MRC-based Medical NER with Multi-task Learning and Multi-strategies

CCL 2022 Proceedings of the 21st China National Conference on Computational Linguistics, pages 836-847, Nanchang, China, October 14 - 16, 2022. (c) Technical Committee on Computational Linguistics, Chinese Information Processing Society of China 836 MRC-based Medical NER with Multi-task Learning and Multi-strategies Xiaojing Du, Yuxiang Jia∗, and Hongying Zan School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou, China zzu_dxj@163.com,{ieyxjia,iehyzan}@zzu.edu.cn Abstract Medical named entity recognition (NER), a fundamental task of medical information ex- traction, is crucial for medical knowledge graph construction, medical question answering, and automatic medical record analysis, etc. Compared with named entities (NEs) in gen- eral domain, medical named entities are usually more complex and prone to be nested. To cope with both flat NEs and nested NEs, we propose a MRC-based approach with multi-task learning and multi-strategies. NER can be treated as a sequence labeling (SL) task or a span boundary detection (SBD) task. We integrate MRC-CRF model for SL and MRC-Biaffine model for SBD into the multi-task learning architecture, and select the more efficient MRC-CRF as the final decoder. To further improve the model, we employ multi-strategies, including adaptive pre-training, adversarial training, and model stacking with cross validation. Experiments on both nested NER corpus CMeEE and flat NER corpus CCKS2019 show the effectiveness of the MRC-based model with multi-task learning and multi-strategies. 1 Introduction With the fast development of medical digitalization, more and more medical documents are generated, including electronic medical records, medical reports, etc. Medical information ex- traction, including medical named entity recognition (NER), becomes increasingly important to applications like knowledge graph construction, question answering system, and automatic electronic medical record analysis. Medical NER is a task to automatically recognize medical named entities, like body (bod), disease, clinical symptom (sym), medical procedure, medical equipment, drug, medical examination item, etc., from medical texts. Medical named entities are usually long, nested and polysemous, which pose great challenges to medical NER. For example, in Fig 1, the two “bod” entities “延髓”(medulla oblongata) and “脊髓”(spinal cord) are nested in the “sym” entity “延髓和脊髓受损”(damage to the medulla oblongata and spinal cord). Figure 1: An example with nested entity To tackle both flat and nested NER, like (Li et al., 2020b), we take NER as a machine reading comprehension (MRC) problem. In addition, from different views, NER can be treated as a ∗Corresponding author ©2022 China National Conference on Computational Linguistics Published under Creative Commons Attribution 4.0 International License Computational Linguistics sequence labeling (SL) task or a span boundary detection (SBD) task. We integrate MRC-CRF model for SL and MRC-Biaffine model for SBD into the multi-task learning (MTL) architecture. There is no nested NEs composed of entities of the same type in the datasets, so we select the more efficient MRC-CRF as the final decoder. To further improve the model, we employ multi- strategies (MS), including adaptive pre-training, adversarial training, and model stacking with cross validation. The main contributions of this paper are as follows: • We improve MRC-CRF for medical NER with Biaffine through a multi-task learning archi- tecture, which …

Content truncated to 500 words