Abstract
Graph neural networks denote a group of neural network models introduced forthe representation learning tasks on graph data specifically. Graph neuralnetworks have been demonstrated to be effective for capturing network structureinformation, and the learned representations can achieve the state-of-the-artperformance on node and graph classification tasks. Besides the differentapplication scenarios, the architectures of graph neural network models alsodepend on the studied graph types a lot. Graph data studied in research can begenerally categorized into two main types, i.e., small graphs vs. giantnetworks, which differ from each other a lot in the size, instance number andlabel annotation. Several different types of graph neural network models havebeen introduced for learning the representations from such different types ofgraphs already. In this paper, for these two different types of graph data, wewill introduce the graph neural networks introduced in recent years. To be morespecific, the graph neural networks introduced in this paper include IsoNN,SDBN, LF&ER, GCN, GAT, DifNN, GNL, GraphSage and seGEN. Among these graphneural network models, IsoNN, SDBN and LF&ER are initially proposed for smallgraphs and the remaining ones are initially proposed for giant networksinstead. The readers are also suggested to refer to these papers for detailedinformation when reading this tutorial paper.