基于特征指标推荐系统混淆托攻击半监督检测模型

Semi-supervised detection of obfuscated shilling attack for recommend system based on feature index vector

本文利用标记用户和特征指标，计算高信任度标记用户特征指标向量的特征值进而压缩输入用户概貌，使得攻击概貌和普通概貌的特征指标值富集到空间的两端最终识别攻击用户。实验证明，该检测算法对混淆托攻击模型在不同填充率下有较高的检测正确率。

This paper calculated characteristic value with high made attack profile and normal profile gather into both ends of th results show that the detection has a higher detection accuracy for trust user label e space to detect confused shilling to compress user profile and attack user. The experimental attack in different fill rate.