Autonomic Clustering in Temporal Network Graphs

Abstract

In this paper, we examine the use of autonomic clustering algorithms on temporal graph topologies representing mobile communication networks. We introduce basic group-based mobility scenarios including periods of overlapping group clusters and present both emulation and simulation models of these scenarios. From extracted temporal graph models, we demonstrate how periods of clustering overlap introduce specific challenges in the autonomic clustering of temporal graph models. We perform several group mobility experiments on classes of autonomic clustering approaches and we focus in on some high quality clustering algorithm performers including: Spectral clustering, multilevel clustering, and information theoretic clustering. We present quality metrics and examine basic measures of accuracy and stability and further demonstrate challenges associated with both measuring quality and effectively partitioning evolving graphs. We then demonstrate improvements in detecting the temporal “ground truth” clustering by the use of a time-windowed, weighted graph representation. We conclude with a discussion of future areas of work and summarize initial experiments.