Shilling Attack Detection Scheme in Collaborative Filtering Recommendation System Based on Recurrent Neural Network

Abstract

With the prosperity of the modern electronic business, merchandise recommendation system has become an important tool for online shopping. However, the threat of shilling attack caused by injecting fake rating records into the system cannot be ignored. To deal with shilling attacks, many methods especial user profile-based detection methods have been proposed. But there are still remainder challenging problems in those methods: (1) detection attributes need to be designed in advance; (2) instability of detection effect when faced with variety of attack models; (3) other aspects such as low accuracy, high computing cost and failure in detecting some special shilling attacks. Therefore, a shilling attack detection scheme based on neural network is proposed in this paper in order to address these challenging problems. In this scheme, the LSTM model is used to learn the historical rating records, and then predict the ratings for the next period when the desired accuracy is achieved. Chi-square test is used to determine whether the item is under attacks by comparing the predicted ratings and the actual ones within the time period. The simulation of experimental results on MovieLens 20M dataset show that our proposal is feasible and effective, and it improves the detection performance.