基于项目流行度和新颖度分类特征的托攻击检测算法
An Shilling Attack Detection Algorithm Based on Item Popularity and Novelty Degree Features
作者：于洪涛(燕山大学 信息科学与工程学院, 河北 秦皇岛 066004;河北省计算机虚拟技术与系统集成重点实验室, 河北 秦皇岛 066004)；周倩楠(燕山大学 信息科学与工程学院, 河北 秦皇岛 066004;河北省计算机虚拟技术与系统集成重点实验室, 河北 秦皇岛 066004)；张付志(燕山大学 信息科学与工程学院, 河北 秦皇岛 066004;河北省计算机虚拟技术与系统集成重点实验室, 河北 秦皇岛 066004)
Author：YU Hongtao(School of Info. Sci. and Eng., Yanshan Univ., Qinhuangdao 066004, China;Key Lab. for Computer Virtual Technol. and System Integration of Hebei Province, Qinhuangdao 066004, China)；ZHOU Qiannan(School of Info. Sci. and Eng., Yanshan Univ., Qinhuangdao 066004, China;Key Lab. for Computer Virtual Technol. and System Integration of Hebei Province, Qinhuangdao 066004, China)；ZHANG Fuzhi(School of Info. Sci. and Eng., Yanshan Univ., Qinhuangdao 066004, China;Key Lab. for Computer Virtual Technol. and System Integration of Hebei Province, Qinhuangdao 066004, China)
收稿日期：2016-09-17 年卷（期）页码：2017,49(1):176-183
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：托攻击;项目流行度;项目新颖度;Boosting技术;集成检测
Key words：shilling attacks;item popularity;item novelty;Boosting technology;ensemble detection
基金项目：国家自然科学基金资助项目（61379116）；河北省自然科学基金资助项目（F2015203046）；河北省高等学校科学技术研究重点资助项目（ZH2012028）
中文摘要
针对有监督检测方法在检测托攻击时准确率不高的问题，提出一种基于项目流行度和新颖度分类特征的托攻击检测算法。首先，根据真实概貌和攻击概貌在选择评分项目方式上不同，从流行度和新颖度角度，提出有效区分正常用户和攻击用户的特征；然后，基于这些特征提出一种集成检测框架，通过Boosting提升技术产生多个差异较大的基分类器，并且通过融合带有权重的基分类器的预测值得到最终的检测结果。实验结果表明，基于项目流行度和新颖度分类特征的托攻击检测算法能够提高攻击检测的准确率和召回率。
英文摘要
The existing supervised detection algorithms have low precision when they detect shilling attacks.To address this problem,a shilling attack detection algorithm based on the features of popularity and novelty degree was proposed.Firstly,according to the difference between genuine and attack profiles in choosing items to rate,several features were extracted,which can effectively distinguish normal and attack users in perspective of popularity and novelty.Secondly,an ensemble detection framework based on these features was proposed.The Boosting technology was used to generate different base classifiers and the detection results were obtained by combining the predicted results of the base classifiers with weight.The experimental results showed that the shilling attack detection algorithm based on the features of popularity and novelty degree can improve the precision and recall of attack detection.
上一条：LeakDetector:隐私泄漏自动化检测方法
【关闭】
