The Effects of LLM-Empowered Chatbots and Avatar Guides on the Engagement, Experience, and Learning in Virtual Museums

Virtual museums, as a form of serious game, transcend spatiotemporal constraints of traditional museums through advanced computer technologies, enabling immersive and educational user experiences. With the support of large language models (LLMs) in particular, users can have exploratory interactions with conversational agents. This paper presents a comparative study examining user engagement, experience, and learning across three interaction methods: text labels, chatbots, and avatar guides in a virtual museum. Our study revealed that users were more engaged with the LLM-empowered avatar guide than chatbots and text labels. Additionally, the overall user experience with avatar guides was superior to that of chatbots and text labels. However, no significant differences were observed in users’ learning outcomes across the three interaction methods. We discussed the characteristics of these interaction methods and proposed design recommendations for virtual museums and serious games.