More Than the Sum of Its Parts – Holistic Ontology Alignment by Population-Based Optimisation

Ontology alignment is a key challenge to allow for interoperability between heterogeneous semantic data sources. Today, most algorithms extract an alignment from a matrix of the pairwise similarities of ontological entities of two ontologies. However, this standard approach has severe disadvantages regarding scalability and is incapable of accounting for global alignment quality criteria that go beyond the aggregation of independent pairwise correspondence evaluations. This paper considers the ontology alignment problem as an optimisation problem that can be addressed using nature-inspired population-based optimisation heuristics. This allows for the deployment of an objective function which can be freely defined to take into account individual correspondence evaluations as well as global alignment constraints. Moreover, such algorithms can easily be parallelised and show anytime behaviour due to their iterative nature. The paper generalises an existing approach to the alignment problem based on discrete particle swarm optimisation, and presents a novel implementation based on evolutionary programming. First experimental results demonstrate feasibility and scalability of the presented approaches.