Robustness

By its nature, social network data is burdened by the underlying complexity of its subject matter and its collection procedures. While network analysts have made substantial progress in developing techniques to evaluate and analyze the data they have collected, errors in social-based data remains an inherent problem. In spite of the widely recognized existence of this problem, analysts continue to infer a great deal from their error-prone network data. One can only contemplate how mistaken past analyses may in fact be and what the impact on subsequent conclusions may have been. Mistaken data is a known problem but it remains one with unexplored impact.

The Robustness Project is motivated by the recognition that error is ubiquitous in social network data and that we actually have little comprehension of the impact of this error. Irrespective of calls being made over the years to study this problem, there has been little attention to the specific question of the robustness of common network measures -- certainly more attention is needed.

This project aims to increase our knowledge of the impact of erroneous network data on our network measures. Specifically, we seek to quantify the robustness of network centrality measures, relative to the network's topology and other characteristics, vis-à-vis inherently incorrect network data. We project that, given the known characteristics of a social network, analysts may ultimately quantify the impact of these errors on centrality measures and adjust their analyses accordingly. Eventually analysts may some day harvest more accurate information obtained from the likely /true/ network rather than from the erroneous /observed/ network

For more detailed information, papers, demos, data and results from Robustness Research; Visit the Robustness Project Working Page at: http://www.cs.cmu.edu/~terrill/robustness/