GNN-EADD: Graph Neural Network-Based E-Commerce Anomaly Detection via Dual-Stage Learning

Abstract

E-commerce platforms face significant challenges in detecting anomalous products, including counterfeit goods and fraudulent listings, which can undermine user trust and platform integrity. This paper presents Graph Neural Network-based E-commerce Anomaly Detection via Dual-stage Learning (GNN-EADD), a novel approach leveraging graph neural networks for anomaly detection in large-scale e-commerce ecosystems. Our key contributions include: 1) A heterogeneous graph representation incorporating products, sellers, and buyers as nodes with their relationships as edges; 2) A novel dual-stage learning framework combining unsupervised graph embedding with semi-supervised fine-tuning; and 3) An attention mechanism that effectively captures complex patterns within network structures. Extensive experiments on a large-scale Amazon dataset demonstrate that GNN-EADD significantly outperforms state-of-the-art baselines in terms of anomaly detection accuracy, precision, and recall, while showing robustness to various types of anomalies and scalability to large networks.

Keywords

• Graph neural networks
• e-commerce
• anomaly detection
• heterogeneous graphs
• graph attention networks