Biomedical named entity recognition based on fusion multi-features embedding.

Area of Science:

• Biomedical informatics
• Natural Language Processing

Background:

• The growing volume of biomedical literature necessitates advanced text mining.
• Named entity recognition is crucial for applications like knowledge graphs and medical Q&A systems.

Purpose of the Study:

• To effectively recognize biomedical entities by integrating diverse features.
• To leverage multi-feature embeddings for enhanced prediction accuracy.

Main Methods:

• Generated deep contextual word-level, local character-level, and part-of-speech features.
• Utilized BiLSTM for dependency information and CRF for optimal tagging sequences.

Main Results:

• The proposed model surpassed state-of-the-art methods across six datasets.
• Achieved superior performance in identifying four types of biomedical entities.

Conclusions:

• Fusing multi-feature embeddings positively impacts named entity recognition.
• The method enhances semantic understanding by considering multiple relevant factors.