Jointly Events Extraction for Database Alarm based on Dynamic Matching Strategy and GCN

Event extraction is one of the important processes in building event knowledge graphs. However, existing models cannot combine syntactic construction of text and structured knowledge of event to execute the identification and classification of trigger word and argument. In order to solve this problem, a jointly events extraction model for database fault alarm based on dynamic matching strategy and Graph Convolution Neural Networks (GCN) is proposed. Specifically, dynamic matching strategy is designed to label the event nodes which include the feature of event construction for the text, thus solving the classification of the ambiguous and unseen trigger words. Then, to get better syntactic features, GCN is used to aggregate the semantic features. The experimental results on Oracle alarm dataset show that Precision, Recall and F1 of our model respectively are 76.3%, 72.2% and 74.2%., which are better than JMEE and GREE.