G-Pare: A Visual Analytic Tool for Comparative Analysis of Uncertain Graphs
There are a growing number of machine learning algorithms which operate on graphs, spawning a need for tools that help users make sense of the output of these algorithms. In this work, we examine machine learning algorithms which predict the labels for the nodes in a graph, which we refer to as node labeling algorithms. Node labeling algorithms take as input a graph and output a probability distribution over the labels of the nodes in the graph. Examples include predicting which customers will recommend products to their friends in a viral marketing campaign, predicting the topics of publications in a citation network, or predicting the political affiliations of people in a social network. We refer to the output of a node labeling algorithm as an uncertain graph, and are interested in the problem of comparing and contrasting the output of different algorithms. We have developed G-Pare, a visual analytic tool for comparing uncertain graphs. G-Pare provides several different views which allow users to obtain a global overview of the algorithms' output, as well as focused views that show subsets of nodes of interest. By providing an adaptive exploration environment, G-Pare guides the users to places in the graph where two algorithms' predictions agree and places where they disagree. This enables the user to follow cascades of misclassifications by comparing the algorithms' outcome with the ground truth. After describing the features of G-Pare, we illustrate its utility through several use cases based on networks from different domains.
Hossam Sharara, PhD Student, Computer Science
Awalin Sopan, PhD Student, Computer Science
Galileo Namata, PhD Student, Computer Science
Lise Getoor, Associate Professor, University of Maryland
Lisa Singh, Associate Professor, Georgetown University
G-PARE: A Visual Analytic Tool for Comparative Analysis of Uncertain Graphs
Hossam Sharara, Awalin Sopan, Galileo Namata, Lise Getoor, Lisa Singh
IEEE Conference on Visual Analytics Science and Technology, 2011 (VAST '11).
This work was supported in part by the FODAVA program from NSF Grant #CCF0937094.