CG-HC: Candidates Grouping and Hierarchical Clustering based Group shilling Attacks detection in Recommender System

Abstract

In the past ten years, a variety of techniques for spotting shilling assaults in recommender systems have been put forth. The current strategy for detecting shilling attacks focuses on detecting single attackers, but rarely addresses group shilling attacks, whereas team of attackers uses false profiles to manipulate a digital recommender system's results. In some technique’s researchers used users individual and group features separately to form the groups. In this article, we developed a three-phase technique for identifying group shilling attacks. In first phase, we build tight candidate groups using user behaviour features. In second phase, we determined the degree of suspicion for each user in the group and using a hierarchical clustering technique splitted each group into two clusters. Finally, we applied group suspicious measures to find the attack groups. On real-world datasets like Netflix and Amazon reviews, the developed technique beats the baseline when compared with baseline detection methods.