A NOVEL APPROACH FOR IDENTIFYING GROUP SHILLING ATTACKS IN RECOMMENDATION SYSTEMS

Abstract

Existing shilling attack detection algorithms are largely focused on identifying individual attackers in online recommender systems, and they seldom handle group shilling assaults, in which a group of attackers colludes to influence an online recommender system's output by providing fraudulent profiles. In this paper present a method for identifying shilling as a group assaults that incorporates both bisecting K-means clustering and hierarchical clustering methods in this research. First the rating track is extracted from each item’s and divides it into candidate groups based on a predetermined time period. Second, the degree of item’s attention and user interaction parameters are used to determine candidate group suspicious degrees. Finally, this work generate attack groups by grouping candidate groups based on their suspicious degrees using the bisecting K-means approach and the hierarchical clustering algorithm. Experiments are carried out using Amazon datasets. There are 103,297,638 ratings from 480,186 users on 17,770 products in the dataset. The proposed solution outperforms traditional approaches. The performance metrics used to evaluate the proposed systems are accuracy, precision, recall and f1-score.The proposed system provides 98% of accuracy, 90 % of precision, 98% of recall and 99% of f1-score.The goal of this research is to compare the bisecting k-means method with hierarchical clustering method and the results show hierarchical clustering based (proposed) method out performs bisecting k-means clustering method.