**2.2 Graph neural networks**

GNNs were mentioned for the first time and further elaborated by [16]. The goal of a GNN is to learn a node's representation of the acquisition of its information by propagation. Currently, there are many deep learning tasks that need to process data with graph structures. Convolutional neural networks (CNNs) [17] have been successfully developed in the field of computer vision [18, 19] but are unable to process graph structured data [20]. The method used in this paper is called a graph convolutional network (GCN). A GCN can aggregate similar samples by propagating neighbor information, giving it the ability to infer, and there is no need to consider the sequence. GCNs have appeared in many top machine learning conferences and many applications across different tasks and domains, such as manifold learning [21, 22], computer vision [23–25], text classification [26, 27], hashing [28, 29], and hyperspectral image classification [30, 31].
