CrowdLLM: Building LLM-Based Digital Populations Augmented with Generative Models

2025-12-02

AI Summary

Existing LLM-driven digital population approaches struggle to simultaneously ensure behavioral authenticity and group diversity. This paper proposes a collaborative framework integrating large language models (LLMs) with generative models, jointly optimizing heterogeneous, cross-domain behavioral signals (e.g., voting, rating) to achieve high-fidelity modeling of real-world population distributions. Key contributions include: (1) leveraging a generative model to calibrate statistical biases in LLM outputs, thereby enhancing individual behavioral diversity and aggregate population representativeness; and (2) introducing a distribution-matching mechanism enabling scalable, task-agnostic simulation across domains—including social simulation, crowdsourcing, and recommender systems. Experiments demonstrate that our method closely approximates empirical population distributions across multiple domains, outperforming pure-LLM baselines on both accuracy and diversity metrics. The framework establishes a novel paradigm for constructing low-cost, high-fidelity synthetic populations.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The emergence of large language models (LLMs) has sparked much interest in creating LLM-based digital populations that can be applied to many applications such as social simulation, crowdsourcing, marketing, and recommendation systems. A digital population can reduce the cost of recruiting human participants and alleviate many concerns related to human subject study. However, research has found that most of the existing works rely solely on LLMs and could not sufficiently capture the accuracy and diversity of a real human population. To address this limitation, we propose CrowdLLM that integrates pretrained LLMs and generative models to enhance the diversity and fidelity of the digital population. We conduct theoretical analysis of CrowdLLM regarding its great potential in creating cost-effective, sufficiently representative, scalable digital populations that can match the quality of a real crowd. Comprehensive experiments are also conducted across multiple domains (e.g., crowdsourcing, voting, user rating) and simulation studies which demonstrate that CrowdLLM achieves promising performance in both accuracy and distributional fidelity to human data.

Problem

• Enhances digital population diversity and fidelity
• Integrates LLMs with generative models for accuracy
• Reduces costs and improves representation in simulations

Innovation

• Integrates pretrained LLMs with generative models
• Enhances diversity and fidelity of digital populations
• Achieves accuracy and distributional fidelity to human data