What is it about?
While Large Language Models are excellent at processing language, their ability on graph analysis is not well understood. We test these models on community search—the task of finding groups within a network. We found that they struggle to provide reliable answers. Our work introduces a new method where two agents collaborate to solve this problem. It significantly improve the LLMs' ability to solve community search problems without fine-tuning.
Featured Image
Photo by Logan Voss on Unsplash
Why is it important?
This work is the first to bridge LLMs with community search in graph analysis. Our unique CS-Agent framework enables LLMs to self-correct and refine results through multi-agent dialogue, without any fine-tuning. This provides a novel, robust, and immediately applicable solution for complex graph analysis tasks.
Perspectives
hope this work inspires readers to see the untapped potential of LLMs in understanding complex graph structures. It's exciting to pioneer a path where language models can collaboratively reason about and refine their analysis, making sophisticated graph insights more accessible.
jiahao hua
Nanjing University of Science and Technology
Read the Original
This page is a summary of: CS-Agent: LLM-based Community Search via Dual-agent Collaboration, November 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3746252.3761335.
You can read the full text:
Resources
Contributors
The following have contributed to this page
