A Joint Model for Chinese Medical Entity and Relation Extraction based on Graph Convolutional Networks

Mining large-scale medical entities and entity relationships from electronic medical record (EMR) is of great significance for the construction of medical knowledge graph, medical intelligent assistant diagnosis and other applications. Most existed methods regarded the medical entity recognition and entity relation classification as independent subtasks and hence use pipeline models to solve the two tasks, which may suffer error propagation and are not able to utilize the interactions between subtasks. Furthermore, much less work has been done on Chinese EMRs. This paper proposes a joint model for Chinese medical entities and their relation extraction utilizing graph convolution network (GCN). Experimental results on our manually annotated Chinese EMR dataset show that the model presented in this paper improved significantly the performance of medical entity recognition and entity relation classification.