Joint event extraction model based on dynamic attention matching and graph attention networks
Jiajun Cheng, Wenjie Liu, Zhifan Wang, Zhijie Ren, Xingwen Li
Abstract
Event extraction is one of the important processes in event knowledge graph construction. However, extant event extraction models are confronted with the challenge of handling vague and unfamiliar event trigger words as well as noise that is prevalent in text. To address this issue, this study proposes a joint event extraction model that leverages dynamic attention matching and graph attention network. Specifically, the dynamic attention matching mechanism is employed to identify event nodes that contain text event structure features and to integrate event structure knowledge for constructing event pattern subgraph that correspond to the text, thereby resolving the problem of ambiguous and unknown trigger word classification. To better discriminate between semantic information and event structure information and to mitigate the impact of noise in text, we introduce a graph attention network that integrates event structure features for aggregating feature embedding of node neighbors. Experiment results on the ACE2005 dataset demonstrate that our proposed model attains competitive performance in comparison to existing methods.