An Event Relation Extraction Model for Database Alarm Based on Type-Guided Graph Attention Network and CNN

Event relation extraction is a fundamental task in event knowledge graph. However, most existing models fail to use the dependency-type knowledge in the syntactic dependency tree of the sentence to assist in extracting the global information. In order to solve this problem, we propose an event relation extraction model for database alarm based on Type-Guided Graph Attention Network and CNN(TGGAT-CNN). Specifically, the TGGAT is designed to add the dependency-type knowledge into the computation of the graph attention mechanism, thus effectively aggregating semantic context among sentence to extract the global information. Then, to enhance the semantic representation of the sentence, the CNN is used to extract local information of the sentence. The experimental results on Oracle alarm dataset show that Pre, Rec and F1 of our model respectively are 81.5%, 83.7% and 83.4%, which are better than the state-of-the-art models.