Enhancing biomedical named entity recognition with parallel boundary detection and category classification

Area of Science:

• Natural Language Processing
• Bioinformatics
• Computational Biology

Background:

• Biomedical Named Entity Recognition (BioNER) is crucial for advanced applications.
• BioNER faces challenges due to nested structures and category correlations in entities.
• Existing models struggle to balance nested structure handling and category knowledge integration.

Purpose of the Study:

• To present a novel parallel BioNER model, BEAN.
• To address the unique properties of biomedical entities.
• To balance nested structures and category correlations in BioNER.

Main Methods:

• Developed a parallel BioNER model named BEAN.
• Utilized a triaffine model exploiting head, tail, and contextualized features for boundary detection.
• Introduced a multi-label classification model for category extraction without boundary guidance.

Main Results:

• BEAN achieves state-of-the-art performance on five public NER datasets, including four biomedical datasets.
• Demonstrated effectiveness in handling nested structures and category correlations.
• Showcased balanced performance between entity boundary detection and category classification.

Conclusions:

• BEAN is the first BioNER model to handle nested structures and category correlations in parallel.
• The model effectively detects entity boundaries and classifies categories.
• BEAN offers an efficient approach to BioNER, advancing the field.