Deep graph neural networks for link prediction

Abstract

Graph neural networks (GNNs) is a form of machine learning architecture that uses many neurons to learn a given information which is similar to how a human brain works. It is also known as deep GNNs when there are many layers of information processing within the neural network architecture. GNNs can be used for many machine learning tasks and can be used for learning networks such as citation networks. In this project, the main focus will be the investigation of inference performance of GNN models for link prediction task. Research in GNNs in the recent years has been agile but there is not enough experiments and discussions on the different hyperparameters and architectures that are being implemented. A literature review of the different GNN models and architectures was conducted. Comparisons between using different hyperparameters and architectures will be conducted for analyzing and discussing the strengths and weaknesses of the different configurations and frameworks. Upon investigations of the results, it was determined that the different datasets, model parameters and hyperparameters affects the inference performance differently for GNN models.