Fraud detection at eBay

Fraud detection is a key research topic for e-commerce, addressing challenges like dynamic heterogeneity and interlinked fraudulent patterns. Existing efforts include rule-based and machine learning systems, but graph-based approaches are increasingly critical. This paper presents the first systematic review of fraud detection in real-world e-commerce environment like eBay, leveraging multi-source data such as transaction logs and user behavior, dealing with challenges of information heterogeneity, scalability, graph dynamics, explainability, and adaptability. We also highlight eBay's efforts in designing explainable fraud detection systems with graph neural networks (GNNs) tailored to deployment needs and offer insights and recommendations for advancing research. • Challenges: Info heterogeneity, scalability, graph dynamics, & explainability. • An in-depth review of fraud detection landscape with 180 publications. • Innovative Solutions. • xFraud deals with heterogeneity and provides human-readable explanations. • BRIGHT offers real-time GNN inference for faster fraud detection. • FinDeepBehaviorCluster leverages user behavior data for unsupervised fraud detection. • BehaviorClustering uses GNN-learned embeddings to extract user behavior embeddings.