RealMedDial: A Real Telemedical Dialogue Dataset Collected from Online Chinese Short-Video Clips

Proceedings of the 29th International Conference on Computational Linguistics, pages 3342–3352 October 12–17, 2022. 3342 RealMedDial: A Real Telemedical Dialogue Dataset Collected from Online Chinese Short-Video Clips Bo Xu1, Hongtong Zhang1, Jian Wang1, Xiaokun Zhang1, Dezhi Hao1, Linlin Zong2∗, Hongfei Lin1, Fenglong Ma3 1School of Computer Science and Technology, Dalian University of Technology, China 2School of Software, Dalian University of Technology, China 3College of Information Sciences and Technology, Pennsylvania State University, USA {xubo, wangjian, llzong, hflin}@dlut.edu.cn {dutzht, kp5861213, kun}@mail.dlut.edu.cn, fenglong@psu.edu Abstract Intelligent medical services have attracted great research interests for providing auto- mated medical consultation. However, the lack of corpora becomes a main obstacle to re- lated research, particularly data from real sce- narios. In this paper, we construct RealMed- Dial, a Chinese medical dialogue dataset based on real medical consultation. RealMedDial contains 2,637 medical dialogues and 24,255 utterances obtained from Chinese short-video clips of real medical consultations. We col- lected and annotated a wide range of meta- data with respect to medical dialogue includ- ing doctor profiles, hospital departments, dis- eases and symptoms for fine-grained analysis on language usage pattern and clinical diagno- sis. We evaluate the performance of medical response generation, department routing and doctor recommendation on RealMedDial. Re- sults show that RealMedDial are applicable to a wide range of NLP tasks with respect to med- ical dialogue. 1 Introduction The COVID-19 pandemic has dramatically changed how outpatient care is delivered in health- care practices. To decrease the risk of transmitting the virus to either patients or healthcare workers within their health practice, providers are deferring or selectively prohibiting in-person visits, but fortu- nately, they are usually converting in-person visits to telemedicine visits (Mann et al., 2020). During the telemedicine, patients describe their symptoms of suffered diseases and/or adverse reactions of the taking drugs to doctors, while doctors provide medical consultations through online video con- ferences. Although telemedicine is convenient and timely for disease diagnoses, the continuous growth of telemedicine visits significantly increases the burden and workload of doctors, and meanwhile, the health conditions of remote patients become ∗Corresponding author increasingly difficult to be tracked. Thus, how to relieve the burden of doctors and effectively track the patient’s health conditions remains an open re- search question. Researchers from related fields are trying to solve this issue by developing medical dialogue systems to serve as virtual doctors, which greatly facilitates users to obtain medical and healthcare information. Recent advances in medical dialogue systems have benefited medical applications such as psychological consultation (Das et al., 2022), elderly care (Keshmiri et al., 2019), and disease pre-diagnosis (Nasreen et al., 2021). To build effec- tive medical dialogue systems, related studies are focusing on optimizing medical dialogue from var- ious aspects, including automatic diagnosis (Wei et al., 2018; Xu et al., 2019), medical information extraction (Zhang et al., 2020), medical slot fill- ing (Shi et al., 2020), and medical conversational summarization (Joshi et al., 2020). Although these researches have improved the performance of med- ical dialogue, this challenging task is still facing great …

Content truncated to 500 words