Accession Number : AD1051068

Title :   Measuring, Understanding, and Responding to Covert Social Networks: Passive and Active Tomography

Descriptive Note : Technical Report,01 Dec 2010,22 Nov 2016

Corporate Author : Harvard University Cambridge United States

Personal Author(s) : Blitzstein,Joseph K ; Wolfe,Patrick J

Full Text :

Report Date : 11 Nov 2017

Pagination or Media Count : 35

Abstract : Primary overarching goal: Answer the question How, and under what conditions, can we detect the presence of structure in networksstructure that is not well explained by background models? The MURI has significantly advanced our theoretical and practical understanding of how to model background network clutter, leading to principled approaches to foreground sub-network detection. Before the MURI, no frameworks existed for network detection theory or goodness-of-fit, nor were models and algorithms coupled to sound sociological principles. In particular, in standard statistical theory, confidence intervals quantify thresholds for rejecting the null hypothesis, which is the detection of a significant signal in this context. Such methods had never been used to achieve confidence values prior to this, but the MURI team obtained the first such confidence values for networks. Additionally, the MURI team proved and published the first known detection-theoretic theorem to formally test for signal presence by quantifying if the observed network structure is consistent with the fitted clutter model. In addition this work lead to 1.) First ever computationally scalable algorithms to capture social dynamics from network analysis, resulting from fast leader-follower algorithm. 2.) First ever flexible model-free approaches to signal detection in networks. 3.) New interpretations of overlapping community structure, resulting from modernized mixed membership models. 4.) New exploitations of latent social foci. In particular, we showed the mismatch of existing social network signal detection algorithms to social processes, modified to remove normality, orthogonality, created new models and simulation experiments, leading to new testbed.5. New sociologically principled algorithms (versus abstract network principled) for subnetwork detection.

Descriptors :   social networks , statistical analysis , social media , group dynamics , network science , data mining

Subject Categories : Sociology and Law
      Information Science
      Statistics and Probability
      Operations Research
      Computer Systems

Distribution Statement : APPROVED FOR PUBLIC RELEASE