A New Mechanism for Detecting Shilling Attacks in Recommender Systems Based on Social Network Analysis and Gaussian Rough Neural Network with Emotional Learning

Abstract

A recommender system is an integral part of any e-commerce site. Shilling attacks are among essential challenges in recommender systems, which use the creation of fake profiles in the system and biased rating of items, causing the accuracy to decrease and the correct performance of the recommender system in providing recommendations to users. The target of attackers is to change the rank of content or items corresponded to their interests. Shilling attacks are a threat to the credibility of recommender systems. Therefore, detecting shilling attacks it necessary to in recommender systems to maintain their fairness and validity. Appropriate algorithms and methods have been so far presented to detect shilling attacks. However, some of these methods either examine the rating matrix from a single point of view or use low-order interactions or high-order interactions. This study aimed to propose a mechanism using users' rating matrix, rating time, and social network analysis output of users' profiles by Gaussian-Rough neural network to simultaneously use low-order and high-order interactions to detect shilling attacks. Finally, several experiments were conducted with three models: mean attack, random attack, and bandwagon attack, and compared with PCA, Semi, BAY, and XGB methods using precision, recall, and F1-Measure. The results indicated that the proposed method is more effective than the comparison methods regarding attack detection and overall detection, which proves the effectiveness of the proposed method.