Year: 2026
Type: article
Abstract: Financial and e-commerce fraud detection remains challenging because fraudulent entities may exploit structural camouflage and coordinated activities. We propose JK-DHGNN, a dual-view framework that combines observed pairwise relations with learned higher-order affiliations. A Top-$K$ dynamic hypergraph generator suppresses low-confidence assignments, a Jacobi spectral filter aggregates multi-order graph signals, and a Jacobi polynomial-based KAN-like classifier provides adaptive nonlinear prediction. Under stratified random splits, JK-DHGNN achieves the highest mean AUC and AP on YelpChi and Amazon, while remaining above several conventional and hypergraph baselines on Elliptic. Ablation analyses reveal dataset-dependent component behavior: Jacobi filtering and auxiliary supervision are most beneficial on the review graphs, whereas dynamic hypergraph learning is critical on the sparse Elliptic transaction graph. Random edge perturbation and a transductive time-aware case study further characterize the model's stability and limitations. These results support dynamic high-order structure learning for fraud detection while highlighting the need for dataset-adaptive integration of spectral and classifier modules.
Source: Scientific Reports
Authors: Danyang Li, Jie Shen, XiangBeng Yang, Zhenkai Qin
Institutions: Beijing Police College, Guangxi Police College
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7212239915