Core Concepts
Proposing GraphERE for enhanced event relation extraction through graph-based embeddings and dynamic graphs.
Abstract
The content introduces GraphERE, a framework for joint event-event relation extraction. It addresses the limitations of current methods by incorporating event argument and structure features using static AMR and IE graphs. The model utilizes Node Transformer and Task-specific Dynamic Event Graphs to extract multiple event relations simultaneously. Experimental results on the MAVEN-ERE dataset demonstrate GraphERE's superior performance over existing methods.
Directory:
Abstract
Events describe entity state changes.
Multiple events in a document connected by various relations.
Current ERE works face challenges in representing event features and interconnections between relations.
Introduction
ERE is crucial for identifying semantic relationships among events.
Various types of event relations exist in real-world scenarios.
Data Extraction Methods
"Experimental results on the latest MAVEN-ERE dataset validate that GraphERE significantly outperforms existing methods."
Related Work
Previous research focuses on specific types of event relations.
Approach
Task Formulation: Document-level Event-Event Relation Extraction task defined.
Experiments
Evaluation conducted on the MAVEN-ERE dataset with comparisons to baselines.
Ablation Analysis
Removal of components affects model performance, highlighting their importance.
Analysis for IE Graph and AMR Graph
Impact of IE and AMR graphs on model performance analyzed.
Data Scale Analysis
Model performance analyzed with varying data scales.
Stats
実験結果は、最新のMAVEN-EREデータセットで、GraphEREが既存の手法を大幅に上回ることを検証しています。