Amr Elsisy made an oral presentation on "A synthetic network generator for covert network analytics," co-authored by Aamir Mandviwalla, Boleslaw K. Szymanski, and Thomas Sharkey at Networks'21 Satellite

Amr Elsisy made an oral presentation on "A synthetic network generator for covert network analytics," co-authored by Aamir Mandviwalla, Boleslaw K. Szymanski, and Thomas Sharkey at Networks'21 Satellite on Communities in Networks on July 1st, 2021. The talk focuses on covert social networks (also known as hidden) networks, such as terrorist or criminal networks. Their structures, memberships and activities are illegal. Thus, data about covert networks is often incomplete and partially incorrect, making interpreting structures and activities of such networks challenging. For legal reasons, real data about active covert networks is inaccessible to researchers. To address these challenges, we introduce here a network generator for synthetic networks that are statistically similar to a real network but void of personal information about its members. The generator uses statistical data about a real or imagined covert organization network. It generates randomized instances of the Stochastic Block model of the network groups but preserves this network organizational structure. The direct use of such anonymized networks are for training on them the research and analytical tools for finding structure and dynamics of covert networks. Since these synthetic networks differ in their sets of edges and communities, they can be used as a new source for network analytics. First, they provide alternative interpretations of the data about the original network. The distribution of probabilities for these alternative interpretations enables new network analytics. The analysts can find community structures which are frequent, therefore stable under perturbations. They may also analyze how the stability changes with the strength of perturbation. For covert networks, the analysts can quantify statistically expected outcomes of interdiction. This kind of analytics applies to all complex network in which the data are incomplete or partially incorrect. The proposed generator can help resolve the issue of data sharing, by maintaining the anonymity and privacy of networks. We can control how similar the generated networks are to the original network. Using the BWRN model, we generate a certain number N = 1000 synthetic networks, and compare their structures (communities) to each other and to the original network, Caviar. We test the quality of the generated networks by using Normalized Mutual Information to compare community structures, and by detecting leadership similarity in the generated networks using Jaccard metric. We find that the BWRN model preserves the community structure, and the leadership hierarchy better than the WRG model. Lastly, we can also measure the entropy of the generated networks, to better understand how it can differ between the generated networks. Entropy can be measures for individual nodes, to see how their behavior changes in different versions of the networks, or it can be measures for whole communities, to quantity how communities are changing in different versions of the network, and finally it can also be applied to edges, to see how the interactions between nodes or communities are changing in different versions of the network.