Abstract: The essence of detecting shilling attacks is to classify real users and fake users. The existing detection algorithms have poor detection robustness against bandwagon attacks and segment attacks with options. To solve this problem, by analyzing the different distributions of ratings of real users and fake users, combined with ID3 decision tree, a shilling attack detection algorithm based on user score dispersion is proposed. The algorithm measures the dispersion of user scores by using three features: extreme score ratio, de-extreme score variance, and user score standard deviation, and uses it as the classification feature of the ID3 decision tree algorithm.  According to the information gain of different features, the feature is selected as the classification attribute, and the classifier is trained. Experiments show that the algorithm has a good detection effect on all kinds of shilling attacks, and has good robustness.

Key words: recommendation system, shilling attack detection, score dispersion, decision tree