融合局部语义和全局结构信息的健康问句分类

doi:10.19665/j.issn1001-2400.2020.02.002

摘要/Abstract

摘要：

针对已有的中文医疗健康问句分类方法的不足,提出了一种融合句子局部语义信息和全局结构信息的中文医疗健康问句分类方法。首先,通过卷积神经网络和独立循环神经网络获得医疗健康问句的局部语义表示和全部结构表示;然后,利用自注意力机制将得到的局部语义表示和全局结构表示向量融合,生成医疗健康问句的最终语义表示向量;最后,通过分类层分类并输出分类结果。实验结果表明,这种方法可以有效地提高模型的语义表示能力,且有效地改善了梯度消失和梯度爆炸问题,在中文医疗健康问句数据集上具有很好的性能。

关键词: 中文医疗健康问句分类, 局部语义表示, 全局结构表示, 卷积神经网络, 独立循环神经网络

Abstract:

Considering the shortcomings of existing research methods in the Chinese medical health questions classification task, this paper proposes a new health questions classification method that incorporates the health questions’ local semantic information and global structural information. We first obtain the questions’ local semantic representation and global structural representation by the convolutional neural network (CNN) and independent recurrent neural network (IndRNN). Then, we extract the questions’ semantic representation, and especially we get the questions’ semantic representation by fusing the local semantic representation and global structural representation using a self-attention mechanism. Finally, we classify the semantic representation of the medical health question through the softmax layer and output classification result. Experimental results show that this method leads to a good performance in the Chinese medical health questions dataset, and that it effectively improves the semantic representation ability of the model and significantly resolves the gradient vanishing and gradient explosion problems.

Key words: Chinese medical health question classification, local semantic representation, global structural representation, convolution neural network, independently recurrent neural network

中图分类号:

TP391

张志昌,张治满,张珍文. 融合局部语义和全局结构信息的健康问句分类[J]. 西安电子科技大学学报, 2020, 47(2): 9-15.

ZHANG Zhichang,ZHANG Zhiman,ZHANG Zhenwen. Classifying health questions with local semantic and global structural information[J]. Journal of Xidian University, 2020, 47(2): 9-15.

图/表 9

图1

图2

图3

表1

表2

表3

两种融合方法的比较"

句子	真实标签	SC-C-IndRNN	SA-C-IndRNN	$a se$	$a st$
245阳性提示感染乙肝后痊愈,请问多久以后体检可以完全为阴性呢?以后还有必要注射疫苗吗?乙肝五项指标对照表怎么看?	治疗	流行病学	治疗	0.43	0.57

表3

图4

表4

图5

参考文献 17

[1]	CAMPILLOS L, BOUAMOR D, BILINSKI E , et al. Description of the Patient GenesysDialogue System[C]// Proceedings of the 2015 16th Annual Meeting of the Special Interest Group on Discourse and Dialogue. New York: Association for Computational Linguistics, 2015: 438-440.
[2]	ROBERTS K, KILICOGLU H, FISZMAN M , et al. DecomposingConsumer Health Questions[C]// Proceedings of the 2014 Workshop on Biomedical Natural Language Processing. New York: Association for Computational Linguistics, 2014: 29-37.
[3]	PATRICK J, LI M . AnOntology for Clinical Questions About the Contents of Patient Notes[J]. Journal of Biomedical Informatics, 2012,45(2):292-306.
[4]	ROBERTS K, RODRIGUEZ L, SHOOSHAN S E , et al. ResourceClassification for Medical Questions[C]// AMIA 2017 Annual Symposium Proceedings. Bethesda: AMIA, 2016: 1040-1049.
[5]	GUO H H, NA X, HOU L , et al. Classifying Chinese Questions Related to Health Care Posted by Consumers via the Internet[J]. Journal of Medical Internet Research, 2017,19(6):e220.
[6]	LLANOS L C, ROSSET S, ZWEIGENBAUM P . AutomaticClassification of Doctor-patient Questions for a Virtual Patient Record Query Task[C]// Proceedings of the 2017 Workshop on Biomedical Natural Language Processing. New York: Association for Computational Linguistics. 2017: 333-341.
[7]	HUGHES M, LI I, KOTOULAS S , et al. MedicalText Classification Using Convolutional Neural Networks[J]. Computer Science, 2017,235:246-250.
[8]	ZHOU C T, SUN C L, LIU Z Y , et al. A C-LSTM Neural Network for Text Classification[J/OL]. Computer Science, 2015, https://arxiv.org/pdf/1511.08630.pdf.
[9]	高云龙, 左万利, 王英 , 等. 基于稀疏自学习卷积神经网络的句子分类模型[J]. 计算机研究与发展, 2018,55(1):179-187.
	GAO Yunlong, ZUO Wanli, WANG Ying , et al. Sentence Classification Model Based on Sparse Self-learning Convolutional Neural Network[J]. Journal of Computer Research and Development, 2018,55(1):179-187.
[10]	TAI K S, SOCHER R, MANNING C D . Improved Semantic Representations from Tree-structured Long Short-term Memory Networks[C]// Proceedings of the 2015-3rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing of the Asian Federation of Natural Language Processing: 1. New York: Association for Computational Linguistics, 2015: 1556-1566.
[11]	MOU L L, PENG H, LI G , et al. Discriminative Neural Sentence Modeling by Tree-based Convolution[C]// Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. New York: Association for Computational Linguistics, 2015: 2315-2325.
[12]	ZHAO J, ZHAN Z, YANG Q , et al. Adaptive Learning of Local Semantic and Global Structure Representations for Text Classification[C]// Proceedings of the2018 27th International Conference on Computational Linguistics. New York: Association for Computational Linguistics, 2018: 2033-2043.
[13]	LI S, LI W, COOK C , et al. Independently Recurrent Neural Network (IndRNN): Buildinga Longer and Deeper RNN[C]// Proceedings of the 2018 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2018: 5457-5466.
[14]	张金刚, 方圆, 袁豪 , 等. 一种识别表情序列的卷积神经网络[J]. 西安电子科技大学学报, 2018,45(1):150-155.
	ZHANG Jinggang, FANG Yuan, YUAN Hao , et al. A Convolutional Neural Network for Identifying Expression Sequences[J]. Journal of Xidian University, 2018,45(1):150-155.
[15]	GUO H, NA X, LI J . Qcorp:an Annotated Classification Corpus of Chinese Health Questions[J]. BMC Medical Informatics and Decision Making, 2018,18(S1):16-16.
[16]	GUO H, LI J, DAI T . Consumer Health Information Needs and Question Classification: Analysis of Hypertension Related Questions Asked by Consumers on a Chinese Health Website[J]. Studies inHealth Technology and Informatics, 2015,216:810-814.
[17]	郭海红, 李姣, 代涛 . 中文健康问句分类与语料构建[J]. 情报工程, 2016,2(6):39-49.
	GUO Haihong, LI Jiao, DAI Tao . ChineseHealth Question Classification and Corpus Construction[J]. Technology Intelligence Engineering, 2016,2(6):39-49.

类别	训练集数量	开发集数量	测试集数量	问题平均长度	问题最大长度
诊断	1050	413	259	84	1626
治疗	1429	562	352	121	2098
择医	88	35	22	130	1306
流行病学	272	107	67	120	904
健康生活方式	182	72	44	78	485

方法	准确率	召回率	值
SVM	0.79	0.77	0.77
CNN	0.77	0.77	0.78
LSTM	0.74	0.74	0.74
IndRNN	0.82	0.81	0.82
SC-C-IndRNN	0.85	0.82	0.84
SA-C-IndRNN	0.86	0.86	0.86

用户输入	真实类别	预测类别
婴儿身上有白斑和皮屑	诊断	治疗
孕妇高血压吃什么通便	治疗	流行病学
产后高血压怎么办?应注意什么? 哪些东西不能吃?	健康生活方式	流行病学