Fusion framework: Conditional-aware one-stage nested event extraction model.

Area of Science:

• Biomedical Natural Language Processing
• Computational Biology
• Bioinformatics

Background:

• Biomedical event extraction is crucial for understanding biological processes from text.
• Existing models struggle with complex event structures like overlapping and nested events.
• Accurate identification of event triggers, arguments, and roles is challenging.

Purpose of the Study:

• To introduce CA-NEE, a novel one-stage model for biomedical event extraction.
• To address limitations in handling overlapping and nested biomedical events.
• To improve the accuracy of trigger and argument classification in complex scenarios.

Main Methods:

• Developed a Conditional-Aware one-stage model (CA-NEE).
• Integrated an event-type-aware conditioning mechanism with token-pair relation modeling.
• Employed Conditional Layer Normalization (CLN) for dynamic token representation adaptation.
• Utilized a parallel word-pair scorer for simultaneous span and role prediction.

Main Results:

• CA-NEE demonstrated consistent performance gains in Trigger Classification (TC) and Argument Classification (AC).
• Significant improvements were observed on complex overlapping and nested event structures.
• Evaluations on GENIA11 and GENIA13 datasets confirmed the model's effectiveness over baselines.

Conclusions:

• CA-NEE provides an effective and efficient solution for biomedical event extraction.
• The model's architecture successfully handles intricate event structures.
• This approach advances the field of automated biomedical information extraction.