HAG: Hierarchical Demographic Tree-based Agent Generation for Topic-Adaptive Simulation

TL;DR

This work tackles the challenge of topic-adaptive, high-fidelity agent population generation for Agent-Based Modeling by balancing macro-level demographic distributions with micro-level individual consistency. It introduces HAG, a two-stage framework that first builds a Topic-Adaptive Demographic Distribution Tree using a World Knowledge Model to capture hierarchical joint attribute probabilities, and then grounds leaf nodes in real-world data with targeted agentic augmentation to fill gaps. To evaluate such populations, the authors construct a multi-domain benchmark (Bluesky, Amazon, IMDB) and propose PACE, a rigorous framework for Population Alignment and Sociological Consistency, including both distributional and semantic metrics. Across experiments with multiple LLMs, HAG significantly reduces population alignment errors (average 37.7%) and enhances sociological consistency (average 18.8%) compared with baselines, demonstrating improved macro-structure fidelity and micro-level realism. The work advances high-fidelity, topic-aware ABM and provides a benchmark and evaluation protocol to spur further research in grounded, sociologically plausible agent generation.

Abstract

High-fidelity agent initialization is crucial for credible Agent-Based Modeling across diverse domains. A robust framework should be Topic-Adaptive, capturing macro-level joint distributions while ensuring micro-level individual rationality. Existing approaches fall into two categories: static data-based retrieval methods that fail to adapt to unseen topics absent from the data, and LLM-based generation methods that lack macro-level distribution awareness, resulting in inconsistencies between micro-level persona attributes and reality. To address these problems, we propose HAG, a Hierarchical Agent Generation framework that formalizes population generation as a two-stage decision process. Firstly, utilizing a World Knowledge Model to infer hierarchical conditional probabilities to construct the Topic-Adaptive Tree, achieving macro-level distribution alignment. Then, grounded real-world data, instantiation and agentic augmentation are carried out to ensure micro-level consistency. Given the lack of specialized evaluation, we establish a multi-domain benchmark and a comprehensive PACE evaluation framework. Extensive experiments show that HAG significantly outperforms representative baselines, reducing population alignment errors by an average of 37.7% and enhancing sociological consistency by 18.8%.

Figures (6)

• Figure 1: The necessity of Topic-Adaptive in simulation.
• Figure 2: Illustration of the HAG framework. It utilizes the World Knowledge Model to construct a Topic-Adaptive Demographic Distribution Tree, and generates population based on the tree via filtering real users and agentic data augmentation. Evaluation of the generated population is from two aspects: population alignment and sociological consistency.
• Figure 3: Visualization of the Population Manifold Structure in Latent Sociological Space (t-SNE).
• Figure 4: Heatmaps of joint attribute distributions.
• Figure 5: Demonstration of the case study.