TFD: Trust-Based Fraud Detection in SIoT With Graph Convolutional Networks

With the rapid development of Artificial Intelligence (AI) technology, its application in areas such as Social Internet of Things (SIoT) and e-commerce has become increasingly widespread. However, there are many uncertainties and potential risks in the widespread deployment of AIoT, and miscreants take advantage of these vulnerabilities to commit fraud on SIoT, which raises serious public concerns about security and privacy. Our work is dedicated to distinguishing fraudsters from consumers through fraud. Given the superior ability of Graph Neural Networks (GNN) in modeling network environments, it has become an important tool in the field of fraud detection. In this work, we propose a fraud detection method based on potential trust between users in SIoT, which combines the graph structure of SIoT with pairwise trust relationships between users. Specifically, we construct an end-to-end framework that utilizes Graph Convolutional Networks (GCNs) to learn and mine potential trust patterns in different social relationships. Further, a trust-aware neighborhood difference aggregation method is introduced by us to significantly distinguish normal consumers from fraudsters. To cope with the class imbalance problem, we employ an imbalance-aware loss function for optimization. Extensive experiments on two datasets confirm that our model is very effective compared to existing techniques.