A Joint Personality-Emotion Framework for Personality-Consistent Conversational Agents

Personality and emotion are key to engaging interactions with conversational agents, yet their integration into LLM-based systems remains challenging. Current approaches often treat personality and emotion separately and, when combined, they lack mechanisms to ensure that emotional expression remains consistent with the agent’s personality. This can lead to unpredictable or contradictory behavior, undermining user trust and believability. We propose a novel framework that jointly models personality and emotion through a transparent and controllable mechanism. By warping the valence–arousal space based on personality traits, we create a 3D topology that guides emotion updates over time while preserving coherent personality expression. A blending parameter enables fine-grained control over the trade-off between personality consistency and emotional adherence. In simulations, our method preserved up to 77% of personality consistency otherwise lost with standard approaches. This framework offers both a theoretical and a practical basis for building more consistent and believable LLM-driven agents.