Scholarlink

<p>User churn in OHCs have hence garnered widespread academic attention for a better understanding of engagement patterns, which, in turn, facilitates better platform design and management. This study applies inverse reinforcement learning (IRL) to capture intrinsic drivers shaping user engagement, which offers a novel predictive framework on churn. Based on a dataset of 1081 users from a breast cancer OHC, we implemented an IRL model whose states refer to user reception of informational support, emotional support, and companionship and actions refer to engagement behaviors in the form of thread initiation, replying, and self-replies. The IRL-based reward values, which reflect intrinsic drivers of user engagement, improve churn prediction accuracy across short-, mid-, and long-term time frames with F1 scores improving by up to 19.85%. Our findings contribute to the literature on user churn and retention in OHCs, and offer managerial implications for platform managers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Treat or quit: churn prediction in online health communities based on inverse reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年6月20日

查看全文
  • Xi Wang,
  • Yaqi Zhang,
  • Hui Li,
  • Zhiya Zuo

摘要

User churn in OHCs have hence garnered widespread academic attention for a better understanding of engagement patterns, which, in turn, facilitates better platform design and management. This study applies inverse reinforcement learning (IRL) to capture intrinsic drivers shaping user engagement, which offers a novel predictive framework on churn. Based on a dataset of 1081 users from a breast cancer OHC, we implemented an IRL model whose states refer to user reception of informational support, emotional support, and companionship and actions refer to engagement behaviors in the form of thread initiation, replying, and self-replies. The IRL-based reward values, which reflect intrinsic drivers of user engagement, improve churn prediction accuracy across short-, mid-, and long-term time frames with F1 scores improving by up to 19.85%. Our findings contribute to the literature on user churn and retention in OHCs, and offer managerial implications for platform managers.