AEDNet: Adaptive Edge-Deleting Network For Subgraph Matching

Subgraph matching is to find all subgraphs in a data graph that are isomorphic an existing query graph. NP-hard problem, yet has found its applications many areas. Many learning-based methods have been proposed for matching, whereas few designed subgraph matching. The problem generally more challenging, mainly due the different sizes between two graphs, resulting considerable large space of solutions. Also extra edges connecting matched nodes may lead graphs having adjacency structures and often being identified as distinct objects. Due edges, learning based fail generate sufficiently similar node-level embeddings nodes. This study proposes novel Adaptive Edge-Deleting Network (AEDNet) method trained end-to-end fashion. In AEDNet, sample-wise adaptive edge-deleting mechanism removes ensure consistency structure nodes, while unidirectional cross-propagation ensures features We applied on six datasets with varying from 20 2300. Our evaluations open demonstrate AEDNet outperforms state-of-the-arts much faster than exact graphs.