Event Detection Based on Multi-hop Multi-channel Graph Neural Network

The goal of event detection is to identify instances of various event types in text. While graph neural networks based on dependency parse trees have shown promise for enhancing event detection, their effectiveness is still limited. One challenge is that dependency trees do not automatically create connections between trigger words and other key terms, which are essential for accurately recognizing and classifying triggers. Moreover, syntax-based methods generally focus on aggregating information from only the nearest neighbors in the dependency graph, despite the fact that relevant words are often located multiple hops away. In this paper, we introduce a novel approach, Multi-hop Multi-channel Graph Neural Network(MMGNN) for event detection. Experimental results confirm the effectiveness of our model in advancing performance on this task.