Deep learning methods for biomedical named entity recognition: a survey and qualitative comparison.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• The exponential growth of biomedical literature necessitates efficient information extraction.
• Biomedical Named Entity Recognition (BioNE) is crucial for understanding and utilizing this data.
• Accurate BioNE identification supports downstream biomedical text mining tasks.

Purpose of the Study:

• To comprehensively review deep learning-based methods for Biomedical Named Entity Recognition (BioNER).
• To classify existing deep learning approaches for BioNER.
• To discuss datasets, future trends, and opportunities in the field.

Main Methods:

• Categorization of deep learning methods into four groups: single neural network-based, multitask learning-based, transfer learning-based, and hybrid model-based.
• Analysis of datasets used for training and testing BioNER models.
• Review of the application of these methods across multiple biomedical domains.

Main Results:

• Deep learning methods have achieved state-of-the-art performance in BioNER.
• The effectiveness of BioNER methods is influenced by dataset size and type.
• Various deep learning architectures offer distinct advantages for BioNER tasks.

Conclusions:

• Deep learning approaches are pivotal for advancing BioNER.
• Future research should focus on optimizing methods based on dataset characteristics and exploring novel architectures.
• Continued development in BioNER is essential for unlocking insights from biomedical texts.