors' capabilities, performance, and usage conditions that allow the discovery of different data depending on the purpose and context (Xue et al., 2019d). Since different sensor ontologies are developed independently and for different requirements, a concept in one sensor ontology could be described with different terminologies or in different context in another sensor ontology, which raises the heterogeneity problem to a higher level. Ontology matching is an effective technique to solve the sensor ontology heterogeneity problem by determining the semantically identical entities in heterogeneous sensor ontologies (Xue, & Wang, 2016). The obtained sensor ontology alignment A is a correspondence set, and each correspondence inside is a 4-tuple 
, where e and 
are the entities of two sensor ontologies. n∈[0,1] is a confidence value holding for the correspondence between e and 
, and r is the relationship between e and 
, which refers to equivalence in this work.
Due to the complex nature of the ontology matching process, Evolutionary Algorithm (EA) has become a state-of-the-art methodology for matching the ontologies. However, there existing different aspects of solution that are partially or wholly in conflict, and the single-objective EA may lead to unwanted bias to one of them and reduce the solution's quality. Multi-objective EA (MOEA) can estimate different aspect of solutions simultaneously, and produce a set of solutions which contains a number of non-dominated solutions, none of which can be further improved on any one objective without degrading it in another. MOEA based ontology matching technique is a recently introduced, innovative, and efficient methodology to address the ontology matching problem (Acampora et al., 2014). However, due to the complexity of the ontology matching process, ontology alignments generated by an automatic matcher should be checked by users to ensure their quality (Shvaiko & Euzenat, 2013). To improve the quality of sensor ontology alignment, in this work, we propose an Interactive MOEA (IMOEA)-based semi-automatic ontology matching technique. In particular, our contributions are as follows:
- 1.
A semi-automatic ontology matching framework is proposed for matching sensor ontologies;
- 2.
A multi-objective optimal model is constructed for the sensor ontology matching problem;
- 3.
A t-dominance rule is proposed and a problem-specific IMOEA is designed to effectively solve the sensor ontology matching problem.
The rest of the paper is organized as follows. Section 2 describes related works. Section 3 defines the ontology matching problem. Section 4 provides the details of IMOEA. Section 5 shows the experimental results, and section 6 relates our conclusions.