Biomedical ontology alignment with BERT

Conference item

Existing machine learning-based ontology alignment systems often adopt complicated feature engineering or traditional non-contextual word embeddings. However, they are often outrun by the rule-based sys- tems despite the model complexity. This paper proposes a novel ontology alignment system based on a contextual embedding model named BERT, aiming to suficiently utilize the text semantics implied by ontologies. Our results on two biomedical alignment tasks demonstrate that, despite us- ing the to-be-aligned classes alone as the input, our system outperforms the leading systems: LogMap and AML.

Authors: He, Y; Chen, J; Antonyrajah, D; Horrocks, I

• Publication status: Published
• Peer review status: Peer reviewed
• Host title: Proceedings of the 16th International Workshop on Ontology Matching co-located with the 20th International Semantic Web Conference (ISWC 2021)
• Volume: 3063
• Pages: 1-12
• Publication date: 2021-03-01
• Event title: 16th International Workshop on Ontology Matching co-located with the 20th International Semantic Web Conference (ISWC 2021)
• Event location: Online
• Event website: http://om2021.ontologymatching.org/
• Event start date: 2021-10-25
• Event end date: 2021-10-25
• ISSN: 1613-0073
• Language: English
• Keywords: FFR