An Unsupervised Algorithm for Detecting Shilling Attacks on Recommender Systems

Abstract

Shilling attack is one of the significant security problems involved in recommender systems. Developing detection algorithms against shilling attacks has become the key to guaranteeing both the preciseness and robustness of recommender systems. Considering the low degree of unsupervised features the existing algorithms suffer from, this paper proposes an iterative Bayesian inference genetic detection algorithm (IBIGDA) through the introduction of the quantitative metric for the group effect of attack profiles and the corresponding object function for genetic optimization. This algorithm combines the posterior inference for the adaptive parameters with the process of attack detection, thus relaxes the dependence of the detection performance on the relating prior knowledge of the systems. Experimental results show that this algorithm can effectively detect shilling attacks of typical types.