Seasonal Stochastic Blockmodeling for Anomaly Detection in Dynamic Networks

TL;DR

This paper introduces a seasonal stochastic blockmodel for dynamic networks to identify anomalies by modeling time-varying structures driven by seasonal processes, with applications in geospatial and sociotechnological systems.

Contribution

It presents a novel statistical model that captures seasonal dependencies in dynamic networks and provides an inference method to detect anomalies based on learned normal patterns.

Findings

Model effectively captures seasonal patterns in dynamic networks.

Inference procedure successfully recovers normal seasonal processes.

Preliminary experiments indicate potential for anomaly detection.

Abstract

Sociotechnological and geospatial processes exhibit time varying structure that make insight discovery challenging. To detect abnormal moments in these processes, a definition of `normal' must be established. This paper proposes a new statistical model for such systems, modeled as dynamic networks, to address this challenge. It assumes that vertices fall into one of k types and that the probability of edge formation at a particular time depends on the types of the incident nodes and the current time. The time dependencies are driven by unique seasonal processes, which many systems exhibit (e.g., predictable spikes in geospatial or web traffic each day). The paper defines the model as a generative process and an inference procedure to recover the `normal' seasonal processes from data when they are unknown. An outline of anomaly detection experiments to be completed over Enron emails and New York City taxi trips is presented.