A neural classification method for supporting the creation of BioVerbNet.

Area of Science:

• Computational linguistics
• Biomedical informatics

Background:

• VerbNet is a computational verb lexicon crucial for Natural Language Processing (NLP) tasks.
• Biomedical text processing requires a similar specialized verb resource.
• BioVerbNet aims to be a VerbNet tailored for biomedical verbs.

Purpose of the Study:

• To develop BioVerbNet, a verb lexicon for the biomedical domain.
• To overcome the time-consuming nature of manual verb classification.
• To leverage state-of-the-art neural representation models for automated expansion.

Main Methods:

• Started with a small, manually classified set of biomedical verbs.
• Applied a neural representation model for class-based optimization.
• Utilized PubMed abstracts and PubMed Central Open Access subset for data.

Main Results:

• The automatically expanded classification showed promising results against BioSimVerb.
• Human validation confirmed high accuracy by linguists and biologists.
• The method successfully included novel verbs and classes, facilitating cost-effective development.

Conclusions:

• This study is the first to apply neural representation learning to biomedical verb classification.
• The developed automatic classification method is efficient and accurate.
• The released classification can readily support biomedical application tasks.