Information assurances and threat identification in networked organizations

摘要

We present a brief report on a controlled experiment that provides valuable statistics to network-oriented defence analysts involved in threat identification. These statistics estimate the accuracy of the top-central actor findings that have been derived from relational data classically found in real-world datasets, such as those collected on distributed, covert organizations. Our experiment involved cellular social-networks with four types of data error: missing links, missing actors, extra links, and extra actors. We provide statistical results for top threat identification from the perspective of four traditional measures of network centrality: degree, betweenness, closeness and eigenvector. The results from our experiment provide a statistical estimate of the accuracy of the top-1 and top-3 actors as indicated by the observed data. Using these statistics a quantitative indication of reliability can be provided along with defence intelligence estimates of covert-organization leadership derived from relational network data. We provide lookup tables for the specific situations created for this experiment, from which other conditions may be loosely estimated. This work has highly practical implications for operational analysts and consumers of such analyses, particularly in the terrorist network and drug-trafficking domains. This work also lays the groundwork for developing more intricate estimates of reliability for other network-related, analytic tasks of analysts — from more extensive key-actor identification tasks to assessing the statistical reliability of the centrality measures in and of themselves.