Scholarlink

<p>As Large Language Models (LLMs) become increasingly prevalent across various domains, prompt engineering has emerged as a key approach for shaping model behavior and response generation. One specific type of prompt is the persona pattern, which instructs the model to assume a particular role or identity. This study focuses on examining the impact of persona-based prompts on students’ experiences with LLMs, particularly investigating whether the model’s interpretation of such role-playing prompts can enhance the user experience. We conduct an experiment involving 331 students from three different academic colleges. Participants interact with two versions of an LLM built on ChatGPT: a baseline model without any role specification and a version prompted with a persona. Each participant completes four nutrition-related tasks varying in question type (fact-based or guidance-based) and content category (moral-related or non-moral-related), and evaluates the responses from both LLM versions in terms of Understanding, Trust, and Satisfaction. Specifically, for fact-type questions, both models score similarly on average in Understanding, but the original model rates higher in Trust and Satisfaction. For guidance-type questions, the persona-based model averages higher in Understanding, while the original model remains higher in Trust and Satisfaction. We find that while the persona-based model can enhance perceived Understanding, particularly in guidance-type tasks, they may reduce Trust and Satisfaction compared to baseline models because the persona role uses certain emotional expressions and inappropriate metaphors. Moreover, participants unfamiliar with prompt engineering may not benefit from such personalization as intended. This study represents an exploratory examination of how persona-based prompting influences user experience in a nutrition education context. The findings suggest that role-playing personas may affect students’ perceptions of LLM-generated educational content, highlighting the need for effective personalization strategies in educational LLMs. Future studies should examine the generalizability of these effects across disciplines and learning contexts.</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

Exploring the impact of persona-based large language models on user experience in educational settings

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Chia-Chi Wang,
  • Feng-Chi Liu,
  • Jingnan Xie,
  • Chun-Ming Lai,
  • Ching-Sheng Lin

摘要

As Large Language Models (LLMs) become increasingly prevalent across various domains, prompt engineering has emerged as a key approach for shaping model behavior and response generation. One specific type of prompt is the persona pattern, which instructs the model to assume a particular role or identity. This study focuses on examining the impact of persona-based prompts on students’ experiences with LLMs, particularly investigating whether the model’s interpretation of such role-playing prompts can enhance the user experience. We conduct an experiment involving 331 students from three different academic colleges. Participants interact with two versions of an LLM built on ChatGPT: a baseline model without any role specification and a version prompted with a persona. Each participant completes four nutrition-related tasks varying in question type (fact-based or guidance-based) and content category (moral-related or non-moral-related), and evaluates the responses from both LLM versions in terms of Understanding, Trust, and Satisfaction. Specifically, for fact-type questions, both models score similarly on average in Understanding, but the original model rates higher in Trust and Satisfaction. For guidance-type questions, the persona-based model averages higher in Understanding, while the original model remains higher in Trust and Satisfaction. We find that while the persona-based model can enhance perceived Understanding, particularly in guidance-type tasks, they may reduce Trust and Satisfaction compared to baseline models because the persona role uses certain emotional expressions and inappropriate metaphors. Moreover, participants unfamiliar with prompt engineering may not benefit from such personalization as intended. This study represents an exploratory examination of how persona-based prompting influences user experience in a nutrition education context. The findings suggest that role-playing personas may affect students’ perceptions of LLM-generated educational content, highlighting the need for effective personalization strategies in educational LLMs. Future studies should examine the generalizability of these effects across disciplines and learning contexts.