题目
Exploring Word Segmentation and Medical Concept Recognition for Chinese Medical Texts
中文医学文本分词与医学概念识别研究
香港中文大学
摘要
问题:要处理电子病历,可是医学数据集缺乏。
贡献:收集了电子病历 – ACEMR(有人工标注的),运行模型BiLSTM,BERT,ZEN; 还有系统:WMSeg,TwASP
实验结果证明了建立一个专用医疗数据集的必要性,并表明利用额外资源的模型在这两个任务中达到最佳性能,为未来医学领域模型选择研究提供了一定的指导。
https://github.com/cuhksz-nlp/ACEMR (还未更新完毕)
背景与相关技术
中文分词(CWS)和医学概念识别是处理中文电子病历(EMRs)的两项基本任务,并在理解中文电子病历的下游任务中发挥着重要作用。
之前的研究【1】~【7】,都是构造自己的数据集并进行研究的 。
收集了一个数据集ACEMR:500 EMRs (7K sentences)
并且基于这个数据集做了一些实验。
ACEMR语料
数据收集
来自5个科室:呼吸系统、胃肠病学、泌尿学、妇科和心脏病学
包括信息:
CWS和医学概念注释
分类:
方法
EMR的CWS
看成序列标记任务(看实验结果)
Medical Concept Recognition
看成序列标记任务(看实验结果)
这里引入了一些句法的内容,这个是展示了这个内容:POS,依赖关系,句法结构
实验
数据集:ACMER,CTB6
train与test:
CWS结果:
这里有三种情况:
情况1:CTB Only,表示在CTB6中训练,在ACEMR测试集中验证;
情况2:CTB+ACEMR,
情况3:ACEMR Only
为什么要把CTB加入来呀?不是太明白!!是想说明这个算法好呢?还是想说明这个数据集好呀?
MCR结果:
子类那个略过。
加入一语法信息得到的效果:
还等语料公布了测试一下。
总结:
冲着题目给读完了,读完了其实没有什么感觉。
不过可以看一下它的引用内容,从引用来的内容去学习一下。
腾讯的Directional Skip-Gram: Explicitly Distin guishing Left and Right Context for Word Embeddings,https://ai.tencent.com/ailab/nlp/zh/embedding.html;
Improving Chinese Word Segmentation with Wordhood Memory Networks
作者:Yuanhe Tian, Yan Song, Fei Xia, Tong Zhang, Yonggang Wang
论文地址:https://www.aclweb.org/anthology/2020.acl-main.734/
GitHub:https://github.com/SVAIGBA/WMSeg
Joint Chinese Word Segmentation and Part-of-speech Tagging via Two-way Attentions of Auto-analyzed Knowledge
作者:Yuanhe Tian, Yan Song, Xiang Ao, Fei Xia, Xiaojun Quan, Tong Zhang, Yonggang Wang
论文地址:https://www.aclweb.org/anthology/2020.acl-main.735/
GitHub:https://github.com/SVAIGBA/TwASP
参考:
【1】Junjie Xing, Kenny Zhu, and Shaodian Zhang. 2018.
Adaptive Multi-task Transfer Learning for Chinese Word Segmentation in Medical Text. In Proceedings of the 27th International Conference on Computational Linguistics, pages 3619–3630.
【2】Qi Wang, Yangming Zhou, Tong Ruan, Daqi Gao,
Yuhang Xia, and Ping He. 2019. Incorporating Dictionaries into Deep Neural Networks for the Chinese Clinical Named Entity Recognition. Journal of biomedical informatics, 92:103133.
【3】Yan Xu, Yining Wang, Tianren Liu, Jiahua Liu, Yubo
Fan, Yi Qian, Junichi Tsujii, and Eric I Chang. 2014. Joint Segmentation and Named Entity Recognition using Dual Decomposition in Chinese discharge summaries. Journal of the American Medical Informatics Association, 21(e1):e84–e92.
【4】Dong Xu, Meizhuo Zhang, Tianwan Zhao, Chen Ge,
Weiguo Gao, Jia Wei, and Kenny Q Zhu. 2015. Datadriven Information Extraction from Chinese Electronic Medical Records. PloS one, 10(8):e0136270.
【5】Yu-Bing Li, Xue-Zhong Zhou, Run-Shun Zhang, YingHui Wang, Yonghong Peng, Jing-Qing Hu, Qi Xie, Yan-Xing Xue, Li-Li Xu, Xiao-Fang Liu, et al. 2015. Detection of Herb-Symptom Associations from Traditional Chinese Medicine Clinical Data. EvidenceBased Complementary and Alternative Medicine,2015.
【6】Shaodian Zhang, Tian Kang, Xingting Zhang, Dong Wen, Noémie Elhadad, and Jianbo Lei. 2016. Speculation Detection for Chinese Clinical Notes: Impacts of Word Segmentation and Embedding Models. Journal of biomedical informatics, 60:334–341.
【7】Bin He, Bin Dong, Yi Guan, Jinfeng Yang, Zhipeng
Jiang, Qiubin Yu, Jianyi Cheng, and Chunyan Qu.2017. Building a Comprehensive Syntactic and Semantic Corpus of Chinese Clinical Texts. Journal of biomedical informatics, 69:203–217.