Ontology Alignment using Biologically-inspired Optimisation Algorithms

Ontologies describe real-world entities in terms of axioms, i.e. statements about them, and have become an established instrument for formally modelling and representing knowledge. The diversity of available ontologies results in a heterogeneous landscape where ontologies can overlap in their content. Such an overlap can be caused by ontologies modelling the same or similar domains created by different ontology designers, or with different views of a domain. If overlapping ontologies are to be used in a semantic application, sophisticated methods are required to overcome this heterogeneity. Identifying the overlap of ontologies is tackled by the discipline of ontology alignment. An alignment between two ontologies denotes a set of correspondences between ontological entities. In this book, the ontology alignment problem is considered an optimisation problem. Thereby, optimality is defined in terms of an objective function that evaluates candidate alignments according to ontology modellingand domain-specific criteria, e.g. significance and similarity of entity identifiers, or logical implications of an alignment. This optimisation problem is solved using biologicallyinspired optimisation techniques, exemplary demonstrated by a novel Evolutionary Algorithm and an adapted Discrete Particle Swarm Optimisation algorithm. The Evolutionary Algorithm implements concepts from Evolutionary Programming and Extremal Optimisation and operates on a newly developed data structure for representing alignments. The Discrete Particle Swarm Optimisation algorithm extends an existing algorithm for a structurally similar problem. The presented approach is the first to systematically apply biologicallyinspired optimisation algorithms to the problem of ontology alignment. These algorithms have several advantages, which address relevant issues of the ontology alignment problem: First, the inherent parallelisability of biologically-inspired optimisation techniques enables the exploitation of distributed computing environments, such as cloud infrastructures. This improves on scalability aspects of the alignment task. Second, biologically-inspired optimisation algorithms are metaheuristics,