基于主成分分析的无先验知识的攻击检测算法

Approach of Attack Detection Algorithm Based on PCA Varselect Without Prior Knowledge

针对现有托攻击检测算法在攻击强度较小时正确识别率低、实施成本高的缺点,在基于主成分分析变量选择(Variableselection using principal component analysis,PCA Var Select)的基础上,提出一种无需事先知道攻击强度的攻击检测算法.首先,利用PCA Var Select方法得到每个用户的主成分值;然后,选取主成分值较小范围的用户集,用来确定被攻击物品和评分向量长度;最后,将对被攻击物品评高分的用户集、具有嫌疑评分向量长度的用户集、主成分值较大范围的用户集三者取交,得到攻击用户集.经过实验验证,该算法在去除PCA Var Select对先验知识依赖的同时,同时也提高了准确率,在面对小规模攻击强度时表现依然良好.

Thinking about the problem that the precision of the existing user profile attack detection algorithms is not high facing the small attack size and the high cost of its implementation. so we propose an approach based on PCA Var Select without prior knowledge. firstly,we use PCA Var Select to get the principal component value of each user; then,we select a smaller range of the principal component value to confirm which item was attacked and the filler size of the shilling attack; last,we get suspected users set from the intersection of three sets,they are the user sets whose attacked item is rated high point,the user sets whose quantity of rated items is the filler size and the user sets whose principal component value in the larger scope. After test,the experimental results showthat the proposed method not only don＇t depend on the prior knowledge but also effectively improve the precision,what＇ s more,it still performs better when facing the smaller attack size.