New Graph Neural Network Boosts Few-Shot Fraud Detection

Liming Liu, Chao Hu, Mingfei Lu, Yiwei Ge, Xingle Li, Heyuan Shi· June 29, 2026 View original

Summary

Researchers introduce ADC-GNN, a novel framework combining diffusion-guided feature augmentation, contrastive learning, and multi-hop spectral attention to improve fraud detection on graphs with sparse and imbalanced labels. The model effectively addresses representation dilution and oversmoothing challenges in real-world transaction systems.

A new research paper presents ADC-GNN, an innovative Attention-guided Diffusion-Contrastive Graph Neural Network designed to enhance fraud detection, particularly in scenarios with limited and imbalanced labeled data. This framework tackles two significant challenges in real-world fraud graphs: the scarcity of verified fraudulent labels and the issue of representation dilution, where traditional graph neural networks can either oversmooth anomalies or suppress crucial high-frequency fraud signals. ADC-GNN integrates a diffusion component for feature-space denoising augmentation, which uses contrastive learning to stabilize node representations. Additionally, a spectral attention module adaptively highlights relevant information across different hop levels and relationships within the graph. The model demonstrated consistent improvements over existing baselines on public benchmarks and a proprietary telecom dataset, especially in few-shot learning settings.

Why it matters

For professionals in finance, cybersecurity, and telecommunications, this research offers a promising new approach to detect fraud more effectively, even when historical fraud data is scarce, which is a common and critical challenge.

How to implement this in your domain

  1. 1Explore integrating ADC-GNN's principles into existing graph-based fraud detection systems.
  2. 2Pilot the diffusion-guided feature augmentation technique to improve representation learning with limited data.
  3. 3Evaluate the effectiveness of multi-hop spectral attention for identifying subtle fraud patterns in graph data.
  4. 4Benchmark ADC-GNN against current fraud detection models using proprietary datasets to assess performance gains.

Who benefits

BFSITelecommunicationsCybersecurityE-commerce

Key takeaways

  • ADC-GNN improves few-shot graph fraud detection by addressing sparse supervision and representation dilution.
  • It uses diffusion-guided feature augmentation and contrastive learning for robust node representations.
  • Multi-hop spectral attention helps identify fraud-relevant cues across graph structures.
  • The model shows significant performance gains over baselines, especially with limited training data.

Original post by Liming Liu, Chao Hu, Mingfei Lu, Yiwei Ge, Xingle Li, Heyuan Shi

"arXiv:2606.28134v1 Announce Type: new Abstract: Graph-based fraud detection is essential for safeguarding large-scale transaction systems, where undetected anomalies may lead to substantial financial losses and security risks. Real-world fraud graphs pose two coupled challenges:…"

View on X

Originally posted by Liming Liu, Chao Hu, Mingfei Lu, Yiwei Ge, Xingle Li, Heyuan Shi on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses