A Graph Neural Network (GNN) is a class of artificial neural networks for processing data that can be represented as graphs. In the more general class of Graph Neural Networks, the graph structure (nodes and edges) is directly used as input. GNNs are designed to capture the dependencies in graphs via message passing between the nodes of graphs. Unlike standard neural networks that expect data in a fixed grid (like images) or sequence (like text), GNNs can operate on irregular data structures like social networks, molecular structures, and knowledge graphs.
The concept of GNNs was first introduced by Scarselli et al. in 2008. However, modern variants like Graph Convolutional Networks (GCNs) gained significant traction around 2016-2017 with work by Kipf and Welling.
GNNs have become essential in fields requiring relational reasoning. They are used in drug discovery (predicting molecular properties), recommendation systems (Pinterest's PinSage), traffic prediction (Google Maps), and social network analysis.