The author proposes Disentangled Intervention-based Dynamic graph Attention networks with Invariance Promotion (I-DIDA) to handle spatio-temporal distribution shifts in dynamic graphs by discovering and utilizing invariant patterns.
Dynamic graph neural networks (DGNNs) aim to bridge the gap between traditional static graph neural networks and the inherent temporal dependencies of real-world dynamic graphs, enabling more authentic modeling of complex network evolution.
This paper provides a comprehensive survey of the latest developments in dynamic graph neural networks (DGNNs), covering 81 DGNN models, 12 DGNN training frameworks, and commonly used benchmarks. It introduces a novel taxonomy to categorize DGNN models, presents detailed overviews of existing frameworks, and conducts thorough experimental comparisons of representative DGNN models and frameworks. The analysis and evaluation results identify key challenges and offer principles for future research to enhance the design of DGNN models and frameworks.