Research map / Graph learning / GNN architectures
Targeting Borderline Fraudsters: Multi-View Hypergraph Fraud Detection with LLM-Guided Contrastive Learning
Rui Ou, Kun Zhu, Nana Zhang, Jiangtong Li, Chaochao Chen, Yuhua Xu, Changjun Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add temporal hyper-view HG(1) to THAN (ablation) | MH-LGC · Credit | AUC: 92.97 -> 94.86 | Improved |
| Add frequency hyper-view HG(2) (ablation) | MH-LGC · Credit | AUC: 93.18 -> 94.86 | Improved |
| Add both hyper-views (THAN) (ablation) | MH-LGC · Wikipedia | AUC: 87.67 -> 90.17 | Improved |
| Add LLM-guided contrastive learning (LGC) (ablation) | MH-LGC · Credit | AUC: 90.54 -> 94.86 | Improved |
| Add LGC (ablation) | MH-LGC · Reddit | AUC: 68.94 -> 69.87 | Improved |
| Add LGC (ablation) | MH-LGC · Reddit | Recall: 63.50 -> 65.01 | No clear effect |
| Use LLaMA3-8B instead of LLaMA2-7B in LGC (ablation) | MH-LGC · Credit | AUC: 93.84 -> 94.86 | Improved |
| Use LLaMA3-8B instead of Vicuna-7B in LGC (ablation) | MH-LGC · Reddit | Recall: 64.76 -> 65.01 | No clear effect |
| Use LLM only for borderline fraudsters (4 LLM queries per batch) | MH-LGC vs GAugLLM · training efficiency | time per batch: 64m -> 21s | Improved |
| Add multi-view hypergraph and LGC to TGAT backbone | MH-LGC · Credit | AUC: 86.84 -> 94.86 | Improved |
| Use general LLM-integrated method OFA instead of TGAT | OFA · Wikipedia | AUC: 83.69 -> 82.84 | Hurt |
Closest papers by meaning
- Dynamic Neighborhood Modeling via Node-Subgraph Contrastive Learning for Graph-Based Fraud Detection (AAAI 2025 · similarity 0.90)
- DiG-In-GNN: Discriminative Feature Guided GNN-Based Fraud Detector against Inconsistencies in Multi-Relation Fraud Graph (AAAI 2024 · similarity 0.90)
- Context-aware Graph Neural Network for Graph-based Fraud Detection with Extremely Limited Labels (AAAI 2025 · similarity 0.87)
- Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum (AAAI 2024 · similarity 0.87)
- Provably Powerful Graph Neural Networks for Directed Multigraphs (AAAI 2024 · similarity 0.86)
- Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks Against GNN-Based Fraud Detectors (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
