User Behavior Analysis in Online Health Community Based on Inverse Reinforcement Learning

Abstract

With the rapid development of the Internet, online health communities (OHCs) have gradually become an important platform for the public to seek medical services. User behavior research, an interesting research topic in OHCs, has garnered widespread academic attention in information management. While traditional quantitative and qualitative studies remain foundational, a growing number of scholars have shifted their focus toward harnessing machine learning techniques to delve deeply into OHC user behaviors, aiming to extract actionable insights for community refinement and decision-making. Given the substantial data volume of OHCs, there remains a gap in employing reinforcement learning for relevant studies. This paper embarks on a detailed analysis and predictions of user behaviors in breast cancer OHCs using inverse reinforcement learning. Research findings suggest that users are more likely to initiate new discussions, respond to others, and self-reply after receiving peer replies. Moreover, the rewards obtained through inverse reinforcement learning serve as compelling features for predicting future user behavior. The F1 value of the prediction increased by 13% and 24% in the dataset with time windows of 3 years and 5 years. The findings may contribute to the effective management and sustainable development of OHCs.