Abstract

A wide variety of application domains is concerned with data consisting of entities and their relationships or connections, formally represented as graphs. Within these diverse application areas, a common problem of interest is the detection of a subset of entities whose connectivity is anomalous with respect to the rest of the data. While the detection of such anomalous subgraphs has received a substantial amount of attention, no application-agnostic framework exists for analysis of signal detectability in graph-based data. In this paper, we describe a framework that enables such analysis using the principal eigenspace of a graph's residuals matrix, commonly called the modularity matrix in community detection. Leveraging this analytical tool, we show that the framework has a natural power metric in the spectral norm of the anomalous subgraph's adjacency matrix (signal power) and of the background graph's residuals matrix (noise power). We propose several algorithms based on spectral properties of the residuals matrix, with more computationally expensive techniques providing greater detection power. Detection and identification performance are presented for a number of signal and noise models, including clusters and bipartite foregrounds embedded into simple random backgrounds, as well as graphs with community structure and realistic degree distributions. The trends observed verify intuition gleaned from other signal processing areas, such as greater detection power when the signal is embedded within a less active portion of the background. We demonstrate the utility of the proposed techniques in detecting small, highly anomalous subgraphs in real graphs derived from Internet traffic and product co-purchases.

Original languageEnglish (US)
Article number7112641
Pages (from-to)4191-4206
Number of pages16
JournalIEEE Transactions on Signal Processing
Volume63
Issue number16
DOIs
StatePublished - Aug 15 2015

Fingerprint

Signal processing
Internet

Keywords

  • Graph theory
  • principal components analysis
  • residuals analysis
  • signal detection theory
  • spectral analysis

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Signal Processing

Cite this

Miller, B. A., Beard, M. S., Wolfe, P. J., & Bliss, N. (2015). A Spectral Framework for Anomalous Subgraph Detection. IEEE Transactions on Signal Processing, 63(16), 4191-4206. [7112641]. https://doi.org/10.1109/TSP.2015.2437841

A Spectral Framework for Anomalous Subgraph Detection. / Miller, Benjamin A.; Beard, Michelle S.; Wolfe, Patrick J.; Bliss, Nadya.

In: IEEE Transactions on Signal Processing, Vol. 63, No. 16, 7112641, 15.08.2015, p. 4191-4206.

Research output: Contribution to journalArticle

Miller, BA, Beard, MS, Wolfe, PJ & Bliss, N 2015, 'A Spectral Framework for Anomalous Subgraph Detection', IEEE Transactions on Signal Processing, vol. 63, no. 16, 7112641, pp. 4191-4206. https://doi.org/10.1109/TSP.2015.2437841
Miller, Benjamin A. ; Beard, Michelle S. ; Wolfe, Patrick J. ; Bliss, Nadya. / A Spectral Framework for Anomalous Subgraph Detection. In: IEEE Transactions on Signal Processing. 2015 ; Vol. 63, No. 16. pp. 4191-4206.
@article{a3d9ae756b9b4a488751a3a160ce2405,
title = "A Spectral Framework for Anomalous Subgraph Detection",
abstract = "A wide variety of application domains is concerned with data consisting of entities and their relationships or connections, formally represented as graphs. Within these diverse application areas, a common problem of interest is the detection of a subset of entities whose connectivity is anomalous with respect to the rest of the data. While the detection of such anomalous subgraphs has received a substantial amount of attention, no application-agnostic framework exists for analysis of signal detectability in graph-based data. In this paper, we describe a framework that enables such analysis using the principal eigenspace of a graph's residuals matrix, commonly called the modularity matrix in community detection. Leveraging this analytical tool, we show that the framework has a natural power metric in the spectral norm of the anomalous subgraph's adjacency matrix (signal power) and of the background graph's residuals matrix (noise power). We propose several algorithms based on spectral properties of the residuals matrix, with more computationally expensive techniques providing greater detection power. Detection and identification performance are presented for a number of signal and noise models, including clusters and bipartite foregrounds embedded into simple random backgrounds, as well as graphs with community structure and realistic degree distributions. The trends observed verify intuition gleaned from other signal processing areas, such as greater detection power when the signal is embedded within a less active portion of the background. We demonstrate the utility of the proposed techniques in detecting small, highly anomalous subgraphs in real graphs derived from Internet traffic and product co-purchases.",
keywords = "Graph theory, principal components analysis, residuals analysis, signal detection theory, spectral analysis",
author = "Miller, {Benjamin A.} and Beard, {Michelle S.} and Wolfe, {Patrick J.} and Nadya Bliss",
year = "2015",
month = "8",
day = "15",
doi = "10.1109/TSP.2015.2437841",
language = "English (US)",
volume = "63",
pages = "4191--4206",
journal = "IEEE Transactions on Signal Processing",
issn = "1053-587X",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "16",

}

TY - JOUR

T1 - A Spectral Framework for Anomalous Subgraph Detection

AU - Miller, Benjamin A.

AU - Beard, Michelle S.

AU - Wolfe, Patrick J.

AU - Bliss, Nadya

PY - 2015/8/15

Y1 - 2015/8/15

N2 - A wide variety of application domains is concerned with data consisting of entities and their relationships or connections, formally represented as graphs. Within these diverse application areas, a common problem of interest is the detection of a subset of entities whose connectivity is anomalous with respect to the rest of the data. While the detection of such anomalous subgraphs has received a substantial amount of attention, no application-agnostic framework exists for analysis of signal detectability in graph-based data. In this paper, we describe a framework that enables such analysis using the principal eigenspace of a graph's residuals matrix, commonly called the modularity matrix in community detection. Leveraging this analytical tool, we show that the framework has a natural power metric in the spectral norm of the anomalous subgraph's adjacency matrix (signal power) and of the background graph's residuals matrix (noise power). We propose several algorithms based on spectral properties of the residuals matrix, with more computationally expensive techniques providing greater detection power. Detection and identification performance are presented for a number of signal and noise models, including clusters and bipartite foregrounds embedded into simple random backgrounds, as well as graphs with community structure and realistic degree distributions. The trends observed verify intuition gleaned from other signal processing areas, such as greater detection power when the signal is embedded within a less active portion of the background. We demonstrate the utility of the proposed techniques in detecting small, highly anomalous subgraphs in real graphs derived from Internet traffic and product co-purchases.

AB - A wide variety of application domains is concerned with data consisting of entities and their relationships or connections, formally represented as graphs. Within these diverse application areas, a common problem of interest is the detection of a subset of entities whose connectivity is anomalous with respect to the rest of the data. While the detection of such anomalous subgraphs has received a substantial amount of attention, no application-agnostic framework exists for analysis of signal detectability in graph-based data. In this paper, we describe a framework that enables such analysis using the principal eigenspace of a graph's residuals matrix, commonly called the modularity matrix in community detection. Leveraging this analytical tool, we show that the framework has a natural power metric in the spectral norm of the anomalous subgraph's adjacency matrix (signal power) and of the background graph's residuals matrix (noise power). We propose several algorithms based on spectral properties of the residuals matrix, with more computationally expensive techniques providing greater detection power. Detection and identification performance are presented for a number of signal and noise models, including clusters and bipartite foregrounds embedded into simple random backgrounds, as well as graphs with community structure and realistic degree distributions. The trends observed verify intuition gleaned from other signal processing areas, such as greater detection power when the signal is embedded within a less active portion of the background. We demonstrate the utility of the proposed techniques in detecting small, highly anomalous subgraphs in real graphs derived from Internet traffic and product co-purchases.

KW - Graph theory

KW - principal components analysis

KW - residuals analysis

KW - signal detection theory

KW - spectral analysis

UR - http://www.scopus.com/inward/record.url?scp=84936971207&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84936971207&partnerID=8YFLogxK

U2 - 10.1109/TSP.2015.2437841

DO - 10.1109/TSP.2015.2437841

M3 - Article

VL - 63

SP - 4191

EP - 4206

JO - IEEE Transactions on Signal Processing

JF - IEEE Transactions on Signal Processing

SN - 1053-587X

IS - 16

M1 - 7112641

ER -