Table 3.

Accuracy of semi-supervised node classification on Cora.

Method	90%	80%	70%	60%	50%	40%	30%	20%	10%
GCN	0.842	0.842	0.828	0.828	0.821	0.821	0.807	0.807	0.800
αLoNGAE	0.803	0.793	0.790	0.783	0.780	0.777	0.770	0.767	0.763
GAT	0.824	0.822	0.816	0.808	0.806	0.804	0.798	0.796	0.794
MT-GAT (ours)	0.874	0.864	0.861	0.856	0.855	0.850	0.848	0.832	0.827

The best results are shown in bold, and our MT-GAT with significant improvements over the baselines is shown with underlines.