Modeling U.S. Attitudes Toward China via an Event-Steered Multi-Agent Simulator

2026-06-19 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computational Social Science · Depth: Expert, extended

Summary

Researchers propose the Event-Steered Multi-Agent Simulator (ES-MAS) to model the dynamic evolution of U.S. public attitudes toward China, addressing limitations of static LLM-based simulators. ES-MAS integrates significant geopolitical events and daily news to continuously drive opinion shifts through dynamic agent interactions. A new China–U.S. Relation Evolution (CURE) dataset was constructed, spanning 20 quarters from 2021 to 2025, comprising 258 major events and over 14,000 daily news articles. The simulator features a Dual-Stream Data Integration Engine (DSDIE) for historical alignment and personalized news exposure, alongside a News-Driven Dynamic Interaction (NDDI) module that groups agents by shared news interests. Experiments with N=100 agents over 20 quarterly steps demonstrate ES-MAS's superior performance in reproducing real-world historical trends compared to existing methods.

Key takeaway

For research scientists modeling complex geopolitical opinion dynamics, traditional static LLM-based simulators are insufficient. You should adopt event-steered multi-agent simulation frameworks like ES-MAS, which integrate real-time events and daily news to capture dynamic opinion shifts. This approach, validated on the China-U.S. Relation Evolution dataset, offers significantly higher accuracy in reproducing historical trends, enabling more robust risk assessment and policy insights.

Key insights

Event-Steered Multi-Agent Simulators (ES-MAS) with dual-stream data and dynamic news-driven interactions accurately model complex, real-world opinion evolution.

Principles

• Static rules limit LLM simulator accuracy.
• Events and news drive opinion evolution.
• Dynamic interactions foster consensus.

Method

ES-MAS integrates significant events and personalized daily news via DSDIE. NDDI dynamically groups agents by news interest for interaction, then aggregates individual attitudes to derive macro-level opinion trajectories.

In practice

• Use CURE dataset for geopolitical modeling.
• Employ GPT-4o-mini for agent reasoning.
• Scale simulations up to 100 agents.

Topics

• Multi-Agent Systems
• Large Language Models
• Opinion Dynamics
• Geopolitical Simulation
• China-U.S. Relations
• Event-Driven Simulation
• CURE Dataset

Best for: AI Scientist, Research Scientist

Related on AIssential

• Position: AI Agents Are Not (Yet) a Panacea for Social Simulation — LLM agents for social simulation require explicit environment modeling and rigorous evaluation beyond mere plausibility.
• Mapping the Political Discourse in the Brazilian Chamber of Deputies: A Multi-Faceted Computational Approach — Computational methods can reveal hidden patterns in parliamentary discourse beyond voting records.
• ARMOR-MAD: Adaptive Routing for Heterogeneous Multi-Agent Debate in Large Language Model Reasoning — Adaptive control and genuine model heterogeneity are crucial for efficient and accurate multi-agent LLM reasoning.
• Polar: A Benchmark for Evaluating Political Bias in LLMs — LLM political bias is context-dependent, varying by language, ideology, and geography.
• Think-Before-Speak: From Internal Evaluation to Public Expression in Multi-Agent Social Simulation — TBS framework enables mechanism-sensitive social simulation by making agents' internal reasoning and public expression observable.
• Network Effects and Agreement Drift in LLM Debates — LLM agent opinion dynamics are shaped by network structure, group imbalance, and a systematic "agreement drift" bias.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.