Bipartite Graph Anomaly Detection With Contrastive Learning

Bipartite graphs are widely used to model social fintech interactions, particularly user-transaction and investor-project relationships. The rapid shift toward digital financial ecosystems has heightened the need for secure and reliable artificial intelligence (AI) models to detect anomalies. Existing graph neural network (GNN) models for anomaly detection often concentrate on homogeneous graphs and address node or edge anomalies, limiting their ability to capture the full complexity of bipartite interactions. These limitations reduce their effectiveness in protecting data and trust within social fintech applications. To overcome these challenges, we propose the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">bipartite graph anomaly detection with contrastive learning (CL-BGAD)</i>, an innovative framework designed to enhance AI model security by comprehensively capturing intricate interactions in bipartite graphs. The encoder features a node attention mechanism that prioritizes critical interactions among nodes, while the structure decoder leverages contrastive learning to discern between normal and anomalous connections. This dual strategy ensures robust anomaly detection at both the node and edge levels, contributing to greater adaptability in social fintech applications. Experimental results validate that CL-BGAD consistently outperforms state-of-the-art baselines across diverse real-world datasets, reinforcing the model’s capability to maintain secure and responsible AI-driven systems. The source code and datasets are publicly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/DSI-Lab1/CL-BGAD</uri>.