WorldGraph models changing graphs to predict future states
WorldGraph: Graph-Native World Modeling
Machine LearningArtificial IntelligenceComputer Vision and Pattern Recognition
Summary
Many environments can be seen as networks that change over time, like social connections or transportation routes. Previous models used graphs to help understand or reason about these settings, but didn't treat the changing graph itself as the main thing to predict. The authors create a new way to model how graphs evolve, capturing changes to nodes, edges, and overall structure. They build a benchmark to test such models and propose WorldGraph, which better predicts how these graph worlds change compared to other methods.
What this means in practice
- •For network operators: Predict how network topologies will evolve to improve maintenance and upgrades.
- •For social media platform engineers: Model user and interaction changes over time to better understand and anticipate social dynamics.
Authors
Zezhong Ding, Yipeng Li, Xike Xie
Abstract
World models infer latent states of an environment to capture its underlying dynamics and predict future evolution. Many real-world environments, however, are inherently relational and observed as evolving graphs, where entities, relations, and their properties change over time. Prior graph-related world models use graph structures to organize internal states or support task-specific reasoning, rather than treating an evolving graph itself as the modeled world. We instead study graph world modeling (GWM), where graph evolution itself constitutes the world dynamics. We formulate graph world modeling over observed graph evolution, latent graph states, and heterogeneous graph-transition predictions. Based on this formulation, we construct GWM-Zero, a benchmark covering node-, edge-, and graph-level transitions over eight temporal graph datasets. We propose WorldGraph, which combines a state-aware graph transformer for multi-granularity structural and transition-conditioned evolution modeling with transition-aware GRPO using dynamic grouping and structure-aware verifiable rewards. Extensive experiments on GWM-Zero show that WorldGraph consistently outperforms representative graph representation, temporal graph learning, graph pretraining, and graph world-model baselines across all three transition granularities.