GraphSAC: Detecting anomalies in large-scale graphs

TL;DR

GraphSAC introduces a robust, scalable method for detecting anomalous nodes in large graphs by using random sampling and semi-supervised learning, effectively handling adversarial attacks and attribute/link compromises.

Contribution

The paper proposes GraphSAC, a novel graph-based sampling and consensus approach that improves anomaly detection robustness and scalability in large-scale graphs.

Findings

GraphSAC outperforms state-of-the-art methods in real-world experiments.

It provides theoretical performance guarantees with bounded sample complexity.

The approach is scalable due to linear per-draw complexity and parallelizable sampling.

Abstract

A graph-based sampling and consensus (GraphSAC) approach is introduced to effectively detect anomalous nodes in large-scale graphs. Existing approaches rely on connectivity and attributes of all nodes to assign an anomaly score per node. However, nodal attributes and network links might be compromised by adversaries, rendering these holistic approaches vulnerable. Alleviating this limitation, GraphSAC randomly draws subsets of nodes, and relies on graph-aware criteria to judiciously filter out sets contaminated by anomalous nodes, before employing a semi-supervised learning (SSL) module to estimate nominal label distributions per node. These learned nominal distributions are minimally affected by the anomalous nodes, and hence can be directly adopted for anomaly detection. Rigorous analysis provides performance guarantees for GraphSAC, by bounding the required number of draws. The per-draw complexity grows linearly with the number of edges, which implies efficient SSL, while draws can be run in parallel, thereby ensuring scalability to large graphs. GraphSAC is tested under different anomaly generation models based on random walks, clustered anomalies, as well as contemporary adversarial attacks for graph data. Experiments with real-world graphs showcase the advantage of GraphSAC relative to state-of-the-art alternatives.