Introducing dynamic graphs
Dynamic graphs and temporal GNNs unlock a variety of new applications, such as transport and web traffic forecasting, motion classification, epidemiological forecasting, link prediction, power system forecasting, and so on. Time series forecasting is particularly popular with this kind of graph, as we can use historical data to predict the system’s future behavior.
In this chapter, we focus on graphs with a temporal component. They can be divided into two categories:
- Static graphs with temporal signals: The underlying graph does not change, but features and labels evolve over time.
- Dynamic graphs with temporal signals: The topology of the graph (the presence of nodes and edges), features, and labels evolve over time.
In the first case, the graph’s topology is static. For example, it can represent a network of cities within a country for traffic forecasting: features change over time, but the connections stay the same.
...
