Fig.1

Reviewer #1: (1) Number of labeled nodes to train the policy network. We have compared classification performance w.r.t. Empirically, we use Micro-F1 score of the classification GNN on test sets as28 the evaluation metricMtogenerate the reward signalR. We agree that it would be interesting to evaluate the proposed34 algorithmGPAonothertasks. The graph policy network selects unlabeled nodes for annotation to42 trainthenodeclassification network.