Semi-supervised node classification via adaptive graph smoothing networks

作者：

Highlights：

• A novel graph neural network is proposed, which learns to reduce inter-class edge weight and produces smooth predictions accordingly.

• Theoretical discussions of possible outcomes of the modified graphs under our framework are presented.

• Experiments and ablation studies were conducted to show the effectiveness of our model, comparing with state-of-the-art graph neural networks.

摘要

•A novel graph neural network is proposed, which learns to reduce inter-class edge weight and produces smooth predictions accordingly.•Theoretical discussions of possible outcomes of the modified graphs under our framework are presented.•Experiments and ablation studies were conducted to show the effectiveness of our model, comparing with state-of-the-art graph neural networks.

论文关键词：Adaptive graph smoothing networks,Graph convolutional networks,Semi-supervised learning,Graph node classification

论文评审过程：Received 1 June 2020, Revised 29 November 2021, Accepted 4 December 2021, Available online 11 December 2021, Version of Record 16 December 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108492