基于多特征融合的双通道医疗实体识别  

Dual channel medical entity recognition based on multi-feature fusion

作　　者：廖涛[1] 马文祥 张顺香[1] LIAO Tao;MA Wen-xiang;ZHANG Shun-xiang(College of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《计算机工程与设计》2023年第10期3146-3152,共7页Computer Engineering and Design

基　　金：国家自然科学基金面上基金项目(62076006);安徽高校协同创新基金项目(GXXT-2021-008);安徽省重点研发计划国际科技合作专项基金项目(202004b11020029)。

摘　　要：针对医疗实体识别中词向量特征单一和忽略文本中局部特征的问题,提出一种基于多特征融合的双通道医疗实体识别模型。对医疗文本字形特征和卷积神经网络进行研究,发现构造的外部特征和挖掘的内部特征进行差异融合能够丰富词向量的特征信息;利用注意力机制改进的卷积神经网络实现特征优化选择,区分不同特征的重要性;设计CNN和BiLSTM并行的双通道神经网络,充分考虑文本的局部特征和上下文特征。在CCKS2017数据集上的实验结果表明,该模型能有效提高医疗实体识别的准确率。A two-channel medical entity recognition model based on multi-feature fusion was proposed to solve the problems of single word vector feature and ignoring local features in text in medical entity recognition.By studying the glyph feature and convolution neural network of medical texts,it was found that the feature information of word vectors was enriched by the diffe-rence fusion of the external feature of construction and the internal feature of mining.The improved convolution neural network with attention mechanism was used to optimize feature selection and distinguish the importance of different features.A parallel two-channel neural network with CNN and BiLSTM was designed,taking full account of the local and contextual features of the text.Results of experiments on CCKS2017 dataset show that the model can effectively improve the accuracy of medical entity recognition.

关 键 词：命名实体识别 医疗实体 多头注意力机制 多特征融合 卷积注意力机制 双通道神经网络 条件随机场 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]