基于卷积图神经网络的多粒度表示学习框架
|
张蕾, 钱峰, 赵姝, 陈洁, 杨雪洁, 张燕平
|
Multi⁃granular representation learning framework for Convolutional Graph Neural Networks
|
Lei Zhang, Feng Qian, Shu Zhao, Jie Chen, Xuejie Yang, Yanping Zhang
|
|
表3 半监督节点分类实验结果对比
|
Table 3 Experimental results of semi?supervised node classification by different algorithms
|
|
方法 | 指标 | Cora | CiteSeer | DBLP | PubMed |
---|
GCN | Mi⁃F1 | 0.834 | 0.736 | 0.825 | 0.830 | Ma⁃F1 | 0.808 | 0.683 | 0.767 | 0.827 | M⁃NRL (GCN) | Mi⁃F1 | 0.849 | 0.743 | 0.828 | 0.855 | Ma⁃F1 | 0.839 | 0.688 | 0.781 | 0.850 | Graph SAGE | Mi⁃F1 | 0.824 | 0.729 | 0.815 | 0.841 | Ma⁃F1 | 0.800 | 0.682 | 0.759 | 0.837 | M⁃NRL (SAGE) | Mi⁃F1 | 0.845 | 0.746 | 0.822 | 0.856 | Ma⁃F1 | 0.837 | 0.689 | 0.772 | 0.852 | GAT | Mi⁃F1 | 0.837 | 0.731 | 0.830 | 0.848 | Ma⁃F1 | 0.816 | 0.687 | 0.784 | 0.842 | M⁃NRL (GAT) | Mi⁃F1 | 0.840 | 0.744 | 0.837 | 0.857 | Ma⁃F1 | 0.828 | 0.688 | 0.793 | 0.852 | APPNP | Mi⁃F1 | 0.832 | 0.742 | 0.811 | 0.826 | Ma⁃F1 | 0.813 | 0.674 | 0.705 | 0.824 | M⁃NRL (APPNP) | Mi⁃F1 | 0.840 | 0.749 | 0.828 | 0.861 | Ma⁃F1 | 0.832 | 0.680 | 0.778 | 0.858 | TAGCN | Mi⁃F1 | 0.854 | 0.730 | 0.828 | 0.843 | Ma⁃F1 | 0.842 | 0.670 | 0.774 | 0.839 | M⁃NRL (TAGCN) | Mi⁃F1 | 0.863 | 0.748 | 0.832 | 0.856 | Ma⁃F1 | 0.847 | 0.696 | 0.777 | 0.853 | GFNN | Mi⁃F1 | 0.859 | 0.732 | 0.840 | 0.828 | Ma⁃F1 | 0.851 | 0.670 | 0.788 | 0.820 | M⁃NRL (GFNN) | Mi⁃F1 | 0.868 | 0.752 | 0.850 | 0.855 | Ma⁃F1 | 0.853 | 0.694 | 0.823 | 0.851 |
|
|
|