Dynamic Relation-Attentive Graph Neural Networks for Fraud Detection

TL;DR

DRAG tackles fraud detection on multi-relational graphs by learning a separate node representation for each relation and a self-transformation, all under a dynamic attention framework. It introduces relation-wise and layer-wise adaptive attention to fuse per-relation representations and to combine information across multiple layers, thereby addressing heterophily and revealing both local and global structure. Empirical results on YelpChi and Amazon show that DRAG consistently outperforms or matches state-of-the-art baselines, with ablations confirming the importance of both relation-aware aggregation and multi-layer integration. The work contributes a scalable, adaptable mechanism for dynamic attention in multi-relational GNNs, with practical impact for fraud detection in real-world graphs.

Abstract

Fraud detection aims to discover fraudsters deceiving other users by, for example, leaving fake reviews or making abnormal transactions. Graph-based fraud detection methods consider this task as a classification problem with two classes: frauds or normal. We address this problem using Graph Neural Networks (GNNs) by proposing a dynamic relation-attentive aggregation mechanism. Based on the observation that many real-world graphs include different types of relations, we propose to learn a node representation per relation and aggregate the node representations using a learnable attention function that assigns a different attention coefficient to each relation. Furthermore, we combine the node representations from different layers to consider both the local and global structures of a target node, which is beneficial to improving the performance of fraud detection on graphs with heterophily. By employing dynamic graph attention in all the aggregation processes, our method adaptively computes the attention coefficients for each node. Experimental results show that our method, DRAG, outperforms state-of-the-art fraud detection methods on real-world benchmark datasets.

Figures (3)

• Figure 1: Overview of DRAG. A node representation is computed by each relation and by a self-transformation. Using learnable attention weights, the node representations are aggregated over the relations. The final node representation is learned by aggregating the representations from different layers and used to predict whether the node is a fraud.
• Figure 2: The kernel density estimate plots of $\beta_{ik}^{(l)}$ attention coefficients at each layer in YelpChi ($L=2$) and Amazon ($L=3$).
• Figure 3: $\text{\Large{$\gamma$}}_{il}$ attention coefficients in YelpChi and Amazon.