Predicting unobserved links in incompletely observed networks

Authors:
David J. Marchette;Carey E. Priebe
Affiliations:
Naval Surface Warfare Center, Code Q21, Dahlgren, VA 22448, USA;Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD 21218, USA
Venue:
Computational Statistics & Data Analysis
Year:
2008

Citing 2
Cited 2

On Random Intersection Graphs: The Subgraph Problem

Combinatorics, Probability and Computing
Modeling graphs using dot product representations

Computational Statistics

Graph matching based on dot product representation of graphs

GbRPR'11 Proceedings of the 8th international conference on Graph-based representations in pattern recognition
Graph matching based on spectral embedding with missing value

Pattern Recognition

Quantified Score

Hi-index	0.03

Visualization

Abstract

In this paper we consider networks in which the links (edges) are imperfectly observed. This may be a result of sampling, or it may be caused by actors (vertices) who are actively attempting to hide their links (edges). Thus the network is incompletely observed, and we wish to predict which of the possible unobserved links are actually present in the network. To this end, we apply a constrained random dot product graph (CRDPG) to rank the potential edges according to the probability (under the model) that they are in fact present. This model is then extended to utilize covariates measured on the actors, to improve the link prediction. The method is illustrated on a data set of alliances between nations, in which a subset of the links (alliances) is assumed unobserved for the purposes of illustration.