Institute for Software Research
School of Computer Science, Carnegie Mellon University
Detecting Changes in a Dynamic Social Network
A suite of computational and statistical approaches for detecting change are identified and compared. The Neyman-Pearson most powerful test of simple hypotheses is extended as a cumulative sum statistical process control chart to detect network change over time. Anomaly detection approaches using exponentially weighted moving average or scan statistics investigate performance under conditions of potential time-series dependence. Fourier analysis and wavelets are applied to a spectral analysis of social networks over time. Parameter values are varied for all approaches. The results are put in a computational decision support framework.
This new approach is demonstrated in multi-agent simulation as well as on eight different real-world data sets. The results indicate that this approach is able to detect change even with high levels of uncertainty inherent in the data. The ability to systematically, statistically, effectively and efficiently detect these changes has the potential to enable the anticipation of change, provide early warning of change, and enable faster appropriate response to change.