Chinese Event Extraction Method based on Abstract Meaning Representation Graph

Abstract

Most of Chinese event extraction methods mainly focus on syntactic information and pay scant attention to semantic information. To make full use of semantic information, we take event extraction as a sequence labeling task and propose a Chinese event extraction method based on abstract meaning representation to improve the accuracy and recall rate of event extraction tasks. The model generates abstract meaning representation graph for input sentences, and the graph attention mechanism is used to capture semantic relationships, extract event argument elements and event triggers, and improve the prediction ability of the model. Compared with the baseline models, the accuracy rate, recall rate and F1-score of ACE2005 Chinese set is significant increased, especially in argument classification, with 11.4%, 13.4%, and 12.5% respectively, which verified the effectiveness of our method.