Discovering and Summarising Regions of Correlated Spatio-Temporal Change in Evolving Graphs

Graphs are adept at describing relational data, hence their popularity in fields including network management, Webpage analysis and sociology. However, most of the current graph mining work regards graphs as static and unchanging, even though many graphs are dynamic. In this paper, we introduce a new pattern to discover from evolving graphs, namely regions of the graph that are evolving in a correlated manner. These regions of correlated spatio-temporal change group together graph changes that are topologically near (spatial) and evolve similarly (temporal) to each other. The regions can be used to summarise changes, particularly for graphs that have many simultaneous changes. We have developed an algorithm called cSTAG to summarise changes in dynamic graphs. This new algorithm discovers these regions of correlated change and identifies events that caused these changes. As a demonstration of the effectiveness of our algorithm, we applied cSTAG to summarise the changes to the border gateway protocol connectivity graph during the 2005 Hurricane Katrina disaster. cSTAG was able to identify the reported failures in Louisiana, as well as other simultaneous events