一种基于流行度分类特征的托攻击检测算法  

An Shilling Attack Detection Algorithm Based on Popularity Degree Features

作　　者：李文涛[1,2] 高旻[1,3] 李华[1,2] 熊庆宇[1,3] 文俊浩[1,3] 凌斌 

机构地区：[1]信息物理社会可信服务计算教育部重点实验室 [2]重庆大学计算机学院 [3]重庆大学软件学院 [4]英国朴茨茅次大学工程学院

出　　处：《自动化学报》2015年第9期1563-1576,共14页Acta Automatica Sinica

基　　金：国家重点基础研究发展计划(973计划)(2013CB328903);国家自然科学基金(71102065);重庆市基础与前沿研究计划项目(cstc2015jcyj A40049);中国博士后基金(2012M521680);中央高校基础研究基金(106112014CDJZR095502;CDJZR12090001)资助~~

摘　　要：基于协同过滤的推荐系统容易受到托攻击的危害,如何检测托攻击成为推荐系统可靠性的关键.针对现有托攻击检测手段使用基于评分的分类特征易受混淆技术干扰的局限,本文从用户选择评分项目方式入手,分析由此造成的用户概貌中已评分项目的流行度分布情况的不同,提出用于区分正常用户与虚假用户基于流行度的分类特征,进而得到基于流行度的托攻击检测算法.实验表明该算法在托攻击检测中具有更强的检测性能与抗干扰性.Recommendation systems based on collaborative filtering are vulnerable to shilling attacks, so how to detect attacks becomes crucial to ensure the reliability of these systems. Because the current shilling attack detection methods based on features extracted from rating patterns are susceptible to obfuscation technologies, this paper starts from a statistics analysis of the way users choose items to rate, thus getting the corresponding results of different rated items popularity degree（rated times） distributions in normal users s profiles and spam users profile. Then classification features based on popularity degree are proposed to distinguish these two types of users. Finally, a shilling attack detection algorithm based on popularity features is developed. Experiments show that the detection performance of the algorithm is superior in attack detection precision and interference resistance.

关 键 词：协同过滤 托攻击 项目流行度 幂律分布 基于流行度的分类特征 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP391.3[自动化与计算机技术—计算机科学与技术]