Skip to main content

Table 8 Accuracy (mean and standard deviation over 10 runs) obtained by the proposed methods and competitors, with early-stopping regularization technique

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

Network ML-GAT ML-GCN GrAMME-SG GrAMME-Fusion GAT GCN
Balance \(0.7642 \pm 0.0416\) \(\mathbf {0.7759 \pm 0.0244}\) \(0.5748 \pm 0.1137\) \(0.6555 \pm 0.0711\) \(0.6046\pm 0.1117\) \(0.7282 \pm 0.0451\)
CKM-Social \(0.8771 \pm 0.0228\) \(0.7882 \pm 0.0424\) \(0.7293 \pm 0.0779\) \(0.8882 \pm 0.0250\) \(\mathbf {0.9512 \pm 0.0307}\) \(0.9288 \pm 0.0331\)
Congress \(0.9111 \pm 0.0271\) \(\mathbf {0.9370 \pm 0.0114}\) \({0.8683 \pm 0.0210}\) \(0.8581\pm 0.0107\) \(0.6131 \pm 0.0000\) \(0.6131 \pm 0.0000\)
DKPol \(0.6234 \pm 0.0566\) \(0.6055 \pm 0.0343\) \(0.3345 \pm 0.0587\) \(0.5392 \pm 0.1015\) \(0.5992 \pm 0.1034\) \(\mathbf {0.7789 \pm 0.0421}\)
Leskovec-Ng \(\mathbf {0.9828 \pm 0.0145}\) \(0.8918 \pm 0.0376\) \(0.6673 \pm 0.1151\) \(0.8110 \pm 0.0861\) \(0.9724 \pm 0.0190\) \(0.9746 \pm 0.0223\)
Starwars \(0.7894 \pm 0.0353\) \(\mathbf {0.8121 \pm 0.0106}\) \(0.6156 \pm 0.0109\) \(0.6656 \pm 0.0650\) \(0.7500 \pm 0.0464\) \(0.7318 \pm 0.0982\)
Terrorist-Noordin \(0.7036 \pm 0.0339\) \(\mathbf {0.7214 \pm 0.0674}\) \(0.6473 \pm 0.1009\) \(0.6618 \pm 0.0419\) \(0.6625 \pm 0.0586\) \(0.6893 \pm 0.0620\)
Terrorist-status \(0.5070 \pm 0.0791\) \(0.5554\pm 0.0638\) \(0.4873 \pm 0.1131\) \(0.4764 \pm 0.0505\) \(\mathbf {0.5643 \pm 0.0192}\) \(0.4921 \pm 0.1202\)
Vickers \(\mathbf {0.9571 \pm 0.0417}\) \(0.9143 \pm 0.0438\) \(0.6842 \pm 0.754\) \(0.7211 \pm 0.1907\) \(0.6428\pm 0.1510\) \(0.6714 \pm 0.1132\)
Koumbia-2-mpx \(\mathbf {0.6540 \pm 0.0258}\) \(0.6376 \pm 0.0250\) \({0.5641 \pm 0.0211}\) \({0.6306 \pm 0.1085}\) \(0.5662 \pm 0.0394\) \(0.5784 \pm 0.0110\)
Koumbia-5-mpx \(\mathbf {0.7585 \pm 0.0501}\) \(0.6715 \pm 0.0177\) \(0.5525 \pm 0.0088\) \(0.7480 \pm 0.0401\) \(0.6051 \pm 0.0433\) \(0.6331 \pm 0.0234\)
Koumbia-10-mpx \(\mathbf {0.8080 \pm 0.0445}\) \(0.7019 \pm 0.0493\) na na \(0.6613 \pm 0.0620\) \(0.6681 \pm 0.0305\)
  1. Training, validation and testing set sizes correspond to 5%, 25% and 70% of the entities, respectively
  2. Bold values refer to the best results on each network