An outlier accuracy improvement in shilling attacks using KSOM

Abstract

Due to the rapid technological changes, these days collaborative filtering-based recommender systems are being widely used worldwide. Collaborative filtering approach is more vulnerable from being attacked because of its open nature. The attackers may rate the fake ratings to disturb the systems. In this paper, unsupervised Kohonen Self-Organizing Map (KSOM) clustering technique is used to make a better detection between genuine and fake profiles to reduce profile injection attacks and compared with existing techniques Enhanced Clustering Large Applications Based on Randomized Search (ECLARANS) and Partition Around Medoids (PAM) with variants of attack size. It has been noticed that KSOM outperforms over ECLARANS and PAM techniques with good outlier accuracy.