Active Learning Pipeline to Identify Candidate Terms for a CDSS Ontology.

Area of Science:

• Biomedical Informatics
• Health Information Technology
• Computational Linguistics

Background:

• Ontology is crucial for interoperability in health information and IT.
• Manual ontology construction by human domain experts (HDE) is time-consuming.
• Existing methods face challenges in scalability and long-term maintenance.

Purpose of the Study:

• To explore an active learning approach for automated ontology term identification.
• To integrate automated term identification with manual verification for deep learning model training.
• To enhance the efficiency and sustainability of ontology development and maintenance.

Main Methods:

• Developed an active learning pipeline to identify candidate terms from biomedical publications.
• Incorporated a manual verification step for candidate terms.
• Utilized a deep learning model for term classification and ontology refinement.

Main Results:

• Demonstrated the feasibility of an active learning approach for ontology term extraction.
• Presented preliminary results showing the potential for automating parts of the ontology construction process.
• Highlighted the complementary nature of this approach to manual HDE efforts.

Conclusions:

• Active learning offers a promising strategy to accelerate ontology development.
• This approach can significantly reduce the burden of manual term identification.
• The proposed pipeline supports efficient long-term maintenance of biomedical ontologies.