Demolish falsy ratings in recommendation systems
Ossama H. Embarak · 2019
Many recommendation systems suffer from overrun false rating injections, which affect and shift generated recommendations in collaborative networks. Previous research focuses more on how to distinguish between profiles ignoring individual characteristics. This paper studies different behaviors and attributes of attackers and isolates them from genuine users. This paper suggested a new algorithm to filter two types of attacks, push and nuke attacks, where it evaluates users' ratings according to the proposed algorithm and classify the rating as melt, freeze, or normal according to the calculate variance from the center. The suggested algorithm gave high potentials and was able to filter efficiently such malicious attacks.