AIONER: all-in-one scheme-based biomedical named entity recognition using deep learning.

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Computational Biology

Background:

• Biomedical Named Entity Recognition (BioNER) is crucial for text mining but faces challenges.
• Manual data annotation is expensive and requires domain expertise.
• Data scarcity leads to overfitting and limited generalizability in current BioNER models.

Purpose of the Study:

• To develop a novel all-in-one (AIO) scheme to improve BioNER model accuracy and stability.
• To introduce AIONER, a general-purpose BioNER tool utilizing deep learning and the AIO schema.
• To address the limitations of single-entity type recognition and data scarcity in BioNER.

Main Methods:

• Proposed an all-in-one (AIO) scheme integrating external annotated resources.
• Developed AIONER, a deep learning-based BioNER tool implementing the AIO schema.
• Evaluated AIONER on 14 benchmark BioNER tasks and three independent tasks.

Main Results:

• AIONER demonstrated effectiveness and robustness across 14 BioNER benchmark tasks.
• AIONER outperformed state-of-the-art methods, including multi-task learning.
• AIONER successfully recognized novel entity types and processed large-scale biomedical text efficiently.

Conclusions:

• The AIO scheme and AIONER tool significantly enhance BioNER performance.
• AIONER offers a robust and generalizable solution for biomedical text mining.
• AIONER provides practical advantages for large-scale biomedical data processing.