arxiv_cl 92% Match Research Paper AI Researchers,Social Scientists,Data Scientists,Simulation Modelers,Economists 19 hours ago

Rethinking LLM Human Simulation: When a Graph is What You Need

graph-neural-networks › graph-learning

📄 Abstract

Abstract: Large language models (LLMs) are increasingly used to simulate humans, with applications ranging from survey prediction to decision-making. However, are LLMs strictly necessary, or can smaller, domain-grounded models suffice? We identify a large class of simulation problems in which individuals make choices among discrete options, where a graph neural network (GNN) can match or surpass strong LLM baselines despite being three orders of magnitude smaller. We introduce Graph-basEd Models for human Simulation (GEMS), which casts discrete choice simulation tasks as a link prediction problem on graphs, leveraging relational knowledge while incorporating language representations only when needed. Evaluations across three key settings on three simulation datasets show that GEMS achieves comparable or better accuracy than LLMs, with far greater efficiency, interpretability, and transparency, highlighting the promise of graph-based modeling as a lightweight alternative to LLMs for human simulation. Our code is available at https://github.com/schang-lab/gems.

Key Contributions

This paper proposes GEMS (Graph-basEd Models for human Simulation), a Graph Neural Network-based framework that matches or surpasses LLM performance in simulating human discrete choices, while being three orders of magnitude smaller. By framing simulation tasks as link prediction problems on graphs, GEMS leverages relational knowledge effectively, offering greater efficiency, interpretability, and transparency compared to LLMs for specific simulation problems.

Business Value

Enables cost-effective and understandable simulation of human behavior for applications like market analysis, policy testing, and personalized user experiences, without the high computational cost of LLMs.

Paper Metadata

Innovation Type

Algorithmic Framework

Deployment Feasibility

High, particularly for specific types of simulation tasks.

Limitations Addressed

LLMs being unnecessarily large and computationally expensive for certain simulation tasks,Lack of interpretability and transparency in LLM-based simulations,Need for more efficient and domain-grounded simulation models

Performance Gains

Achieves comparable or better accuracy than LLMs with significantly greater efficiency, interpretability, and transparency.

Technical Tags

LLM simulationhuman simulationgraph neural networksdiscrete choicelink predictionrelational knowledgelightweight alternativeGEMSinterpretabilitytransparency

Research Topics

Graph Neural NetworksHuman Behavior ModelingAgent-Based SimulationLLM AlternativesDecision Making

Methods & Architectures

Graph-basEd Models for human Simulation (GEMS)Link Prediction FormulationGraph Neural NetworksLanguage Representations (as needed) Graph Neural Networks (GNNs)Large Language Models (LLMs)

Applications & Tasks

Social Science Simulation Market Research Behavioral Economics Personalized Recommendations Overkill of LLMs for human simulationLack of efficiency and interpretability in LLM simulationsNeed for domain-grounded models Simulating Human ChoicesPredicting Survey ResponsesModeling Decision-Making Processes

Datasets & Benchmarks

Datasets

Three simulation datasets

Benchmarks

Comparable or better accuracy than LLMs • Three orders of magnitude smaller

AccuracyEfficiency (Model Size, Computation)InterpretabilityTransparency

Related Fields

Graph Neural NetworksMachine LearningSocial SciencesAgent-Based ModelingNatural Language Processing

Keywords

GNNLLMhuman simulationdiscrete choicelink predictiongraphrelational knowledgeefficiencyinterpretabilitybehavioral modelingagent-based modelingsimulation

Academic Context

#Graph Neural Networks#Human Behavior Modeling#Agent-Based Simulation#LLM Alternatives#Decision Making

Commercial Potential

Potential Products

Simulation Software for Social SciencesBehavioral Analytics PlatformsPersonalized Recommendation Engines

Target Industries

TechnologyMarket ResearchFinanceSocial MediaUrban Planning

Use Case Examples

Simulating consumer choices for new product launchesModeling traffic flow based on individual driver decisionsPredicting user engagement on social media platforms

Competitive Edge

Offers a specialized, efficient, and interpretable alternative to LLMs for human simulation tasks involving discrete choices.

Market Opportunity

Growing market for simulation and behavioral analytics tools.

Revenue Models

Licensing of simulation softwareconsulting services.

Resource Requirements

Compute Needs

Significantly lower than LLMs.

Data Requirements

Graph-structured data representing relationships and choices.

Deployment Constraints

Applicable primarily to simulation problems involving discrete choices and relational structures.

Scalability

GNNs are generally scalable, and the framework is designed for efficiency.

Production Readiness

Maturity Level

Algorithmic Framework

Time to Market

1-2 years for integration into simulation platforms

Licensing

Likely open-source framework.

Patent Potential

Moderate (for the GEMS framework)

View Full Paper Back to Papers