Data Dialogue with Shijing Si, Duke University: Natural Language Processing of the User-Generated MyChart Messages

Natural Language Processing of the User-Generated MyChart Messages
Abstract: Online portals like MyChart allow patients to communicate with healthcare providers conveniently. Therefore, a large number of messages
are sent to healthcare providers everyday, which significantly increase the workload of doctors. We investigate the use of both unsupervised and supervised
learning tools to patient-generated messages. Specifically, 1.) we utilize topic modeling to the massive messages to discover topics and their trends; 2.) we
leverage the state-of-the-art BERT to construct an automatic message triage. Our deep learning classifier outperforms a few baseline methods, reaching a
F1-score of 80.7%. Click here to register: https://duke.zoom.us/j/94764832177