Characterizing Attitudinal Network Graphs through Frustration Cloud
Motivation
Social Networks model community dynamics. How to describe and identify inequity and bias in that community?
Approach
Propose an innovative, scalable graph-processing algorithm that scales social balance theory to automatically quantify disbalance and to predict outcomes in social and content networks.
Faculty
Dr. Jelena Tesic
Dr. Lucas Rusnak
Students
Maria Tomasso, Ph. D. Student
Eric Hull, Research Assistant  
Ben Bond, Research Assistant
Alumini
Joshua Mitchell, M.Sc, Ryan Zamora, an Constance Angeley.
Summer Programs
Rachel Liang, HCMS
Allen Wu, HCMS
Abstract
Attitudinal Network Graphs are network graphs where edges capture an expressed opinion, specifically, if two vertices connected by that edge are agreeable or antagonistic.  Attitudinal Network Graphs can be characterized as a whole by measuring the level of consensus of entire graph, and through status and influence of each vertex in the graph. In this paper we characterize vertices and edges with respect to entire graph, and propose metrics to quantify vertex influence on consensus and strength of its expressed attitudes. We propose to expand the notion of frustration index of a signed graph to frustration cloud, a collection of nearest balanced states for a given network. The frustration cloud resolves the consensus problem with minimal sentiment disruption, taking all possible consensus views over the entire network into consideration. A frustration cloud based approach removes the brittleness of traditional network graph analysis, as it allows one to examine the consensus on entire graph.  A spanning-tree-based balancing algorithm captures the variations of balanced states and global consensus of the network, and enables us to measure vertex status and attitude.  The proposed approach provides a parsimonious account of the differences between strong and weak statuses and influences of a vertex in a large network, as demonstrated on sample Highland tribe and Wikipedia administrator election data.  The method accurately models the Highland tribe network in terms of true alliances, successfully predicts administrator election outcome consistent with Wikipedia's real election outcomes, and provides deeper analytic insights into attitudinal network graph outcome, such as most influential vertices and outliers.
Link to poster
2020_graphB_poster.pdf
Social balance theory defines balance or imbalance of sentiment relation in relations with social network theory. Sentiments can result in the emergence of two groups, where discrepancy can exist between two groups. Social and content networks model the community dynamics. We ask the question: "How can we describe and identify inequity and bias in a community?" We answer with an innovative scalable graph processing algorithm that scales social balance theory to automatically quantify disbalance and to predict outcomes in social and content networks.
Dr. Jelena Tesic
Assistant Professor, Computer Science
Texas State University
Dr. Lucas Rusnak
Assistant Professor, Mathematics
Texas State University

Applications

Analysis of Wikipedia Elections

Social network constructed from Wikipedia dataset of over 7000 users participating in administrator election, with the known outcomes of the elections (elected or not elected). Status graph shows strong correlation with election outcome. Outcome anomalies are defined relative to the promotional outcomes to identify questionable and lacking promotions.

Analysis of Healthcare Data

We analyzed an anonymous health dataset of ≈ 12,000 doctor/patient pairs and ≈ 55 diagnoses. This method establishes a standard of care and diagnostic hierarchy from the Diagnostics Network with one thousand simulated consensus scenarios.

Analysis Software

We provide a user friendly method to intepret results with our interactive data analytics program. You can search by individual user, select a graph traversal method, select the number of trees used to generate the metric, and even click on individual users on the graph.

Give us your data!
Now more than ever, we are looking to test our algorithm on various real world data.
Donate Data