Generalizing Weisfeiler-Lehman Kernels to Subgraphs

The introduction of WLKS marks a significant advancement in subgraph representation learning, addressing the shortcomings of existing graph neural networks (GNNs) that often yield suboptimal results for subgraph-level tasks. By applying the Weisfeiler-Lehman algorithm on induced k-hop neighborhoods, WLKS captures intricate interactions within and between subgraphs, offering a more expressive and efficient alternative. In extensive experiments across eight real-world and synthetic benchmarks, WLKS demonstrated its superiority by significantly outperforming leading approaches on five datasets and achieving a remarkable reduction in training time, ranging from 0.01x to 0.25x compared to state-of-the-art methods. This development not only enhances the capabilities of GNNs but also opens new avenues for solving complex problems in various domains, emphasizing the importance of efficient and expressive learning methods in the evolving landscape of artificial intelligence.