Divisive Betweenness Centrality Clustering on Graphs Weighted by Timestamps

Abstract: Data visualization is important to obtain a better understanding of many networks. However, for larger networks it is hard to perceive the visualization of the network. In order to simplify such visualizations of networks, graph clustering algorithms may be used to identify subgraphs of close vertices and group them together. In this paper we study graph clustering algorithms applied to event structures from continuous integration infrastructures. These event structures are special in the sense that they can be viewed as a directed acyclic graph with edges weighted by the duration between connected events. We have chosen to cluster this graph with variations of divisive graph clustering algorithms utilizing betweenness centrality measures, a measurement which originates from sociology. While some of the algorithms we tested produced acceptable clusters, our conclusion is that more theory behind the event structures is needed in order to achieve greater graph clustering qualities.