A graph neural network (GNN) is a model that captures dependencies of graphs via message passing between the nodes of graph. As opposed to standard neural network, a graph neural network retains a state representing information from its neighborhood with arbitrary depth.
A graph is a data structure that is comprised out of nodes, which are vertices, and edges connected together to represent information without a definite beginning or end. The nodes occupy an arbitrary position in space, and are often clustered according to similar features when they are plotted in a 2D space.
Graphs can either be directed and undirected. Directed graphs hold connections that dictate direction between the nodes, and can be both unidirectional or bidirectional in nature. With undirected graphs, connection order doesn't matter. Graphs can represent social media networks, molecules, and various other subjects.
Each node has a set of features that define it, which can be age, gender, political leaning, etc. Edges can connect nodes together that have similar features, this shows a type of interaction or relationship between them.
