LLM-Based Multi-Agent Systems are Scalable Graph Generative Models

2024-10-13

AI Summary

Existing social graph generation methods face a fundamental trade-off: rule-based approaches lack realism and generalizability, while deep learning–based methods require extensive labeled data and struggle with large-scale, dynamic modeling. To address this, we propose GraphAgent-Generator (GAG), the first framework to leverage large language models (LLMs) as zero-shot, scalable graph generators—exploiting their inherent sociocultural commonsense knowledge via multi-agent collaborative simulation for dynamic, fine-grained, text-attributed social graph generation. GAG is the first method to simultaneously satisfy seven macroscopic network properties and achieve high-fidelity microscopic structure in a zero-shot setting. It improves structural graph metrics by 11%, scales to graphs with up to 100K nodes and 1M edges, and achieves a parallel speedup of 90.4×. Node classification experiments empirically confirm strong alignment between textual semantics and graph topology.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The structural properties of naturally arising social graphs are extensively studied to understand their evolution. Prior approaches for modeling network dynamics typically rely on rule-based models, which lack realism and generalizability, or deep learning-based models, which require large-scale training datasets. Social graphs, as abstract graph representations of entity-wise interactions, present an opportunity to explore network evolution mechanisms through realistic simulations of human-item interactions. Leveraging the pre-trained social consensus knowledge embedded in large language models (LLMs), we present GraphAgent-Generator (GAG), a novel simulation-based framework for dynamic, text-attributed social graph generation. GAG simulates the temporal node and edge generation processes for zero-shot social graph generation. The resulting graphs exhibit adherence to seven key macroscopic network properties, achieving an 11% improvement in microscopic graph structure metrics. Through the node classification benchmarking task, we validate GAG effectively captures the intricate text-structure correlations in graph generation. Furthermore, GAG supports generating graphs with up to nearly 100,000 nodes or 10 million edges through large-scale LLM-based agent simulation with parallel acceleration, achieving a minimum speed-up of 90.4%. The source code is available at https://github.com/Ji-Cather/GraphAgent.

Problem

• Social Graph Modeling
• Rule-based Approaches
• Deep Learning Limitations

Innovation

• GraphAgent-Generator
• Dynamic Social Graph Generation
• Efficient Large-scale Modeling