A Multiperspective Fraud Detection Method For Multiparticipant ECommerce Transactions

Abstract

In the rapidly evolving landscape of digital finance, multi-participant e-commerce transactions—involving buyers, sellers, payment gateways, and logistics providers—have introduced complex vulnerabilities that traditional fraud detection systems often fail to address. This paper proposes a Multiperspective Fraud Detection Method specifically designed to analyze the intricate behavioral patterns and data interdependencies inherent in multi-party environments. Unlike conventional models that focus primarily on binary buyer-seller interactions, our approach integrates a heterogeneous graph-based framework to capture the multi-dimensional relationships between all stakeholders. By leveraging Graph Neural Networks (GNNs) combined with temporal sequence analysis, the method identifies anomalous subgraphs and suspicious transaction chains that signify sophisticated collusion or account takeover attempts.The research introduces a novel weighting mechanism that balances data from various perspectives, including transaction frequency, IP geolocation consistency, and historical reputation scores across different platforms. This multiperspective lens allows for the detection of "low-and-slow" attacks that typically bypass threshold-based security measures. Experimental results, conducted on real-world large-scale e-commerce datasets, demonstrate that the proposed method significantly outperforms baseline models in terms of Precision, Recall, and F1-Score. Furthermore, the system exhibits high robustness against data sparsity and cold-start problems common in new participant accounts. By providing a holistic view of the transaction lifecycle, this method offers a scalable and proactive solution for securing global e-commerce ecosystems, reducing financial losses while maintaining a seamless user experience for legitimate participants.