Ji-Cather/GraphAgent

Implements multi-agent simulation using AgentScope with configurable LLM backends (OpenAI, vLLM) to generate text-attributed temporal graphs that match real-world structural and textual properties. The framework validates generated graphs against seven macroscopic network properties and demonstrates 11% improvement in microscopic structure metrics compared to existing graph generation methods. Supports diverse domains—social media, e-commerce, and citation networks—with parallel execution via distributed launcher architecture and optional user-prompt-driven configuration for rapid environment setup.