페르소나 기반 One-shot MBTI 프롬프트 엔지니어링

초록

Recently, large-scale language models are utilized in the diverse kinds of domains. However, It is still required to overcome thelimitations of conventional generic and neutral response styles of a conversational agent as well as large-scale language models that arebiased toward specific MBTI types. This paper proposes a dialogue generation technique that effectively reflects a conversational agent’sMBTI-based personality traits to provide personalized responses to users. This paper designs a system that conveys an agent’s personalitytraits with only few prompt examples by utilizing one-shot prompt engineering technique. The proposed system is composed of learningand inference phases; during the learning phase, the agent’s dialogues are evaluated and calibrated to ensure consistency with the agent’spersona and MBTI characteristics. During the inference phase, the optimized MBTI prompts are selectively applied to enable the agentto maintain rich contextual understanding and consistent personality. This approach presents the potential to deliver more natural andpersonalized interactive experiences in various fields, including virtual assistants, educational tutors, and entertainment characters

키워드