DeepEventMine

DeepEventMine extracts nested and overlapping biomedical events from raw literature using an end-to-end neural architecture to identify triggers, entities, roles, and modifications for downstream knowledge discovery.

Key Features:

• End-to-End Neural Architecture: Operates on raw text without reliance on external syntactic tools.
• Nested Event Extraction: Detects nested and overlapping event structures within sentences and captures complex relationships between entities and actions.
• Directed Acyclic Graph (DAG) Representations: Constructs multiple DAG representations of events within a single sentence to represent nested event structures.
• BERT-based Transformer Foundation: Leverages BERT to detect nested entities, triggers, roles, and modifications in an end-to-end manner.
• Performance on Benchmark Tasks: Demonstrated superior performance across seven biomedical nested event extraction tasks.
• Entity-independent Event Detection: Capable of detecting events even when gold-standard entities are not provided.

Scientific Applications:

• Biomedical Knowledge Discovery: Extracts detailed event structures from research articles to support identification of complex biological processes and interactions.
• Data Mining and Integration: Mines large-scale biomedical text for nested events that can be integrated into knowledge bases or used for downstream analyses.
• Drug Discovery and Development: Provides extracted event information about molecular mechanisms and pathways relevant to identifying potential therapeutic targets.

Methodology:

Processes raw text through a neural network architecture built on top of BERT, constructs DAG structures to represent nested events, operates without external syntactic tools, and can detect events without gold-standard entities.