Overview of the ROCLING 2023 Shared Task for Chinese Multi-genre Named Entity Recognition in the Healthcare Domain

Abstract This paper describes the ROCLING-2023 shared task for Chinese multi-genre named entity recognition in the healthcare domain, including task description, data preparation, performance metrics, and evaluation results. Among eight registered teams, six participating teams submitted a total of 16 runs. This shared task demonstrates current NLP techniques for dealing with Chinese named entity recognition in multi-genre texts. All data sets with gold standards and evaluation scripts used in this shared task are publicly available for future research. Keywords: named entity recognition, information extraction, health informatics, Chinese language processing 1 Introduction Named Entity Recognition (NER) is an NLP task that involves extracting information of concern known as named entities (e.g., person, location, and organization). The NER task is a sequence labeling problem that jointly recognizes the entity boundaries and category labels. Chinese NER is difficult to process due to the lack of clear delimiters such as spaces between characters and conventional features like capitalization. Since named entity boundaries are also word boundaries for the Chinese language, incorrect word segmentation will cause error propagation. For example, “葡萄糖六磷酸鹽去氫酶” (Glucose-6- Phosphate Dehydrogenase; G6PD) is a proper name in the healthcare domain, which in a particular context may be incorrectly segmented into three words: “葡萄糖” (Glucose), “六磷酸鹽” (Hexaphosphate) and “去氫酶” (Dehydrogenase), resulting in failure to recognize it as a chemical- type named entity. Therefore, character-based methods have been found to outperform word- based approaches for breaking through this word segmentation limitation in Chinese NER (Zhang and Yang, 2018). The ROCLING-2022 shared task (Lee et al., 2022a) focuses on Chinese healthcare NER to atomically identify healthcare named entities such as symptoms, chemicals, diseases, and treatments. In this shared task, the most frequently used system architecture is BiLSTM-CRF, which usually achieves promising results, resulting in identical findings from related studies for NER in the English language (Chiu and Nichols, 2016; Ma and Hovy, 2016). Due to the greater challenge in the healthcare domain for Chinese NER, the ROCLING 2023 shared task features a Chinese healthcare NER task that focuses on healthcare texts written in three different genres as follows: 1) formal texts (FT) including health news and articles written by professional editors or journalists; 2) social media (SM) including texts from users in medical question/answer forums; and 3) Wikipedia articles (WA) created and edited by volunteers worldwide. Chinese healthcare named entities may be used in different word forms in different written genres. For examples, “後天免疫缺乏症候群” (Acquired Immunodeficiency Syndrome; AIDS) is commonly used as a spoken language form “愛滋 病” in medical forums. “甘油三酯” (Triglyceride; TG) is a different usage referred to as “三酸甘油 酯”. We organized the ROCLING 2023 shared task for Chinese Multi-genre Named Entity Recognition in the Healthcare domain (denoted as MultiNER-Health), providing an evaluation platform for the development and implementation of Chinese healthcare NER systems. Given a Overview of the ROCLING 2023 Shared Task for Chinese Multi-genre Named Entity Recognition in the Healthcare Domain Lung-Hao Lee, Tzu-Mi Lin, and Chao-Yi Chen Department of Electrical Engineering National Central University lhlee@ee.ncu.edu.tw, 110521987@cc.ncu.edu.tw, 110581007@cc.ncu.edu.tw The 35th …

Content truncated to 500 words