Performance Analysis of Classifiers in the Detection of Injection Attacks by the Association of Graph-Based Method and Generic Detection Attributes

The rising prevalence of recommender systems in personalization recommendation services, exemplified by platforms like TripAdvisor, Amazon, Netflix, and YouTube, underscores the critical need for robust and user-friendly abnormality detection techniques. This demand is driven by the inherent vulnerabilities of recommender systems and their susceptibility to various forms of attacks due to their open nature. The primary objective is to develop a unified detection system that offers a new perspective on various severe dangers. The process begins with creating an association graph based on the innovative ranking matrix. This design is based on the underlying reason for user ratings and the atomic propagation rules governing linked networks. The trustworthiness of link behaviours within the specified network of linked association networks is then assessed. The study uses a factor graph model to evaluate the credibility of interconnected networks, identifying suspicious users and items through generic attributes analysis. The efficacy of the proposed detection methodology is validated through wide-ranging trials conducted on real data simulating profile injection attacks and co-visitation injection attacks. Finally, the attacks are classified with various classifiers such as Support Vector Machines (SVM), Naive Bayes Classifier (NBC), Logistic Regression (LR), Voting Classifier (VC), Gradient Boosting Classifier (GBC), and Multiple Linear Perceptron (MLP). The performance of the results has overcome some of the state-of-the-art approaches.