Graph Fraud Detection on YelpChi (AUC, Recall, G-Mean)
92.08AUCLGSPF
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LGSPFOptimization=End-to-end2026.05 | 92.08 | 83.46 | 83.25 | |
| RGTAN2026.05 | 91.94 | 77.43 | 75.06 | |
| GTAN2026.05 | 91.72 | 56 | 73.63 | |
| w/o SemanticsDescription=LGSPF ablation removing natural language instructions2026.05 | 90.95 | 77.75 | 64.18 | |
| w/o Joint-OptDescription=LGSPF ablation without end-to-end optimization (frozen GNNs)2026.05 | 90.76 | 79.15 | 82.96 | |
| BWGNN2026.05 | 88.76 | 65.79 | 76.95 | |
| PC-GNN2026.05 | 81.84 | 73.48 | 72.9 | |
| CARE-GNN2026.05 | 77.27 | 70.46 | 70.44 | |
| w/o LLMDescription=LGSPF ablation replacing LLM backbone with MLP classifier2026.05 | 77.26 | 59.47 | 73.95 | |
| GraphSAGE2026.05 | 73.74 | 63.25 | 61.46 | |
| Instruct-LLMMode=Instruction-tuned2026.05 | 71.24 | 67.1 | 74.44 | |
| GCN2026.05 | 63.38 | 59.06 | 59.31 | |
| GAT2026.05 | 57.04 | 53.63 | 54.29 | |
| Zero-shot LLMMode=Zero-shot2026.05 | 41.45 | 1.72 | 2.84 |