Research Article

Semisupervised Classification with High-Order Graph Learning Attention Neural Network

Table 1

Comparison of results of various classification algorithms. Test accuracy (±standard deviation) on various classification data sets, expressed as a percentage. All experiments were repeated using 5 different random seeds. For each data set, the highest accuracy score is marked in bold. We compare HGLAT with several supervised benchmarks and semisupervised learning methods. HGLAT achieved the best results on 4 of the data sets.

Alg\datasetWineCancerDigitsCiteseerCora

LogReg92.1(1.3)93.3(0.5)85.5(1.5)62.2(0.0)60.8(0.0)
RF93.7(1.6)92.1(1.7)83.1(2.6)60.7(0.7)58.7(0.4)
RBF SVM94.1(2.9)91.7(3.1)86.9(3.2)60.2(0.0)59.7(0.0)
Linear SVM93.9(1.6)90.6(4.5)87.1(1.8)58.3(0.0)58.9(0.0)

LP89.8(3.7)76.6(0.5)91.9(3.1)23.2(6.7)37.8(0.2)
SemiEmb91.9(0.1)89.7(0.1)90.9(0.1)68.1(0.1)63.1(0.1)

kNN-GCN93.2(3.1)93.8(1.4)91.3(0.5)70.9(0.5)81.2(0.5)
kNN-GAT94.5(2.5)93.1(1.2)90.5(0.8)72.1(0.7)82.6(0.7)

HGLAT97.5(0.3)94.8(1.1)91.0(0.7)74.4(0.5)84.7(0.4)