Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters

Abstract

Graph Neural Networks (GNNs) have been widely applied to fraud detection problems in recent years, revealing the suspiciousness of nodes by aggregating their neighborhood information via different relations. However, few prior works have noticed the camouflage behavior of fraudsters, which could hamper the performance of GNNbased fraud detectors during the aggregation process. In this paper, we introduce two types of camouflages based on recent empirical studies, i.e., the feature camouflage and the relation camouflage. Existing GNNs have not addressed these two camouflages, which results in their poor performance in fraud detection problems. Alternatively, we propose a new model named CAmouflage-REsistant GNN (CARE-GNN), to enhance the GNN aggregation process with three unique modules against camouflages. Concretely, we first devise a label-aware similarity measure to find informative neighboring nodes. Then, we leverage reinforcement learning (RL) to find the optimal amounts of neighbors to be selected. Finally, the selected neighbors across different relations are aggregated together. Comprehensive experiments on two real-world fraud datasets demonstrate the effectiveness of the RL algorithm. The proposed CARE-GNN also outperforms state-of-the-art GNNs and GNN-based fraud detectors. We integrate all GNN-based fraud detectors as an opensource toolbox 1. The CARE-GNN code and datasets are available at https://github.com/YingtongDou/CARE-GNN. CCS CONCEPTS • Security and privacy → Web application security; • Computing methodologies → Neural networks.

Fraud Detection in Online Product Review Systems via Heterogeneous Graph Transformer

“…GraphConsis [4] suffers from inflexible filtering thresholds and unsupervised similarity measures. CAREGNN [5] cannot handle multiple node types and…”

“…For general graph neural networks, we select GCN [15], GAT [17], GraphSAGE [16], GeniePath [16]. For popular GNN-based fraud detectors, we select SemiGNN [8], GraphConsis [4] and CAREGNN [5]. GCN, GAT, GraphSAGE, and GeniePath are run on homogeneous graphs.…”

“…The current approach is to model the data as a heterogeneous information network so that similarities in characteristics and structure of fraudsters can be discovered. Due to the effectiveness in learning the graph representation, graph neural networks (GNNs) have already been introduced into fraud detection areas including product review [2]- [5], mobile application distribution [6], cyber crime identification [7] and financial services [8], [9]. How-ever, most existing GNN based solutions just directly apply homogeneous GNNs, ignoring the underlying heterogeneous graph nature and camouflage node behaviors.…”

“…How-ever, most existing GNN based solutions just directly apply homogeneous GNNs, ignoring the underlying heterogeneous graph nature and camouflage node behaviors. This problem has drawn great attention with many solutions proposed [4], [5], [10]. GraphConsis [4] found that there are three inconsistency problems in fraud detection and CAREGNN [5] further proposed two camouflage behaviors.…”

Financial Crime & Fraud Detection Using Graph Computing: Application Considerations & Outlook

“…[53] explored dynamic considerations in graph networks, which is essential in fraud applications. [54] proposed solutions to expose camouflaged fraudsters through GNNs. Later studies aimed to improve the structural limitations of the graph networks through attention mechanisms for higher efficiency [55].…”