Beyond Flat Personas: Facilitating Reflective Dialogue via Identity-Based Multi-Persona Agents

Abstract

In contemporary society, individuals express their multifaceted identities through multiple SNS accounts, multiple profile settings, and metaverse avatars, revealing the limitations of conventional dialogue systems built around a flat persona. This study designs identity-based multi-personas that reflect users' multiple identities and narrative experiences, and integrates them into large language model-based conversational agents to investigate their impact on user experience through both quantitative and qualitative analyses. A user study involving 30 participants revealed that multi-personas with enhanced identity alignment had significant effects on self-awareness, self-reflection, self-acceptance, and self-authenticity compared to baseline personas. The findings indicate that stronger persona alignment fosters emotional expression and immersive engagement. This research suggests that conversational agents capable of understanding users more deeply can move beyond mere information delivery tools to serve as companions that support users' inner growth. It further proposes future directions for designing LLM-based dialogue systems that account for identity alignment and ethical autonomy.