Joint event extraction model based on dynamic attention matching and graph attention networks.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Knowledge Representation

Background:

• Event extraction is crucial for building event knowledge graphs.
• Current models struggle with vague/unfamiliar event triggers and text noise.

Purpose of the Study:

• To propose a joint event extraction model addressing challenges in trigger word classification and noise reduction.
• To enhance the accuracy and robustness of event extraction for knowledge graph construction.

Main Methods:

• A joint event extraction model integrating dynamic attention matching and graph attention networks.
• Dynamic attention identifies event nodes and integrates event structure knowledge for pattern subgraph construction.
• Graph attention network aggregates node neighbor embeddings, distinguishing semantic and structural information while mitigating noise.

Main Results:

• The proposed model demonstrates competitive performance on the ACE2005 dataset.
• Successfully addresses challenges of ambiguous trigger words and text noise in event extraction.

Conclusions:

• The novel model effectively improves event extraction accuracy and robustness.
• Dynamic attention and graph attention networks offer a promising approach for event knowledge graph construction.