A Named Entity Recognition Corpus for Vietnamese Biomedical Texts to Support Tuberculosis Treatment

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3601–3609 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3601 A Named Entity Recognition Corpus for Vietnamese Biomedical Texts to Support Tuberculosis Treatment Uyen T.P. Phan1,2, Phuong N.V. Nguyen3, Nhung T.H. Nguyen4 1Faculty of Information Technology, University of Science, Ho Chi Minh city, Vietnam 2 Vietnam National University, Ho Chi Minh city, Vietnam 3Pham Ngoc Thach University of Medicine, Vietnam 4Department of Computer Science, University of Manchester, UK ptpuyen@fit.hcmus.edu.vn, nnvanphuong@pnt.edu.vn, nhung.nguyen@manchester.ac.uk Abstract Named Entity Recognition (NER) is an important task in information extraction. However, due to the lack of labelled corpora, biomedical NER has scarcely been studied in Vietnamese compared to English. To address this situation, we have constructed VietBioNER, a labelled NER corpus of Vietnamese academic biomedical text. The corpus focuses specifically on supporting tuberculosis treatment, and was constructed by collecting scientific papers and grey literature related to tuberculosis symptoms and diagnostics. We manually annotated a small set of the collected documents with five categories of named entities: Organisation, Location, Date and Time, Symptom and Disease, and Diagnostic Procedure. Inter-annotator agreement ranges from 70.59% and 95.89% F-score according to entity category. In this paper, we make available two splits of the corpus, corresponding to traditional supervised learning and few-shot learning settings. We also provide baseline results for both of these settings, in addition to a dictionary-based approach, as a means to stimulate further research into Vietnamese biomedical NER. Although supervised methods produce results that are far superior to the other two approaches, the fact that even one-shot learning can outperform the dictionary-based method provides evidence that further research into few-shot learning on this text type would be worthwhile. Keywords: Named Entity Recognition, Vietnamese Biomedical Texts, Corpus Annotation 1. Introduction According to WHO, prior to the coronavirus (COVID- 19) pandemic, tuberculosis (TB) was the leading cause of death from a single infectious agent, ranking above HIV/AIDS (WHO, 2021). Meanwhile, Nguyen et al. (2020) reported that TB burden in Vietnam still re- mains high, despite government efforts to eradicate the disease. A system to support tuberculosis treatment for the Vietnamese language could therefore be extremely valuable in helping to improve the situation. Ideally, the system should be able to provide answers to the fol- lowing questions: • In which geographical areas is TB present? • When have cases of TB been reported? • Which methods are used to diagnose TB? • Which symptoms/diseases are relevant to TB? • Which organisations suggest/authorise a diagnos- tic procedure? A key element in such a system would be a named entity recognition (NER) tool, which could automati- cally identify mentions of the important entities (e.g. locations, diseases and dates/times) that are needed to answer the above questions. Unfortunately, such a tool does not currently exist for the Vietnamese lan- guage, and developing one is very problematic, due to the lack of supporting resources. Approaches to NER can be classified into three main groups: rule-based, dictionary-based and machine …

Content truncated to 500 words