Identifying Service Gaps from Public Patient Opinions Through Text Mining

Nowadays, healthcare systems have become increasingly patientcentered and the unstructured, open-ended and patient-driven feedback has drawn a significant attention from medical and healthcare organizations. Based on this, we are motivated to harness various machine learning algorithms to process such a large amount of unstructured comments posted on public patient opinion sites. We first used sentiment analysis to automatically predict the concerns of patients from the training set which was already labelled. Then, with the help of the clustering, we extracted the hot topics related to a specific domain to reflect the service issues that patients concern most. Through experimental studies, the performance of different algorithms and the influence of different parameter were compared. Finally, refering to the survey and previous studies, the results were analyzed to obtain the conclusions.