融合用户行为与评论关系的双通道电商欺诈检测方法

凤丽洲,白至纯,王友卫

Dual-channel E-commerce fraud detection method integrating user behavior and review relationships

Lizhou FENG,Zhichun BAI,Youwei WANG

表 2 不同欺诈检测方法在实验用数据集上的性能对比

Tab.2 Performance comparison of different fraud detection methods on experimental datasets %

方法 Pet Garden Instrument Yelp AUC Rm F1 AUC Rm F1 AUC Rm F1 AUC Rm F1 GCN 49.89 50.00 49.12 49.70 50.00 48.45 55.11 51.43 45.88 52.35 50.00 49.10 GAT 50.48 50.32 49.53 49.04 50.00 48.75 58.29 52.77 48.75 51.70 50.21 49.29 GraghSAGE 74.71 50.00 49.89 73.19 50.13 49.45 76.22 55.43 57.68 67.43 50.00 49.29 GEM 64.36 59.16 53.94 54.39 52.11 51.77 66.72 63.25 52.79 64.87 49.89 49.09 GeniePath 53.99 50.00 50.03 55.77 51.02 48.97 53.21 50.00 49.58 51.20 50.00 49.79 FdGars 63.93 52.00 50.14 56.12 51.31 48.58 52.58 50.78 47.88 52.37 51.32 50.03 GraphConsis 79.93 74.88 57.68 74.58 71.59 54.26 77.11 72.29 55.98 69.18 52.10 49.28 PC-GNN 67.86 61.90 51.42 63.66 59.07 53.90 63.67 59.34 50.01 61.08 54.77 52.82 EC-GNN 87.14 79.08 62.05 88.07 78.52 67.03 88.89 81.09 65.21 86.73 79.03 54.26