Bayesian anomaly detection methods for social networks

TL;DR

This paper introduces a two-stage Bayesian approach for detecting anomalies in dynamic social network graphs, combining simple probabilistic models with network inference to efficiently identify unusual behavior.

Contribution

The paper proposes a novel two-stage Bayesian method that efficiently detects anomalies in large, evolving social networks by combining counting process models with network inference tools.

Findings

Effective detection of anomalies demonstrated on simulated data.

Method successfully applied to real social network data.

Reduces computational complexity by focusing on potentially anomalous nodes.

Abstract

Learning the network structure of a large graph is computationally demanding, and dynamically monitoring the network over time for any changes in structure threatens to be more challenging still. This paper presents a two-stage method for anomaly detection in dynamic graphs: the first stage uses simple, conjugate Bayesian models for discrete time counting processes to track the pairwise links of all nodes in the graph to assess normality of behavior; the second stage applies standard network inference tools on a greatly reduced subset of potentially anomalous nodes. The utility of the method is demonstrated on simulated and real data sets.