A Knowledge-driven Generative Model for Multi-implication Chinese Medical Procedure Entity Normalization

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1490–1499, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1490 A Knowledge-driven Generative Model for Multi-implication Chinese Medical Procedure Entity Normalization Jinghui Yan1∗, Yining Wang2,3, Lu Xiang2,3, Yu Zhou2,3,5, Chengqing Zong2,3,4 1School of Computer Science and Information Technology & Beijing Key Lab of Traffic Data Analysis and Mining, Beijing Jiaotong University, Beijing, China 2National Laboratory of Pattern Recognition, Institute of Automation, CAS, Beijing, China 3University of Chinese Academy of Sciences, Beijing, China 4CAS Center for Excellence in Brain Science and Intelligence Technology, Beijing, China 5 Beijing Fanyu Technology Co., Ltd jh yan@bjtu.edu.cn {yining.wang, lu.xiang, yzhou, cqzong}@nlpr.ia.ac.cn Abstract Medical entity normalization, which links medical mentions in the text to entities in knowledge bases, is an important research topic in medical natural language processing. In this paper, we focus on Chinese medical procedure entity normalization. However, non- standard Chinese expressions and combined procedures present challenges in our problem. The existing strategies relying on the discrim- inative model are poorly to cope with nor- malizing combined procedure mentions. We propose a sequence generative framework to directly generate all the corresponding medi- cal procedure entities. we adopt two strate- gies: category-based constraint decoding and category-based model refining to avoid unreal- istic results. The method is capable of linking entities when a mention contains multiple pro- cedure concepts and our comprehensive exper- iments demonstrate that the proposed model can achieve remarkable improvements over ex- isting baselines, particularly significant in the case of multi-implication Chinese medical pro- cedures. 1 Introduction Named entity normalization (NEN), which is also known as entity linking, is one of the fundamental tasks within natural language processing (Hachey et al., 2013; D’Souza, 2015; Fang et al., 2016; Wu et al., 2018). Medical entity normalization is a typ- ical problem of NEN in the medical domain, which aims at linking references or mentions of medical terminology to standard entities in a given medi- cal knowledge base (KB) such as the International Statistical Classification of Diseases and Related Health Problems 9th Revision (ICD-9). Due to the nature of the domain, although differ- ent occupational or writing habits can result in stan- ∗ Contribution during internship at National Laboratory of Pattern Recognition, Institute of Automation, CAS dard entities having different literal expressions, the linked standard entities of a given medical men- tion should always be unique. Thus, unlike NEN as applied generally(Hachey et al., 2013; Luo et al., 2015; Wu et al., 2018; Aguilar et al., 2019), in the medical domain, the main challenge is not ambiguity – the same entity mention may be linked to different concepts, it is variation – the same un- derlying concept can be linked by different entity mentions. However, different from the normaliza- tion task of the medical entity with simple nomi- nal structure, such as disease (Kang et al., 2012; D’Souza and Ng, 2015) or anatomical body (Wang et al., 2019), Chinese medical procedure normaliza- tion have to face the challenge of multi-implication – a …

Content truncated to 500 words