Table 1.
Summary of the advantages and disadvantages of the relevant work on Internet fraud transaction detection.
Table 2.
Term definition.
Table 3.
Symbol table.
Fig 1.
The overall framework of the THG-OAFN model.
Algorithm 1.
THG-OAFN: Time-aware heterogeneous graph oversampling and active fraud detection algorithm.
Fig 2.
The integration of GRU and GNN.
Algorithm 2.
The GRU-GNN integration and fusion process.
Fig 3.
The structure of the heterogeneous graph-based oversampling model.
Fig 4.
The technical process of SMOTE.
Algorithm 3.
Pseudocode based on a heterogeneous graph oversampling model.
Fig 5.
The structure of the heterogeneous graph fraud detection framework based on multilayer attention.
Algorithm 4.
The pseudocode for multi-layer attention fusion of relationship-neighborhood-information.
Table 4.
The summary of the statistical characteristics of datasets.
Table 5.
Model performance comparison after hyperparameter adjustment (Amazon dataset, 40% training ratio).
Table 6.
The hyperparameter grid search results of the THG-OAFN model (Amazon dataset).
Fig 6.
The influence of oversampling degree on model performance.
Table 7.
The regularization control effects at an oversampling level of 1.2 (Amazon dataset).
Table 8.
Ablation experiment results of the attention heads (AUC/%).
Table 9.
Module-level ablation experiment results.
Fig 7.
Influence of different node embedding dimensions on model performance ((a) AUC; (b) Recall; (c) F1).
Fig 8.
Effects of various training ratios on model performance ((a) AUC; (b) Recall; (c) F1).
Table 10.
Performance comparison between THG-OAFN and the baseline models on the Amazon dataset (70% training ratio).
Fig 9.
Performance of different models on various datasets ((a) Amazon; (b) YelpChi; (c) Credit card fraud detection dataset).
Fig 10.
Performance of diverse datasets on different models ((a) AUC; (b) Recall; (c) F1).
Table 11.
Performance comparison between THG-OAFN and the specialized fraud detection model.
Fig 11.
Node embedding PCA dimension reduction coordinates (sampling 1000 nodes in the Amazon dataset).
Fig 12.
Multi-layer attention weight distribution.
Table 12.
Contribution of gradient backpropagation features (TOP5 features of the credit card fraud detection dataset).
Fig 13.
The average cosine similarity of the proposed method under different nodes ((a) Amazon; (b) YelpChi; (c) Credit card fraud detection dataset).
Table 13.
Comparison of execution time (in seconds).
Table 14.
Cross-domain generalization experiment: Performance comparison between THG-OAFN and baseline models on financial datasets.
Table 15.
Performance comparison of insurance claim fraud detection (F1-score/%).