Customizing graph neural networks using path reweighting

Jianpeng Chen, Yujing Wang, Ming Zeng, Zongyi Xiang, Bitan Hou, Tong Yu, Ole J. Mengshoel, Yazhou Ren

Abstract: Highlights•We propose a novel graph neural network architecture named CustomGNN.•CustomGNN incorporates task-oriented semantic features with generic graph features.•Two unsupervised loss functions are proposed to regularize the learning procedure.•The semantics extracted from graph paths are visualized and analyzed.•CustomGNN can mitigate the issues of over-smoothing, non-robustness and overfitting.