Biomedical event argument detection method based on multi-feature fusion and question-answer paradigm.

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Computational Biology

Background:

• The rapid expansion of biomedical text data presents challenges for information extraction.
• Existing event argument detection methods struggle with irrelevant information and deep semantic understanding.
• Extracting multiple events from complex biomedical texts remains difficult.

Purpose of the Study:

• To develop an advanced event argument detection method for biomedical texts.
• To enhance the accuracy of mining valuable information from complex biomedical data.
• To address limitations in current methods for irrelevant argument interference and semantic association.

Main Methods:

• A novel event argument detection method utilizing multi-feature fusion and a question-answer paradigm.
• Splitting events into question-answer formats to simplify detection complexity.
• Employing syntactic distance and prior knowledge to identify argument templates, reducing irrelevant argument interference.
• Integrating a multi-feature attention mechanism to capture deep semantic features.
• Utilizing post-processing for predefined event structures to generate final biomedical events.

Main Results:

• The proposed model achieved a 62.50% F1 score for event extraction on the MLEE dataset.
• This performance surpasses existing advanced event extraction methods.

Conclusions:

• The developed method demonstrates strong performance in biomedical event extraction.
• It effectively supports the mining of valuable information from biomedical texts.