Emotional Cognitive Modeling Framework with Desire-Driven Objective Optimization for LLM-empowered Agent in Social Simulation

TL;DR

The paper addresses the gap in simulating human-like social systems by introducing an Emotional Cognitive Modeling Framework that embeds emotion into LLM-based agents. It defines a state-perception-desire-objective-decision loop using $State_t(I_t,H_t,SR_t,E_t)$ and a triadic $D_t$, guiding decision prompts through a tripartite $Reward( abla I, abla H, abla SR)$ and a softened policy $\,\pi^*(y|x) = rac{1}{Z^*(x)} \,\pi_F(y|x) \, \exp(Reward(\Delta I,\Delta H,\Delta SR)/\beta)$, thus achieving emotion-aligned, bounded rational behavior. The framework, implemented in a proprietary multi-agent simulator, demonstrates improved ecological validity and human-behavioral similarity of outcomes compared with rule-based, imitation, RL, and GPT-based baselines. Results indicate elevated realism in social emergence and coherent affective decision-making, suggesting significant implications for social science testbeds and human-aligned AI in interactive services, while acknowledging domain generalizability and emotional granularity as future work.

Abstract

The advent of large language models (LLMs) has enabled agents to represent virtual humans in societal simulations, facilitating diverse interactions within complex social systems. However, existing LLM-based agents exhibit severe limitations in affective cognition: They fail to simulate the bounded rationality essential for bridging virtual and real-world services; They lack empirically validated integration mechanisms embedding emotions within agent decision architectures. This paper constructs an emotional cognition framework incorporating desire generation and objective management, designed to achieve emotion alignment between LLM-based agents and humans, modeling the complete decision-making process of LLM-based agents, encompassing state evolution, desire generation, objective optimization, decision generation, and action execution. This study implements the proposed framework within our proprietary multi-agent interaction environment. Experimental results demonstrate that agents governed by our framework not only exhibit behaviors congruent with their emotional states but also, in comparative assessments against other agent types, demonstrate superior ecological validity and generate decision outcomes that significantly more closely approximate human behavioral patterns.

Figures (5)

• Figure 1: Illustration of embedding emotion into decision core of LLM-based agents.
• Figure 2: Our Emotional Cognitive Modeling Framework for LLM-based agents.
• Figure 3: Overall performance of different Rider Agents in O2O experimental scenario.
• Figure 4: Decision & desire percentage of different Rider Agents in O2O experimental scenario.
• Figure 5: Money per day and happiness percentage per day of different Rider Agents in O2O experimental scenario.