UTH-SAD: User Trust-Based Hypergraph Decomposition Model for Shilling Attack Detection

Recommendation systems play a crucial role in enhancing user engagement across e-commerce and social platforms. However, these systems are increasingly vulnerable to shilling attacks, where adversaries manipulate recommendations by injecting fake profiles. Traditional detection methods, relying on manual feature engineering or conventional graph models, struggle to capture sophisticated attack patterns due to evolving adversarial strategies and the complexity of user-item interactions. To address these limitations, we propose a novel method called user trust-based hypergraph decomposition model for shilling attack detection (UTH-SAD). This method integrates user trust relationships, rating prediction errors, and hypergraph spectral features to provide robust shilling attack detection. By leveraging multi-granularity hypergraph analysis, UTH-SAD models higher-order user-item relationships, capturing intricate interactions that traditional methods fail to detect. The method combines trust relationships derived from shared rating behaviors, spectral features from hypergraph analysis, and rating prediction errors to identify malicious profiles effectively. This approach not only enhances the detection accuracy of shilling attacks but also provides insights into the complex interactions within recommendation systems, making it a promising solution for securing recommendation systems against evolving adversarial strategies. Extensive experiments conducted on MovieLens and Amazon demonstrate that UTH-SAD achieves superior performance.