An inclusive analysis for performance and efficiency of graph neural network models for node classification

Abstract

Graph Neural Networks (GNNs) have become a prominent technique for the analysis of graph-based data and knowledge extraction. This data can be either structured or unstructured. GNN approaches are particularly beneficial when it comes to examining non-euclidean data. Graph data formats are well-known for their capability to represent intricate systems and understand their relationships. GNNs have significantly advanced the field of research because of their numerous possible applications in machine learning for tasks involving graph-structured data, where relationships between entities play a crucial role. GNN can carry out numerous tasks, such as classifying nodes, categorizing graphs, predicting links or relationships, and much more. Node classification is a widely used and recognized GNN task that has reached state-of-the-art performance on a number of benchmark datasets. In this study, we have provided a comprehensive insight into GNN, its development, and an extensive review of node classification, along with experimental findings and discussions.