Matching ontologies with kernel principle component analysis and evolutionary algorithm

Abstract

Abstract Ontology serves as a structured knowledge representation that models domain-specific concepts, properties, and relationships. Ontology matching (OM) aims to identify similar entities across distinct ontologies, which is essential for enabling communication between them. At the heart of OM lies the similarity feature (SF), which measures the likeness of entities from different perspectives. Due to the intricate nature of entity diversity, no single SF can be universally effective in heterogeneous scenarios, which underscores the urgency to construct an SF with high discriminative power. However, the intricate interactions among SFs make the selection and combination of SFs an open challenge. To address this issue, this work proposes a novel kernel principle component analysis and evolutionary algorithm (EA) to automatically construct SF for OM. First, a two-stage framework is designed to optimize SF selection and combination, ensuring holistic SF construction. Second, a cosine similarity-driven kPCA is presented to capture intricate SF relationships, offering precise SF selection. Finally, to bolster the practical application of EA in the SF combination, a novel evaluation metric is developed to automatically guide the algorithm toward more reliable ontology alignments. In the experiment, our method is compared with the state-of-the-art OM methods in the Benchmark and Conference datasets provided by the ontology alignment evaluation initiative. The experimental results show its effectiveness in producing high-quality ontology alignments across various matching tasks, significantly outperforming the state-of-the-art matching methods.