Combinatorial feature embedding based on CNN and LSTM for biomedical named entity recognition.

Area of Science:

• Computational Biology
• Bioinformatics
• Natural Language Processing

Background:

• Biomedical Named Entity Recognition (NER) is crucial for extracting information from large biomedical datasets.
• Existing NER models struggle with novel and evolving biomedical entity types.
• Deep learning approaches show promise but require improved feature representation in embedding layers.

Purpose of the Study:

• To propose a novel deep learning NER model for enhanced biomedical information extraction.
• To improve the representation of biomedical word tokens using combinatorial feature embedding.
• To address limitations in handling new entity types and long-term dependencies in existing models.

Main Methods:

• Developed a deep learning model based on Bidirectional Long Short-Term Memory (bi-LSTM) and Conditional Random Field (CRF).
• Integrated character-level representations from Convolutional Neural Network (CNN) and bi-LSTM for enhanced feature embedding.
• Incorporated an attention mechanism to improve focus on relevant tokens and mitigate LSTM long-term dependency issues.

Main Results:

• The proposed model achieved an F1-score of 86.93% on the NCBI-Disease dataset, outperforming existing models.
• Demonstrated competitive performance on the JNLPBA dataset with an F1-score of 75.31%.
• The combinatorial feature embedding and attention mechanism effectively improved entity recognition.

Conclusions:

• The proposed deep learning NER model offers a significant advancement in biomedical information extraction.
• The model's architecture effectively handles the complexity and growth of biomedical data.
• This approach provides a more accurate and robust solution for identifying biomedical entities.