Go to ACL ARR 2026 January homepageCharacterChat: Learning towards Conversational AI with Personalized Social Support
Keywords: emotional support dialogue, personalized dialogue
Abstract: The escalating demand for psychological health support highlights the urgent need for scalable and effective interventions. However, traditional Emotional Support Conversation (ESC) frameworks often adopt a generic, "one-size-fits-all" approach, failing to accommodate the diverse personalities of individuals seeking help. To address this limitation, we introduce the \textbf{Social Support Conversation (S2Conv)} framework. This novel approach leverages a diverse pool of character agents and an interpersonal matching mechanism to pair users with persona-compatible virtual companions.
Specifically, we utilize the Myers-Briggs Type Indicator (MBTI) for persona decomposition, constructing the \textbf{MBTI-1024 Bank}—a repository of virtual characters with distinct, granular profiles. To facilitate high-quality interaction, we propose enhanced role-playing prompts integrated with behavior presets and dynamic memory mechanisms, resulting in the creation of the \textbf{MBTI-S2Conv} dataset.
Building upon these foundations, we present \textbf{CharacterChat}, a goal-oriented distributed dialogue system. CharacterChat features a persona- and memory-driven conversational model and a unique interpersonal matching plugin that identifies and dispatches the optimal supporter from the MBTI-1024 Bank based on the user's personality traits.
Empirical results demonstrate CharacterChat's superior efficacy in providing personalized social support, validating the substantial advantages of personality-aware interpersonal matching\footnote{Our code and data will be released upon paper acceptance.}.
Paper Type: Long
Research Area: Dialogue and Interactive Systems
Research Area Keywords: AI / LLM Agents
Contribution Types: NLP engineering experiment
Languages Studied: en
Submission Number: 9497
Loading