Deep Fraud Detection on Non-attributed Graph

TL;DR

This paper introduces a novel framework for fraud detection on non-attributed graphs that leverages structural information and contrastive pre-training to overcome the scarcity of labeled data and limited node features.

Contribution

It proposes a graph transformation method for structural information capture and a contrastive pre-training strategy to improve fraud detection on non-attributed graphs.

Findings

Effective detection performance on large-scale industrial data

Structural graph transformation enhances GNN capabilities

Contrastive pre-training leverages unlabeled data efficiently

Abstract

Fraud detection problems are usually formulated as a machine learning problem on a graph. Recently, Graph Neural Networks (GNNs) have shown solid performance on fraud detection. The successes of most previous methods heavily rely on rich node features and high-fidelity labels. However, labeled data is scarce in large-scale industrial problems, especially for fraud detection where new patterns emerge from time to time. Meanwhile, node features are also limited due to privacy and other constraints. In this paper, two improvements are proposed: 1) We design a graph transformation method capturing the structural information to facilitate GNNs on non-attributed fraud graphs. 2) We propose a novel graph pre-training strategy to leverage more unlabeled data via contrastive learning. Experiments on a large-scale industrial dataset demonstrate the effectiveness of the proposed framework for fraud detection.