The Impact of Global Structural Information in Graph Neural Networks Applications

Graph Neural Networks (GNNs) rely on the graph structure to define an aggregation strategy where each node updates its representation by combining information from neighbours. A known limitation of GNNs is that, as number layers increases, gets smoothed and squashed embeddings become indistinguishable, negatively affecting performance. Therefore, practical GNN models employ few only leverage in terms limited, small neighbourhoods around node. Inevitably, do not capture depending global graph. While there have been several works studying limitations expressivity GNNs, question whether applications structured data require structural knowledge or remains unanswered. In this work, we empirically address giving access models, observing impact it has downstream Our results show that can fact provide significant benefits for common graph-related tasks. We further identify a novel regularization leads average accuracy improvement more than 5% all considered