A Global Pointer based Entity Relation Extraction Method for Chinese Pulmonary Nodule Medical Records

Abstract

To facilitate research on pulmonary nodule medical records for physicians, this paper proposes an entity relation extraction model which bases on Global Pointer, using embedded pre-trained language model RoFormer as upstream encoder, Exponential Moving Average optimization method and Fast Gradient Method for adversarial training. The proposed model can also analyze the parent-child relations on contextual semantics, and then process them into structured data. The experimental results show that this model improves the extraction effect significantly compared with the traditional methods, and the F1 value can reach 86.2% in the Chinese pulmonary nodule medical records dataset.