ABSTRACT
In cyber-physical systems, collaborative recommender schemes are vulnerable to shilling attacks. Various detection techniques were previously proposed to prevent attacks in cyber physical systems. Nevertheless, the supervised techniques are highly effective regarding the detection of those attack types. In this chapter, the authors have studied an unsupervised technique to detect shilling attacks based on user rating behaviors in cyber physical systems to overcome the limitations. Firstly, the latent Dirichlet method of Gibbs is used to derive from the user rating element sequences latent subjects of user expectations. Then it uses the user expectations methods for the mixing transition distribution. It provides many metrics to classify the variation in rating behaviors between users. When the attack size is not determined, the number of users is calculated by measuring the critical point of suspect rating activity between genuine users and attackers. The experimental results from the 1 million data set of MovieLens demonstrate the proposed technique to recall metrics that outperform baseline techniques.
