Scholarlink

<p>This experimental study addresses a research gap by exploring sentiments toward an educational chatbot in massive open online courses using a comprehensive approach, drawing on various data sources and analytic methods to examine both aggregate- and individual-level patterns. Data sources included 1502 chatbot logs providing direct and transactional insight; 174 student open-ended reflections offering more profound qualitative perspectives; and 237 student questionnaire responses analyzed with Random Forest analysis (aggregate-level) and SHapley Additive exPlanations (SHAP; individual-level). Although most chatbot logs were categorized as neutral (88.7%), the reflection data revealed more diverse sentiments, with positive (45.4%) and negative (32.8%) responses. At the aggregate-level, Random Forest analysis identified crucial factors influencing sentiment, including willingness to use the chatbot, social presence, self-regulation, ease of use, and cognitive presence. At the individual-level, SHAP confirmed the importance of the top features while revealing nuanced variations in how other features influenced individual predictions. These findings highlight the value of integrating multiple data sources and analytical perspectives to capture both immediate interactions and overarching reflective sentiments, emphasizing the need for a student-centric approach that balances functionality with enriching learning experiences.</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

Understanding Sentiments Toward an Educational Chatbot in MOOCs: Neutral Interactions, Diverse Reflections, and Key Predictive Factors

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Songhee Han,
  • Hyangeun Ji,
  • Jiyoon Jung,
  • Unggi Lee,
  • Min Liu

摘要

This experimental study addresses a research gap by exploring sentiments toward an educational chatbot in massive open online courses using a comprehensive approach, drawing on various data sources and analytic methods to examine both aggregate- and individual-level patterns. Data sources included 1502 chatbot logs providing direct and transactional insight; 174 student open-ended reflections offering more profound qualitative perspectives; and 237 student questionnaire responses analyzed with Random Forest analysis (aggregate-level) and SHapley Additive exPlanations (SHAP; individual-level). Although most chatbot logs were categorized as neutral (88.7%), the reflection data revealed more diverse sentiments, with positive (45.4%) and negative (32.8%) responses. At the aggregate-level, Random Forest analysis identified crucial factors influencing sentiment, including willingness to use the chatbot, social presence, self-regulation, ease of use, and cognitive presence. At the individual-level, SHAP confirmed the importance of the top features while revealing nuanced variations in how other features influenced individual predictions. These findings highlight the value of integrating multiple data sources and analytical perspectives to capture both immediate interactions and overarching reflective sentiments, emphasizing the need for a student-centric approach that balances functionality with enriching learning experiences.