School of Information Science and Engineering, Yanshan University, Qinhuangdao, China

---

Abstract—Recent research has shown the significant vulnerabilities of collaborative recommender systems in the face of profile injection attacks, in which malicious users insert fake profiles into the rating database in order to bias the system’s output. A single Support Vector Machine (SVM) approach for the detection of profile injection attacks, however, suffers from low precision. With this problem in mind, in this paper we propose a meta-learningbased approach to detect such attacks. In particular, we propose an algorithm to create the diverse base-level training sets through flexible combination of various attack types. Combining the created training sets with SVM, we construct the base-level and meta-level classifiers. Based on these classifiers, we present a meta-learning-based detection algorithm which uses the meta-classifier to integrate the outputs of the base-classifiers and generates the final results of detection. The diversities among the base-classifiers effectively reduce the correlation of the misclassifications and improve the predictive capability of the meta-level. We conduct comparative experiments with a single SVM and the voting-based ensemble method on different-scale MovieLens datasets. The experimental results show that the proposed approach can effectively improve the precision under the condition of holding a high recall.

Index Terms—Meta-learning, support vector machine, profile injection attacks, attack detection, collaborative filtering recommendation.

[PDF]

Cite: Fuzhi Zhang, Quanqiang Zhou, "A Meta-learning-based Approach for Detecting Profile Injection Attacks in Collaborative Recommender Systems," Journal of Computers vol. 7, no. 1, pp. 226-234, 2012.