On the Generation of Medical Dialogs for COVID-19

Authors

Meng Zhou, University of California, San Diego
Zechen Li, University of California, San Diego
Bowen Tan, Carnegie Mellon UniversityFollow
Guangtao Zeng, University of California, San Diego
Wenmian Yang, University of California, San Diego
Xuehai He, University of California, San Diego
Zeqian Ju, University of California, San Diego
Subrato Chakravorty, University of California, San Diego
Shu Chen, University of California, San Diego
Xingyi Yang, University of California, San Diego
Yichen Zhang, University of California, San DiegoFollow
Qingyang Wu, University of California, San Diego
Zhou Yu, Columbia University
Kun Xu, Tencent
Eric Xing, Carnegie Mellon University & Mohamed bin Zayed University of Artificial IntelligenceFollow
Pengtao Xie, University of California, San Diego

Document Type

Conference Proceeding

Publication Title

ACL-IJCNLP 2021 - 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, Proceedings of the Conference

Abstract

Under the pandemic of COVID-19, people experiencing COVID19-related symptoms have a pressing need to consult doctors. Because of the shortage of medical professionals, many people cannot receive online consultations timely. To address this problem, we aim to develop a medical dialog system that can provide COVID19-related consultations. We collected two dialog datasets - CovidDialog - (in English and Chinese respectively) containing conversations between doctors and patients about COVID-19. While the largest of their kind, these two datasets are still relatively small compared with generaldomain dialog datasets. Training complex dialog generation models on small datasets bears high risk of overfitting. To alleviate overfitting, we develop a multi-task learning approach, which regularizes the data-deficient dialog generation task with a masked token prediction task. Experiments on the CovidDialog datasets demonstrate the effectiveness of our approach. We perform both human evaluation and automatic evaluation of dialogs generated by our method. Results show that the generated responses are promising in being doctorlike, relevant to conversation history, clinically informative and correct.

First Page

886

Last Page

896

Publication Date

1-1-2021

Keywords

Automatic evaluation; Dialogue generations; Dialogue systems; Human evaluation; Medical professionals; Overfitting; Prediction tasks; Pressung; Small data set

Comments

IR deposit conditions: none described

OA version available on ACL Anthology

Recommended Citation

M. Zhou et al., “On the generation of medical dialogs for COVID-19,” ACL-IJCNLP 2021 - 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, Proceedings of the Conference, vol. 2, pp. 886–896, 2021, doi: 10.18653/V1/2021.ACL-SHORT.112.