Sentipolis Framework Enhances Emotional Continuity in LLM Social Simulations

A new framework called Sentipolis addresses emotional amnesia in LLM agents used for social simulations by implementing continuous Pleasure-Arousal-Dominance representation. The system integrates dual-speed emotion dynamics and emotion-memory coupling to create emotionally stateful agents. Testing across thousands of interactions with multiple base models and evaluators demonstrated improved emotionally grounded behavior and communication. Emotional continuity received significant boosts through this approach. Performance gains proved model-dependent, with higher-capacity models showing increased believability while smaller models sometimes experienced declines. The framework revealed a human-like tension where emotion-awareness can mildly reduce adherence to social norms. Network-level diagnostics indicated reciprocal, moderately clustered relationship structures that maintain temporal stability. These findings were documented in arXiv preprint 2601.18027v2.