Fraud detection and risk assessment of online payment transactions on e-commerce platforms based on LLM and GCN frameworks (2509.09928v1)

Abstract: With the rapid growth of e-commerce, online payment fraud has become increasingly complex, posing serious threats to financial security and consumer trust. Traditional detection methods often struggle to capture the intricate relational structures inherent in transactional data. This study presents a novel fraud detection framework that combines LLMs (LLM) with Graph Convolutional Networks (GCN) to effectively identify fraudulent activities in e-commerce online payment transactions. A dataset of 2,840,000 transactions was collected over 14 days from major platforms such as Amazon, involving approximately 2,000 U.S.-based consumers and 30 merchants. With fewer than 6000 fraudulent instances, the dataset represents a highly imbalanced scenario. Consumers and merchants were modeled as nodes and transactions as edges to form a heterogeneous graph, upon which a GCN was applied to learn complex behavioral patterns. Semantic features extracted via GPT-4o and Tabformer were integrated with structural features to enhance detection performance. Experimental results demonstrate that the proposed model achieves an accuracy of 0.98, effectively balancing precision and sensitivity in fraud detection. This framework offers a scalable and real-time solution for securing online payment environments and provides a promising direction for applying graph-based deep learning in financial fraud prevention.

• The paper introduces an innovative fusion of LLMs and GCNs to detect fraudulent patterns in imbalanced e-commerce transactions.
• The framework utilizes heterogeneous graph construction and dual-feature encoding via GPT-4o and Tabformer to capture both semantic and structural insights.
• Experimental results demonstrate high accuracy and improved risk assessment, although recall remains challenging due to class imbalance.

Fraud Detection and Risk Assessment of Online Payment Transactions on E-commerce Platforms Using LLM and GCN Frameworks

Introduction

The paper "Fraud detection and risk assessment of online payment transactions on e-commerce platforms based on LLM and GCN frameworks" presents an innovative approach to tackling complex online payment fraud issues. By integrating LLMs and Graph Convolutional Networks (GCNs), the study offers a method for effectively detecting fraudulent activities within e-commerce transactions. Given the dataset's inherent imbalance, with fewer than 6000 fraud cases among 2.84 million transactions, this approach provides a robust framework for addressing the intricacies of fraud detection in financial systems.

Methodology

Graph Construction and Model Definition

The methodology utilizes a heterogenous graph representation where transactions between the nodes (consumers and merchants) form the edges. Each transaction includes attributes such as amount and timestamp. The novelty lies in the integration of GCNs, which learn from neighbors in the graph, capturing both direct and indirect patterns indicative of fraud. A two-layer GCN model aggregates information to enhance local and global feature representations. The weighted loss function is adopted to counteract data imbalance, with an emphasis on accurately identifying the minority class of fraudulent transactions.

Feature Representation and Semantic Integration

Feature extraction relies on both GPT-4o and Tabformer to handle textual and structured data. GPT-4o aids in extracting semantic embeddings from unstructured transaction fields, while Tabformer encodes structured fields, preserving data dependencies. This dual-feature encoding yields comprehensive node and edge representations, thus facilitating nuanced pattern recognition when these features are fused.

Results

Experimental results indicate that the framework achieves high accuracy (0.98) and demonstrates strong performance in fraud detection. However, while the precision for identifying fraud is optimal, recall remains low due to the significant class imbalance. The model's use of a class-weighted loss function during GCN training mitigates this issue to some extent, as highlighted by the detailed analysis which confirms high sensitivity to legitimate transactions.

Implications and Future Work

The integration of LLMs for semantic understanding with the robust structural learning of GCNs opens new possibilities for fraud detection. It demonstrates improvements over traditional methods, particularly in handling complexity and data imbalance. Future work could focus on enhancing recall through dynamic graph modeling and incorporating additional data modalities to further improve model robustness. Additionally, efforts to reduce false positives would be beneficial, increasing the practicality of the framework for real-world deployment.

Conclusion

This study demonstrates a promising fusion of LLMs and GCNs in developing advanced fraud detection systems. The framework enhances e-commerce security by providing a scalable, real-time solution for identifying sophisticated fraud patterns. As online transactions continue to grow, such approaches will be integral in safeguarding financial security, thereby maintaining consumer trust. The paper offers valuable insights into combating fraud through interdisciplinary techniques that leverage graph-based deep learning and language processing.