Finding Shilling Attack in Recommender System based on Dynamic Feature Selection

Abstract

Recommender system is widely used as an important tool in various fields for effectively dealing with information overload, and collaborative filtering algorithm plays a vital role in the system. However, such system is highly vulnerable to malicious attacks, especially shilling attack because of data openness and independence. Therefore, detecting shilling attack has become an important issue to ensure the security of recommender system. Most of existing methods for detecting shilling attack are based on rating classification features and their limitation is that they are easily to be interfered by obfuscation techniques. Moreover, traditional detection algorithms can not handle multiple types of shilling attack flexibly. In order to solve these problems, in this paper, we propose an outlier degree shilling attack detection algorithm based on dynamic feature selection. By considering the differences of user choosing items and taking user popularity as a detection metric, as well as using information entropy to select detection metrics dynamically, a variety of shilling attack models can be dealt with flexibly. Experiments show that the algorithm has stronger detection performance and interference immunity in shilling attack detection.

Shilling attack detection based on improved clustering algorithm

“…We divide each feature vector into two parts according to the real users and the attacking users in the users profile Dt, and calculate the proportion of the feature values of the two parts to the total of all the feature values [11]. The proportion of the attack users eigenvalues is shown in formula (1).…”