Network Analytics Enabled by Generating a Pool of Network Variants from Noisy Data.

Area of Science:

• Network Science
• Social Network Analysis
• Information Security

Background:

• Understanding network structure and function is vital, especially for covert social networks that intentionally obscure their topology.
• Inferring network attributes from incomplete or noisy data presents significant challenges.

Purpose of the Study:

• To develop a robust methodology for inferring the structure and functions of covert social networks using noisy data.
• To address the limitations of single network representations when ground truth is unavailable.

Main Methods:

• Application of a generative model to create a pool of statistically equivalent networks from noisy data.
• Utilizing network variant repetition counts to approximate real-world probabilities.
• Employing Shannon entropy to identify network variants with minimal uncertainty.

Main Results:

• Identified persistent structural patterns within frequently occurring network variants.
• Demonstrated that repeated generation from the best network variant reduces uncertainty (entropy).
• Proposed a heuristic for constructing optimal network variants with minimized operational costs.

Conclusions:

• The generative model effectively infers covert network structures and functions despite noisy data.
• Shannon entropy is a valuable metric for quantifying and reducing uncertainty in network analysis.
• The methodology offers a practical approach for network operators to identify critical nodes for monitoring.