针对协同过滤推荐系统的混淆托攻击模型  

Shilling Attack Model of Obfuscation for Collaborative Filtering Recommender System

作　　者：卫星君[1] 顾清华[2] WEI Xingjun;GU Qinghua(Shaanxi Energy Institute,Xianyang 712000;School of Management,Xi＇an University of Architecture ＆ Technology,Xi＇an 710055)

机构地区：[1]陕西能源职业技术学院,咸阳712000 [2]西安建筑科技大学管理学院,西安710055

出　　处：《计算机与数字工程》2018年第8期1575-1579,1696,共6页Computer & Digital Engineering

摘　　要：为了躲避现有的托攻击检测,攻击者利用混淆技术降低攻击概貌和普通概貌之间的区别,使之成为更多用户的近邻进而影响推荐系统的预测评分。对物品按属性划分,在混淆技术的基础上,给出噪音系数、评分偏移函数和目标偏移函数,提出混淆流行交叉托攻击模型。设计攻击概貌自动产生器注入系统数据库,同最流行项中添加平均项构造的混淆托攻击对比。实验结果表明,该托攻击模型对推荐系统的危害性更大,应加强防范。In order to avoid the existing attack detection,the attacker uses the obfuscation techniques to reduce the difference between the attack profiles and the general profiles,and then to become more users neighbor,the predictive score of the recom-mended system is affected. Items are classified by attributes,the noise coefficient,the score offset function and the target offset function are given on the basis of confusing techniques. The popular crossing model of confusion is proposed. Attack profiles automatic generator is designed to inject system database,this model is compared with average-over-popular. The experimental result shows that the attack model is more harmful to the recommender system,prevention should be strengthened for recommender system.

关 键 词：推荐系统 攻击模型 托攻击 混淆技术 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]