Graph-level learning

Graph-level learning refers to ML tasks and techniques that operate at the level of entire graphs rather than individual nodes or edges within a graph. Graph-level learning focuses on generating predictions, classifications, or insights based on the entire graph or a subgraph structure.

Graph-level prediction

The goal here is to make predictions or classifications for the entire graph rather than individual nodes or edges. For example, given a time-specific user-item interaction graph for e-commerce, we can predict any special events, or patterns in general, based on graph-level learning. In urban planning and transportation management, graph-level prediction can be employed to forecast traffic flow across an entire road network. This can enable the optimization of traffic signal timings, route planning, and infrastructure development.

Figure 2.8 – Event prediction based on a snapshot of an e-commerce graph

Figure 2.8...