Mixture-of-Personas Language Models for Population Simulation

2025-04-07

AI Summary

Pretrained large language models (LLMs) struggle to capture the behavioral diversity of real human populations, limiting their credibility in social science research and human behavior modeling. To address this, we propose the probabilistic Mixture-of-Personas (MoP) prompting framework—a novel, tuning-free method that models subgroup-level personality traits and behavioral exemplars as context-aware, learnable-weighted mixture components, ensuring cross-model generalizability. MoP integrates personality-driven agent design, stratified behavioral exemplar sampling, and weighted response aggregation to achieve fine-grained alignment with target population distributions. Experiments on synthetic data generation demonstrate that MoP significantly improves demographic alignment and behavioral diversity metrics, outperforming state-of-the-art prompting and fine-tuning baselines across multiple dimensions. By enabling scalable, low-barrier population-aligned modeling without architectural modification or parameter updates, MoP establishes a new paradigm for social computing and human-AI collaborative behavioral modeling.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Machine Learning: Mixture of Experts (MoE)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Advances in Large Language Models (LLMs) paved the way for their emerging applications in various domains, such as human behavior simulations, where LLMs could augment human-generated data in social science research and machine learning model training. However, pretrained LLMs often fail to capture the behavioral diversity of target populations due to the inherent variability across individuals and groups. To address this, we propose extit{Mixture of Personas} (MoP), a extit{probabilistic} prompting method that aligns the LLM responses with the target population. MoP is a contextual mixture model, where each component is an LM agent characterized by a persona and an exemplar representing subpopulation behaviors. The persona and exemplar are randomly chosen according to the learned mixing weights to elicit diverse LLM responses during simulation. MoP is flexible, requires no model finetuning, and is transferable across base models. Experiments for synthetic data generation show that MoP outperforms competing methods in alignment and diversity metrics.

Problem

• LLMs lack behavioral diversity in population simulations
• Aligning LLM responses with target population variability
• Generating diverse synthetic data without model finetuning

Innovation

• Probabilistic prompting for population alignment
• Contextual mixture model with persona agents
• No model finetuning, transferable across models