In recent years, the increasing prevalence of nuclear families and the weakening of local communities havemade solo parenting a significant social issue. In this context, there has been growing interest in chatbot systemsthat can offer parenting support anytime and anywhere. In particular, chatbots powered by Large Language Models(LLMs) are expected to address a wide variety of parenting concerns. However, the nature of these concerns isdiverse. Some users simply seek someone to listen, while others wish to organize their thoughts or obtain specificinformation. Because each user’s purpose differs, chatbots are required to respond in a flexible and personalizedmanner. Traditional systems have struggled to provide such adaptability. To address this challenge, we developeda chatbot based on ChatGPT-4o, designed to adapt its responses flexibly according to the user’s underlying intent.We conducted user studies in which real participants engaged with the chatbot, and we analyzed both the chat logsand questionnaire responses to evaluate its effectiveness. Additionally, we carried out comparative experiments usingQwen-2.5-7B-Instruct, a lightweight open-source model, to explore future considerations such as privacy protectionand local deployment. The results demonstrated that our chatbot based on ChatGPT-4o is capable of accuratelyinferring users’ goals and adjusting its conversational strategies accordingly. This capability contributed to greateruser satisfaction and trust. On the other hand, Qwen-2.5-7B-Instruct exhibited limitations in language capabilityand prompt interpretation. The main contributions of this study are threefold: (1) proposing a novel approach todesigning chatbots that can adapt their conversational style based on user intent; (2) the validation of the proposedapproach through user studies to evaluate its effectiveness and identify potential issues; and (3) a comparative analysisof ChatGPT-4o and Qwen-2.5-7B-Instruct to highlight the challenges associated with using lightweight open-sourcemodels.
Ito et al. (Sat,) studied this question.