Table 4.

Graph learning methods.

Methods	Advantages	Limitations
Graph Covolutional Network107	Aggregate graph information and make the use of structural information of graphs.	Less computationally efficient for large graphs. Lack of generalization to graphs with different structure110. Oversmoothing of graph embeddings108.
Graph Attention Network110	Computationally efficient for node-level parallel processing. Don’t need the knowledge of the entire graph structure.	Can’t tell the differences between local and global structures well111.
(Variational) Graph Autoencoder112	Reduce data dimensionality and speed up the training process.	Captures more information from dataset, rather than the relevant information to the problem. And the reconstruction process loses information.
Graph Generative Adversarial Nets113	Can augment dataset and impute missing values.	Instability of gradient updates, and the vanished gradients of generator114, etc.