摘要
传统的托攻击检测方法多采用基于评分值差异的算法,其在小规模情况下易造成误判率过高的问题。通过分析真实用户和攻击用户评分项目选择方式的差异,文中提出了一种基于混合特征值的托攻击检测算法。该算法在Degsim、MeanVar、WDA特征检测指标组成的特征模型基础上,加入了流行项目卡方估计值(Chi-square of popular item,CHIP)、新颖项目卡方估计值(Chi-square of novel item,CHIN)两个特征检测指标,构成一种新的特征模型。该特征模型在传统方法的基础上,提出对项目与流行项目、项目与新颖项目之间的关联程度的考量,依据特征属性选择K-means聚类与阈值判断相结合的分类方法,可有效区分攻击用户和正常用户。实验对比表明,该算法在小规模情况下可有效解决误判率高的问题,具有更好的检测准确度。
Aiming at the problem of high misjudgment rate in shilling attack detection of traditional differential algorithm based on score value,we propose a shilling attack detection algorithm based on the hybrid eigenvalue by analyzing the selection mode difference of score item between real users and false users.The algorithm adds two feature detection indexes CHIP(Chi-square of popular item)and CHIN(Chi-square of novel item)to form a new feature model on the basis of the feature model with three feature detection indexes,Degsim,MeanVar and WDA.In the new feature model,we consider the correlation degrees between the item and popular item,and between the item and novel item based on the traditional differential algorithm,and according to characteristic attribute,adopt the classification method which combines K-means clustering and threshold judgment to distinguish the attacking user from the normal user effectively.Experiment shows that the proposed algorithm can effectively solve the problem of high misjudgment rate in small scale and has better accuracy.
作者
雷梦宁
丁爱玲
王新美
韩佳倩
曹苗
LEI Meng-ning;DING Ai-ling;WANG Xin-mei;HAN Jia-qian;CAO Miao(School of Information Engineering,Chang’an University,Xi’an 710061,China)
出处
《计算机技术与发展》
2021年第10期87-92,共6页
Computer Technology and Development
基金
国家自然科学基金-青年科学基金项目(61806023)。
关键词
推荐系统
托攻击
混合特征
卡方估计值
聚类算法
recommendation system
shilling attack
hybrid feature
Chi-square
clustering algorithm