Graph Fraud Detection on S-FFSD
0.8719AUCLGSPF
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LGSPFOptimization=End-to-end2026.05 | 0.8719 | 0.7561 | 0.7837 | |
| w/o Joint-OptDescription=LGSPF ablation without end-to-end optimization (frozen GNNs)2026.05 | 0.841 | 0.7476 | 0.7588 | |
| w/o SemanticsDescription=LGSPF ablation removing natural language instructions2026.05 | 0.8364 | 0.7432 | 0.7801 | |
| GTAN2026.05 | 0.8299 | 0.6921 | 0.6196 | |
| RGTAN2026.05 | 0.8246 | 0.694 | 0.6279 | |
| w/o LLMDescription=LGSPF ablation replacing LLM backbone with MLP classifier2026.05 | 0.7638 | 0.4912 | 0.6877 | |
| GCN2026.05 | 0.7404 | 0.5777 | 0.6824 | |
| Instruct-LLMMode=Instruction-tuned2026.05 | 0.7188 | 0.4408 | 0.6412 | |
| CARE-GNN2026.05 | 0.7112 | 0.487 | 0.623 | |
| GAT2026.05 | 0.7051 | 0.5611 | 0.6583 | |
| PC-GNN2026.05 | 0.7033 | 0.5705 | 0.6213 | |
| BWGNN2026.05 | 0.7029 | 0.6201 | 0.6199 | |
| GraphSAGE2026.05 | 0.6713 | 0.3814 | 0.6055 | |
| Zero-shot LLMMode=Zero-shot2026.05 | 0.4633 | 0.0304 | 0.1729 |