基于特征指标推荐系统托攻击半监督检测*

推荐系统托攻击检测算法监督学习过度依赖训练集,无监督算法依赖于攻击概貌之间相似性。本文提出一种半监督托攻击检测模型,对标记用户分类计算簇中心,给出中心用户相似度特征属性。对不同攻击选择合适的特征指标,把输入用户划分到不同的簇集中,通过簇集中输入用户全部评分项为最大值的均值与标记用户对该项均值差,确定攻击项。依据特征指标对不同簇集进行两次分类,进而确定攻击对象。实验证明,该检测算法对不同的托攻击有较高的检测准确率。

Semi-supervised Detection of Shilling Attack for Recommender System Based on Characteristic Index

Attack detection algorithm of supervise learning rely on training set for recommended system overly, unsupervised algorithms rely on the similarity between attack profiles. This paper proposes a semi-supervised shilling attack detection model, classification of tagged users, the cluster center is calculated, and the mean user similarity characteristic attribute is given. Characteristic index are selected for different attacks, dividing the input user into different cluster, counting maximum number of rating items about means difference to label users and cluster of input users, determining target item. According to the characteristic index, the different clusters are classified two parts, identifying the attacker object. Experimental result shows that the detection method has higher accuracy for different attack model.