基于图卷积网络融合依存信息的事件检测方法

句子级别细粒度的事件检测任务旨在对触发词进行识别与分类。针对现有事件检测方法中存在的过度平滑及缺乏依存类型信息的问题，提出了一种基于图卷积网络融合依存信息的事件检测方法。该模型首先使用双向长短期记忆网络对句子进行编码，同时根据依存分析构建多阶句法图和依存句法图；然后利用图卷积网络融合句子的依存信息，从而有效地利用多跳信息和依存标签信息。在自动文本抽取数据集上进行实验，在触发词识别和分类这两个子任务中分别取得了81.7%和78.6%的F1值。结果显示，提出的方法能更加有效地捕获句子中的事件信息，提升了事件检测的效果。

Event detection based on dependency information and graph convolutional network

The fine-grained event detection task at the sentence level aims at identifying and classifying triggers. To solve the problem of over-smoothing and lack of dependency type information in existing event detection methods, this paper proposed an event detection method based on dependency information and graph convolutional network(GCN). This method firstly used bi-directional long short-term memory(Bi-LSTM) networks to encode the sentence, and constructed a multi-order syntactic graph and a dependent syntactic graph based on the dependency analysis. Then it used GCN to aggregate the sentences'' dependency information, which effectively utilized the multi-hop information and the dependent type information. On automatic content extraction(ACE) dataset, the proposed method achieved 81.7% and 78.6% F1 values in the two subtasks of trigger identification and classification. Results show that the proposed method can capture event information in sentences more effectively, and improve the effect of event detection.