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

2026-06-15

AI Summary

Large language models (LLMs) struggle to reason over large-scale dynamic graphs due to exponential inference costs and limited context windows, typically supporting only dozens of nodes. This work proposes the first multi-agent framework tailored for dynamic graph reasoning, decomposing the task into two stages: adaptive graph partitioning and spatiotemporally decoupled collaborative reasoning. The former dynamically segments subgraphs according to the model’s capacity, while the latter enables efficient multi-agent coordination by aligning agents with the graph topology. Requiring no external tools, the method scales inference to graphs with thousands of nodes, achieving over 90% accuracy across multiple large-scale settings. It significantly outperforms seven strong baselines and establishes state-of-the-art performance on both established benchmarks and real-world datasets.

Abstract

Large Language Models (LLMs) demonstrate remarkable potential in dynamic graph reasoning, but suffer from a scaling bottleneck: current models can only handle graphs with tens of nodes, constrained by exponential reasoning overhead and finite context windows. While multi-agent systems (MAS) offer collective reasoning and topology-aware orchestration, capabilities naturally suited for graph-structured tasks, their application to dynamic graphs remains unexplored. This paper presents Scaling LLM Reasoning on Dynamic Graphs via Adaptive Spatio-Temporal Multi-Agent Collaboration (AdaSTORM), a framework that reformulates large-scale dynamic graph reasoning into two stages: (i) Adaptive Partitioning, partitioning large-scale dynamic graphs into subregions that match the model's reasoning capacity while minimizing inference cost; and (ii) Collaborative Reasoning, aligning graph partition topologies with a spatio-temporal decoupled multi-agent architecture. AdaSTORM is the first multi-agent framework tailored for dynamic graph reasoning. Extensive experiments show that AdaSTORM successfully breaks through the scaling bottleneck, scaling reasoning to thousand-node graphs with over 90% accuracy across several large-scale dynamic graph settings without external tools, significantly outperforms seven competitive baselines. Furthermore, it achieves state-of-the-art accuracy on existing benchmarks and generalizes robustly to real-world datasets. The source code is available at: https://github.com/irisorchid107/AdaSTORM/.

Problem

• dynamic graphs
• LLM reasoning
• scaling bottleneck
• multi-agent systems
• graph reasoning

Innovation

• dynamic graphs
• multi-agent systems
• adaptive partitioning
• spatio-temporal reasoning
• LLM scaling