EvntExtrc

EvntExtrc extracts biomedical events from scientific literature to enable automated identification and fusion of molecular- and cellular-level interactions in molecular biology and biomedical research.

Key Features:

• Context Awareness and Embedding: Employs bidirectional long short-term memory (Bi-LSTM) networks to model contextual relationships between recognized arguments for accurate event detection.
• Bottom-Up Detection Framework: Identifies events from pre-recognized arguments to provide a scalable and generalizable computational model for biomedical knowledge management tasks.
• Compositional Attribute Utilization: Leverages compositional attributes to derive candidate samples that enhance the training of event classifiers and improve detection robustness.
• Evaluation on BioNLP Shared Task datasets: Evaluated on BioNLPST-BGI and BioNLPST-BB with average F-scores of 0.81 and 0.92 respectively, with the BioNLPST-BB result reported as 0.92 compared to a previous best of 0.56.

Scientific Applications:

• Biomedical Knowledge Management: Facilitates the integration and fusion of knowledge extracted from diverse biomedical literature sources.
• Literature Mining: Enables large-scale mining of scientific literature to identify relevant biological interactions and events.

Methodology:

Uses a bottom-up detection framework starting from recognition of arguments; models argument context embeddings with a Bi-LSTM network; and derives candidate training samples via compositional attribute utilization for event classifiers.