Multi-Scale Feature Fusion-Based Shilling Attack Detection Algorithm

With the rapid development of the internet and information technology, recommendation systems have become a crucial tool for delivering personalized information to users. However, these systems are often vulnerable to shilling attacks, which can significantly degrade the accuracy of recommendations. To address the issues of poor model generalization and suboptimal detection of cold-start fraudulent users in existing shilling attack detection methods, we propose a multi-scale feature fusion detection algorithm. The algorithm leverages a multi-scale neural network to extract users' rating profiles and employs a Variational Autoencoder to uncover latent features in the ratings. By incorporating self-attention mechanisms, probability density distributions, and user preferences, we construct a multi-scale feature fusion neural network model for attack detection. Experimental results demonstrate that the proposed algorithm exhibits robust performance across various attack scenarios, showing superior detection accuracy, especially in identifying cold-start fraudulent users at lower fill rates.