Subgraph Mining for Graph Neural Networks

While Graph Neural Networks (GNNs) are state-of-the-art models for graph learning, they are only as expressive as the first-order Weisfeiler-Leman graph isomorphism test algorithm. To enhance their expressiveness one can incorporate complex structural information as attributes of the nodes in input graphs. However, this typically demands significant human effort and specialised domain knowledge. We demonstrate the feasibility of automatically extracting such information through subgraph mining and feature selection. Our experimental evaluation, conducted across graph classification tasks, reveals that GNNs extended with automatically selected features obtained using subgraph mining can achieve comparable or even superior performance to GNNs relying on manually crafted features.