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

In this work, we propose GraphAgent-Generator (GAG), a social simulation framework tailored for generating social graphs. Inspired by the bipartite modeling approach of affiliation networks, GAG simulates the dynamic expansion of social networks by modeling pairwise interactions between agents and items. We designed the S-RAG algorithm, combined with parallel acceleration, to efficiently generate social graphs. Source Code.