Intelligent Information Networks: Young Investigator Program: Modeling, Sampling, and Analyzing Adversarial Social Networks
Abstract
Over the past decade, complex network analysis has become a central part of a wide range of scientific fields. However, the quality of network analysis depends on the quality of the data, and in certain settings, such as when dealing with a ÔdarkÕ network, individuals in the network may act adversarially when providing data. The purpose of this project is to develop models and techniques for the entire pipeline of network analysis under such scenarios, including modeling, sampling, and analyzing inaccurately or adversarially-provided network data. The research literature contains numerous algorithms for collecting and analyzing sampled networks; however, these methods typically make strong, rigid assumptions about the nature of the sampling process (for example, they may assume that a query on a node returns all edges adjacent to that node, or a uniformly random subset of these edges). These assumptions are realistic when collecting online data through an API or some other automated process, but are less likely to hold when dealing with other applications. The proposed project contains three tasks: (1) Develop models to describe the structure of network data obtained under such conditions, (2) Create sampling algorithms, including interactive methods, for collecting network data in these settings, and (3) Develop a suite of graph analysis algorithms to analyze the resulting network samples, taking into account how the data was collected and the likely errors in the observed network. Task 1: Modeling adversarial social networks. In this task, the PI will characterize how individuals report network data when those individuals are acting in an adversarial manner against those who are collecting the data. The PI will address this task from the perspective of network games, and will perform both a game theoretic analysis as well as an experimental analysis to explore how individuals report their data, and how this is affected by different network configurations and incentives. Task 2: Sampling adversarial social networks. The work in this task will use the models developed in Task 1 to develop techniques for exploring network data. Unlike traditional sampling methods, the methods developed in this task may suggest that the user repeatedly query the same node or region of the graph in order to reinforce portions of the observed graph that may be inaccurate. Task 3: Analyzing adversarial social networks. Network analysis algorithms rarely distinguish between input graphs that are complete and those that are samples; however, by taking into account how the data was collected and any known incompleteness, one can substantially improve the quality of the analysis. The proposed work will focus on the samples generated under inaccurate or adversarial conditions, which poses addtional challenges. The work in this taske will focus on community detection and identification of influential nodes in the network.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1810047
Entities
People
- Sucheta Soundarajan
Organizations
- Army Contracting Command
- Syracuse University
- United States Army