A space-time varying graph for modelling places and events in a network

Modelling topological relationships between places and events is challenging especially because these relationships are dynamic, and their evolutionary analysis relies on the explanatory power of representing their interactions across different temporal resolutions. In this paper, we introduce the Space-Time Varying Graph (STVG) based on the whole graph approach that combines directed and bipartite subgraphs with a time-tree for representing the complex interaction between places and events across time. We demonstrate how the proposed STVG can be exploited to identify and extract evolutionary patterns of traffic accidents using graph metrics, ad-hoc graph queries and clustering algorithms. The results reveal evolutionary patterns that uncover the places with high incidence of accidents over different time resolutions, reveal the main reasons why the traffic accidents have occurred, and disclose evolving communities of densely connected traffic accidents over time.