An End-to-End Progressive Multi-Task Learning Framework for Medical Named Entity Recognition and Normalization

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6214–6224 August 1–6, 2021. ©2021 Association for Computational Linguistics 6214 An End-to-End Progressive Multi-Task Learning Framework for Medical Named Entity Recognition and Normalization Baohang Zhou1,3, Xiangrui Cai2,3, Ying Zhang1,3,∗, Xiaojie Yuan1,3 1 College of Computer Science, Nankai University, Tianjin 300350, China 2 College of Cyber Science, Nankai University, Tianjin 300350, China 3 Tianjin Key Laboratory of Network and Data Security Technology, Tianjin 300350, China zhoubh@mail.nankai.edu.cn, {yingzhang,caixr,yuanxj}@nankai.edu.cn Abstract Medical named entity recognition (NER) and normalization (NEN) are fundamental for con- structing knowledge graphs and building QA systems. Existing implementations for medi- cal NER and NEN are suffered from the er- ror propagation between the two tasks. The mispredicted mentions from NER will directly influence the results of NEN. Therefore, the NER module is the bottleneck of the whole system. Besides, the learnable features for both tasks are beneficial to improving the model performance. To avoid the disadvan- tages of existing models and exploit the gener- alized representation across the two tasks, we design an end-to-end progressive multi-task learning model for jointly modeling medical NER and NEN in an effective way. There are three level tasks with progressive difficulty in the framework. The progressive tasks can re- duce the error propagation with the incremen- tal task settings which implies the lower level tasks gain the supervised signals other than errors from the higher level tasks to improve their performances. Besides, the context fea- tures are exploited to enrich the semantic infor- mation of entity mentions extracted by NER. The performance of NEN profits from the en- hanced entity mention features. The stan- dard entities from knowledge bases are intro- duced into the NER module for extracting cor- responding entity mentions correctly. The em- pirical results on two publicly available med- ical literature datasets demonstrate the superi- ority of our method over nine typical methods. 1 Introduction To dig into the large amount of electronic medical records, there has been an increasing interest in applying information extraction to them. These techniques can generate tremendous benefit for cor- responding research and applications, such as med- ∗Corresponding author. NER NEN Medical Text Medical Text NER NEN Feature Extractor Standard Entity Base ( MeSH or OMIM ) Pipeline Framework Parallel Multi-task Framework Low-level Task End-to-end Progressive Multi-task Framework Medical Text Mid-level Task High-level Task Figure 1: The overall frameworks for medical named entity recognition and normalization. ical knowledge graph (Wu et al., 2019) and QA systems (Lamurias and Couto, 2019). Among the medical text mining tasks, medical named entity recognition and normalization are the most funda- mental tasks. Named entity recognition tries to find the bound- aries of mentions from the medical texts. And named entity normalization maps mentions ex- tracted from the medical text to standard identifiers, such as MeSH and OMIM (Zhao et al., 2019). The initial pipeline implementations for medical NER and NEN have a main limitation: error extractions from …

Content truncated to 500 words