A Joint Graph Neural Model for Chinese Domain Event and Relation Extraction with Character-Word Fusion

One of the most difficult problems in Chinese event extraction is the lack of interaction between characters and words, which is more prominent in domain-specific event extraction tasks. Previous studies focus on event detection and lack a joint model for Chinese event and intra-event relation extraction. This paper proposes a character-word fusion method based on a graph convolutional neural network model, which fully associates characters with related entities and greatly improves the information interaction between characters and words. At the same time, the relation extraction task is combined with the recognition and classification task of event trigger words and arguments, forming an end-to-end method for transforming text into an event graph. Experimental results on the ACE Chinese data set and Chinese domain data set DGRE demonstrate the effectiveness and superiority of the proposed method. At the same time, through the analysis of strategy analysis, ablation experiment, parameter analysis, and other analysis work, each module's role and the model's scalability are verified.