基于深度学习多模型融合的医疗命名实体识别

MEDICAL NAMED ENTITY RECOGNITION BASED ON DEEP LEARNING MULTI-MODEL FUSION

针对医疗命名实体识别中单一模型获取文本语义特征不足的问题,提出基于深度学习多模型融合的医疗命名实体识别方法。提出一种基于变异系数的加权投票算法,构建基于BERT(Bidirectional Encoder Representations from Transformers)-MNER、IDCNN(Iterated Dilated Convolution Neural Networks)-MNER、GAT(Graph Attention Networks)-MNER的融合模型,然后提出一种基于历史信息的实体纠错算法优化融合结果。以2019年全国知识图谱与语义计算大会(CCKS2019)中文电子病历医疗实体识别语料作为实验数据,实验结果表明,该方法获得了较好的识别效果,精确率、召回率和F1值分别达到89.56%、82.77%和86.03%。

Aiming at the problem of insufficient semantic features of a single model in medical named entity recognition,this paper proposes a medical named entity recognition method based on deep learning multi-model fusion.A weighted voting algorithm based on the coefficient of variation was proposed to build a fusion model based on BERT(Bidirectional Encoder Representations from Transformers)-MNER,IDCNN(Iterated Dilated Convolution Neural Networks)-MNER and GAT(Graph Attention Networks)-MNER.Then,entity error correction algorithm based on historical information was used to optimize the fusion result.Taking the Chinese electronic clinical record medical entity recognition corpus of China Conference on Knowledge Graph and Semantic Computing in 2019(CCKS2019)as experimental data,the experimental results show that this method has achieved a good recognition effect,and the precision rate,recall rate and F1 value reached 89.56%,82.77%and 86.03%respectively.