Link Prediction and Node Classification Based on Multitask Graph Autoencoder

Algorithm 2

 

Multitask graph autoencoder.

Input: the network with adjacency matrix , node labels, the parameters

Output: network representation and updated parameter

1: Apply Adam optimizer and ReLU activation function

2: Construct the similarity matrix

3: 

4: Repeat

5:  Based on , apply Equation (1) to obtain and

6:  

7:  

8:  

9:  

10:  Use to backpropagate through the whole network to obtain the parameter

11: Until converge

12: Obtain the network representations