ABSTRACT
This paper proposes GlobAlign-T, an unsupervised method extending global representation and optimal transport to temporal dynamic graphs. It introduces a temporal-aware self-attention mechanism, a time-conditioned hierarchical transport cost, and an efficient sparse temporal optimal transport algorithm, aiming to improve accuracy and efficiency in aligning dynamic graphs.
PAPER · PDF
Loading PDF…
Key findings
GlobAlign-T captures both structural and temporal dependencies across node pairs.
Introduces a transport cost function that respects temporal evolution patterns.
Develops a sparsification strategy for temporal graphs, reducing OT computation complexity.
Limitations & open questions
The paper does not discuss potential limitations of the proposed method.