AdaSTORM: Scaling LLM Reasoning on Dynamic Graphs via Adaptive Spatio-Temporal Multi-Agent Collaboration

The paper introduces AdaSTORM, a novel framework designed to enhance large language model (LLM) reasoning on dynamic graphs by employing adaptive spatio-temporal multi-agent collaboration. It overcomes scaling limitations by partitioning large graphs into manageable subregions and utilizing a multi-agent architecture for collaborative reasoning, enabling effective processing of graphs with thousands of nodes and achieving over 90% accuracy. This advancement is significant for practitioners, as it allows for more efficient reasoning in complex graph-structured tasks without the need for external tools, outperforming existing methods and achieving state-of-the-art results on benchmarks.