Context awareness and embedding for biomedical event extraction.

Area of Science:

• Computational biology
• Bioinformatics
• Biomedical Natural Language Processing (NLP)

Background:

• Biomedical event extraction is crucial for knowledge fusion in molecular biology and research.
• Existing models are limited by event context and single-task applicability.
• A general and scalable computational model is needed for effective biomedical knowledge management.

Purpose of the Study:

• To propose a novel bottom-up detection framework for biomedical event extraction.
• To develop a general and scalable computational model for biomedical knowledge management.
• To improve the accuracy and efficiency of identifying events from recognized arguments in biomedical literature.

Main Methods:

• A bottom-up detection framework identifying events from recognized arguments.
• Utilizing a bidirectional long short-term memory network to model argument context embeddings.
• Deriving candidate samples for training event classifiers based on compositional attributes.

Main Results:

• Achieved average F-scores of 0.81 on BioNLPST-BGI and 0.92 on BioNLPST-BB datasets.
• Nearly doubled the state-of-the-art F-score performance on the BioNLPST-BB dataset (0.92 vs. 0.56).
• Demonstrated the effectiveness of the proposed framework through case studies.

Conclusions:

• The proposed framework offers a significant advancement in biomedical event extraction.
• The method provides a general and scalable solution for biomedical knowledge management.
• The results highlight the potential for improved information extraction in molecular biology and biomedical research.