This paper introduces a novel approach for identifying frequent subgraph patterns by combining two frequently occurring smaller subgraph patterns, and proposes a new metric based on Maximal Independent Sets to efficiently enumerate pattern graphs within a data graph.
The author argues that compressing background nodes in web graphs can significantly enhance efficiency and performance in target node classification tasks. The proposed Graph-Skeleton model effectively condenses background nodes while maintaining target nodes, leading to superior results compared to other compression methods.
The author proposes a novel approach, GC-SNTK, utilizing Kernel Ridge Regression and Structure-based Neural Tangent Kernel for efficient graph condensation, demonstrating superior performance and time efficiency compared to existing methods.