Contextual information contributes to biomedical named entity normalization.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Biomedical Named Entity Normalization (BNEN) is critical for mapping entities to standard identifiers.
• Existing BNEN methods often overlook contextual information, which is vital for disambiguating mentions.

Purpose of the Study:

• To propose a novel BNEN method, CIFSYN, that integrates contextual information fusion.
• To enhance the accuracy of biomedical named entity normalization by leveraging surrounding text.

Main Methods:

• CIFSYN builds upon the IA-BIOSYN framework.
• It comprehensively considers mention context by placing candidates within the same context.
• A contextual information fusion module captures relationships between mentions, candidates, and context.

Main Results:

• CIFSYN achieved high Acc@1 scores across five public datasets (e.g., 0.969 on BC5CDR-Chemical).
• The proposed method significantly outperformed existing state-of-the-art BNEN approaches.
• The context information module provided an average improvement of 0.5% in Acc@1.

Conclusions:

• Incorporating contextual information demonstrably improves biomedical named entity normalization performance.
• CIFSYN offers a more robust approach to BNEN by effectively utilizing contextual cues.