Algorithm 1: RF-RFE-SELLP algorithm

Input: Network $G=\left(V,E\right)$  Output: Parameters of SELLP model 1: Calculate the adjacency matrix $A$ of network G. 2: for each $v_i,v_j\in V$ do 3: $y_n\leftarrow A\left(:\right)$ 4: $x_n\leftarrow \left(x^{S_1}\left(v_i,v_j\right),\dots ,x^{S_{12}}\left(v_i,v_j\right)\right)$ 5:/*$y_n$ is the label, $x_n$ is the feature vector, and $x^{S_l}\left(v_i,v_j\right)$ is the score computed by the similarity index $S_l$.*/ 6: end for 7: for $r\leftarrow 1$ to 12 do 8: Compute the VIM of features, and remove the feature with the lowest one. 9: Update feature ranked list, and calculate the corresponding CVS. 10: end for 11: Obtain the optimal feature subset $F_n$ by comparing CVS. 12: Construct the set $S$ including the feature subset $F_n$ and label $y_n$, and then divide $S$ into the training set $S^T$ and test set $S^P$. 13: for $k\leftarrow 1$ to 5 do 14: for $j\leftarrow 1$ to 3 do 15: Train the base model $C_{jk}$ on the training set $S_{-k}^T$. 16: Predict the probability ${\widehat{y}}_{jk}$ on the test set $S_k^T$. 17: end for 18: end for 19: Construct the set $S_{new}^T=\left\{\left({\widehat{y}}_{1i},{\widehat{y}}_{2i},{\widehat{y}}_{3i},y_i\right),i=1,\dots ,N^T\right\}$ by the prediction results of each base model. 20: Train the fuse model of the second layer on the set $S_{new}^T$. 21: Return the parameters of RF-RFE-SELLP model.