Direction-aware Edge Updating and Argument Combination-based Context Enhancement for Event Detection

Event detection aims to identify event trigger words in text and classify them into predefined event types. Existing methods still face significant challenges in modeling long-distance dependencies and handling overlapping and nested events. To address these issues, this paper proposes a context-enhanced event detection model that integrates direction-aware edge updating and argument combination to better representing complex events. Specifically, based on traditional graph convolutional networks, it introduces a direction-aware edge updating module to effectively capture asymmetric dependency paths and key information within syntactic structures. It also designs a context enhancement module based on argument combination, which fully leverages the sentence-level global context, models syntactic information corresponding to different event types, and handles overlapping and nested events. The proposed model achieves F₁ scores of 84.46% and 70.11% on the ACE-2005 and ERE-EN datasets, respectively, outperforming the best baseline models on each dataset by 1.66% and 5.11%. Compared with the strong baseline GCB-PFCE, it shows a 3.64% improvement in overlapping and nested events, indicating the model’s effectiveness and robustness in modeling complex event structures.