Rikuto Fukushima, Masayuki Hashimoto

Abstract: In modern society, the problem of social isolation and loneliness among the elderly living alone has become increasingly serious due to the progression of aging. Casual dialogue systems have gained attention as a potential solution to this issue. This study aims to build a casual dialogue system that facilitates shared-experience-focused utterances, where the system and the user engage in dialogue based on commonly recognized knowledge and experiences. Grounded in the theory of utterance classification incorporating insights from language education, the proposed system seeks to promote natural interaction through shared understanding. To achieve this goal, we refined our prompt design and conducted experiments focusing on text-based dialogue. This approach enabled us to test a wide variety of prompts, as the system’s performance was evaluated through LLM-to-LLM dialogue results assessed subjectively by the author. As a result, the proposed prompt design generated natural shared-experience-focused utterances while reducing contradictions with self-attributes and misinterpretations of given information. These findings demonstrate the effectiveness of incorporating shared experiences and language-education-based utterance classification into prompt design for improving rapport in LLM-based casual dialogue systems.

Keywords: Prompt, Dialogue systems, Large Language Models.

Date Published: December 30, 2025
DOI: 10.11159/jmids.2025.009

View Article