A trust attack detection algorithm based on improved K-means clustering

In the current unsupervised attack detection, the classification attribute of proxy attack has the problem that the resolution of attack users and ordinary users is not high, which affects the recommendation accuracy. This paper proposes a trust attack detection algorithm based on K-means clustering and user rating interval weighted RDMA (SRDMA). In this algorithm, the user reputation IGR classification feature attribute is introduced, and the two classification features with the highest discrimination are extracted by using the distinction coefficient, and then the traditional K-means clustering algorithm is improved through the user reputation IGR. The improved K-means clustering algorithm can cluster most attackers together. Finally, in view of the fact that there are misjudged real users in the attack user set after clustering, this paper uses user rating interval weighted RDMA algorithm (SRDMA) for secondary classification, so as to gather as few real users as possible, and reduce the impact of trust attack on the recommendation system. The experimental results of real MovieLens show that the proposed attack detection algorithm based on improved K-means clustering and SRDMA can effectively detect the random proxy attack profile model, and the detection effect is obviously better than the comparison algorithm.