Information Presentation Strategies for Recommender-Chatbots

초록

This study examines how information presentation strategies―specifically online review provision and price framing―affect user experience and reuse intention in an LLM-based hotel recommender chatbot. To investigate this, we developed a hybrid chatbot integrating a large language model (LLM), Text-to-SQL, and Retrieval-Augmented Generation (RAG), and conducted a 2 × 2 between-subjects experiment based on real user interactions. Data were collected from 160 participants who interacted with the chatbot and subsequently completed a survey. The results indicate that both online review provision and downward price framing significantly enhance perceived recommender chatbot quality, which subsequently increases user satisfaction and reuse intention, whereas their interaction effect is not significant. In addition, exploratory factor analysis revealed that the multidimensional structure of chatbot quality converged into a single higher-order factor. Mediation analysis further showed that the effects of information presentation strategies on behavioral outcomes were mediated by perceived chatbot quality and satisfaction. These findings suggest that, in conversational recommender environments, information presentation strategies influence user behavior primarily through holistic quality evaluations rather than through direct effects on satisfaction.

키워드