Algorithm 1 Active learning for node classification.

Input: Graph $G=(A,X)$, Query budget K, Initial labels $Y_{\mathcal{L}}$ Output: An improved model $f_{\theta }$ for$i\leftarrow 1$ to $n_q=K$ do  Select the best unlabeled instance $q^{*}$ with an acquisition function g  Retrieve its label $Y_{q^{*}}$  Update label set $Y_{\mathcal{L}}\leftarrow Y_{\mathcal{L}}\cup Y_{q^{*}}$  Retrain the model $\theta \leftarrow {arg\; min}_{\theta }\mathit{l}(f_{\theta }\left(G\right),Y_{\mathcal{L}})$ end for Return $\theta$