Skip to main content

Table 9 Accuracy (mean and standard deviation over 10 runs) obtained by initializing the entity features with mixed distributions

From: Graph convolutional and attention models for entity classification in multilayer networks

Network ML-GAT ML-GCN GAT GCN
Balance \(0.8377 \pm 0.0567\) \(\mathbf {0.9053 \pm 0.0168}\) \(0.6362 \pm 0.0346\) \(0.6714 \pm 0.0489\)
CKM-Social \(0.9473 \pm 0.0231\) \(0.8066 \pm 0.0399\) \(\mathbf {0.9962 \pm 0.0069}\) \(0.9786 \pm 0.0075\)
Congress \(0.8900 \pm 0.1294\) \(\mathbf {0.9468 \pm 0.0165}\) \(0.5458 \pm 0.1108\) \(0.6146 \pm 0.0000\)
DKPol \(0.7935 \pm 0.0245\) \(\mathbf {0.8205 \pm 0.0323}\) \(0.7956 \pm 0.0646\) \(0.7919 \pm 0.0424\)
Leskovec-Ng \(0.9944 \pm 0.0085\) \(0.9208 \pm 0.0297\) \(\mathbf {0.9951 \pm 0.0087}\) \(0.9861 \pm 0.0131\)
Starwars \(0.7435 \pm 0.0438\) \(0.7217 \pm 0.0625\) \(\mathbf {0.7782 \pm 0.0493}\) \(0.7202 \pm 0.0552\)
Terrorist-Noordin \(0.7133 \pm 0.0680\) \(\mathbf {0.7550 \pm 0.0643}\) \(0.7017 \pm 0.0337\) \(0.6983 \pm 0.0569\)
Terrorist-status \(0.4883 \pm 0.0629\) \(0.4700 \pm 0.1018\) \(\mathbf {0.5417 \pm 0.0568}\) \(0.5183 \pm 0.0678\)
Vickers \(\mathbf {0.9773 \pm 0.0321}\) \(0.9636 \pm 0.0469\) \(0.7727 \pm 0.1071\) \(0.8590 \pm 0.0452\)
Koumbia-2-mpx \(0.7162 \pm 0.0395\) \(\mathbf {0.7634 \pm 0.0092}\) \(0.7006 \pm 0.0171\) \(0.7020 \pm 0.0200\)
Koumbia-5-mpx \(\mathbf {0.8497 \pm 0.0092}\) \(0.8046 \pm 0.0212\) \(0.7759 \pm 0.0301\) \(0.7702 \pm 0.0143\)
Koumbia-10-mpx \(\mathbf {0.8579 \pm 0.0148}\) \(0.8228 \pm 0.0159\) \(0.8361 \pm 0.0123\) \(0.8052 \pm 0.0110\)
Koumbia-15-mpx \(\mathbf {0.8551 \pm 0.0075}\) \(0.8223 \pm 0.0105\) \(0.8539 \pm 0.0053\) \(0.8181 \pm 0.0121\)
Koumbia-20-mpx \(\mathbf {0.8561 \pm 0.0157}\) \(0.8305 \pm 0.0053\) \(0.8549 \pm 0.0155\) \(0.8312 \pm 0.0125\)
  1. Bold values refer to the best results on each network