Ensemble Supervised learning (svm) Input: meta-graph generated labels(soft labels Output: ground truth labels(partially labeled ones) EM [Dawid and Skene, 1979 E-step: estimate posterior of label assignment of each meta-graph label M-step: estimate label cluster probabilities, and likelihood of label assignment of each meta-graph label Co-training [Wan et al. SDM15 Train the weight of each meta-graph Update the label assignment of each random walk Meta-graph 1 Meta-graph 2 Meta-graph G Label 1 label 2 label 1 label 2 Label 1 label 2 Doc 1 0.9 0.1 0.1 0.8 0.9 0.2 Doc 2 0.9 0.2 0.8 0.1 0.6 0.5 DOC N 0.20.70.1 0.6 0.3 0.6Ensemble • Supervised learning (SVM) – Input: meta-graph generated labels (soft labels) – Output: ground truth labels (partially labeled ones) • EM [Dawid and Skene, 1979] – E-step: estimate posterior of label assignment of each meta-graph label – M-step: estimate label cluster probabilities, and likelihood of label assignment of each meta-graph label • Co-training [Wan et al., SDM’15] – Train the weight of each meta-graph – Update the label assignment of each random walk 26 Meta-graph 1 Meta-graph 2 … Meta-graphG Label 1 Label 2 Label 1 Label 2 Label 1 Label 2 Doc 1 0.9 0.1 0.1 0.8 0.9 0.2 Doc 2 0.9 0.2 0.8 0.1 0.6 0.5 … Doc N 0.2 0.7 0.1 0.6 0.3 0.6