Dynamic graph learning aims to model how entities and their relations evolve over time, using historical observations to predict future graph states. This is a core task for systems that need to anticipate changes in networks, such as social, biological, or transportation systems.
However, according to a new preprint on arXiv, existing methods do not fully address what the authors call "dynamic graph world modelling." This concept appears to require a more comprehensive understanding of the underlying dynamics, going beyond simple next-step prediction.
The abstract is truncated, so the full scope of the proposed approach is not yet visible. What is clear is that the authors identify a gap between current dynamic graph learning techniques and the broader goal of modelling the world's evolving structure.