Graph neural network for link prediction

Abstract

Link prediction estimates two nodes in a graph are linked or not. Link prediction can use it in different aspects, such as social networks or biological networks. The link prediction method is helpful to find some new or missing relation between other nodes. However, there has no function that is common and powerful that suitable for all kinds of graphs. In this project, the graph neural network (GNN) will be used as the python program's training method. GNN will split the graph into different small parts and train to learn its characteristics and using the result it gets to predict the link is connect or not. Next, the GNN method will compare with other traditional methods to verify that can GNN have a better performance on different graphs. In conclusion, the final result shows that the GNN method has a more stable and relatively high performance than traditional methods on different graphs.