Using machine learning to explore the determinants of service satisfaction with online healthcare platforms during the COVID-19 pandemic

Abstract

This study investigates the determinants of service satisfaction with online healthcare platforms using machine learning (ML) algorithms. By training and testing eleven ML models based on data mined from a leading online healthcare platform in China, we obtained the best-performing ML algorithm for service satisfaction prediction, namely, Light Gradient Boosting Machine. Furthermore, our empirical results indicate that gifts, patient votes, popularity, fee-based consultation volume, gender, and thank-you letters positively impact service satisfaction, while the impacts of consultation volume, free consultation volume, views, waiting time, articles, physician title, and hospital level are negative. We discuss the theoretical and managerial implications.

How does digital transformation promote total factor productivity? Strategy, technology, and application

“…Considering no consensus exists on the best machine learning models for predicting TFP, we adopted 15 machine learning algorithms in this study and then selected the best-performing algorithm. There are two types of machine learning algorithms (Liu, Li, et al, 2023). Individual machine learning algorithms attempt to generate one hypothesis from training data, while ensemble machine learning algorithms attempt to generate and combine multiple hypotheses (Zhu et al, 2017).…”

“…We used a machine learning approach to determine the role of DT in a firm's productive efficiency, though most of the existing literature mainly applied traditional statistical methods (e.g., Li, Wang, et al, 2022;Wen et al, 2022). Since the function form of influence factors such as DT and TFP is ambiguous, machine learning algorithms can offer various function forms compared to traditional statistical methods (Liu, Li, et al, 2023;Zhang et al, 2023). Furthermore, interpretable machine learning models, such as the SHapley Additive exPlanations (SHAP) approach used in this study, can explore the contribution and effects of different dimensions of DT to TFP (Lundberg & Lee, 2017).…”