Ontology Matching Based on BERT and Graph Contrastive Learning
Daoqu Geng, Jie Luo · 2025
Ontology matching is a prerequisite for enabling cross-domain data integration and semantic interoperability. Previous OM(Ontology Matching) methods either only consider the semantic information of ontologies, or the extracted structural features are too simple or need to be obtained from preprocessed data. This does not work well for ontology mapping tasks with high structural relevance. In this paper, we propose a new OM system which does not require any labeled data, generates rich structural feature information of nodes by graph contrastive learning model, fuses the semantic information generated by the BERT model to get the final vectorized representation of the ontology, and finally calculates the Euclidean distance between the concepts to get the similarity between the concepts to complete the ontology matching task. Our evaluation of two matching tasks for biomedical ontology and one matching task for conference ontology shows that our model can generally achieve higher F1 scores than the existing OM system