Identifying the Fraudulent Users for E-commerce Applications Based on the Access Behaviors

The rapid growth of e-commerce has increased fraudulent activities, causing profit losses for both merchants and customers. Previous studies focus on either spatial relations among users or individual user behavior over time, lacking universality for various types of fraud. This work proposes a novel method for identifying fraudulent users on e-commerce platforms by combining temporal behavior patterns and spatial relationships. We use a Long Short-Term Memory (LSTM) model to analyze time-varying user behavior, a Multi-Layer Perceptron (MLP) to integrate multidimensional access features, and a graph neural network to capture spatial patterns. Numerical experiments on real datasets validate the proposed scheme, demonstrating superior performance in detecting two types of fraud compared to existing methods.