融合信息熵与信任机制的防攻击推荐算法研究  

RESEARCH ON ANTI-ATTACK RECOMMENDATION ALGORITHM WITH FUSION OF INFORMATION ENTROPY AND TRUST MECHANISM

作　　者：郝志峰[1] 牛晓龙[1] 蔡瑞初[1] 温雯[1] 

机构地区：[1]广东工业大学计算机科学与技术系,广东广州510006

出　　处：《计算机应用与软件》2015年第3期284-288,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61070033;61100148;61202269);广东省自然科学基金项目(S2011040004804);广东省科技计划项目(2010B050400011);软件新技术国家重点实验室开放课题(KFKT2011B19);广东高校优秀青年创新人才培育项目(LYM11060);广州市科技计划项目(12C42111607;201200000031);番禺区科技计划项目(2012-Z-03-67)

摘　　要：由于对用户偏好信息的过分依赖,致使推荐系统易受到恶意攻击,从而影响系统的推荐质量。提出一个融合信息熵与信任机制的防攻击推荐算法。在考虑了托攻击与正常用户之间的评分变化幅度差异基础上,提出融合信息熵的相似性改进算法,同时引入信任更新机制,在推荐过程中将用户间信任度与相似度有机相结合,通过筛选推荐权重较高的邻居用户方法获得可靠推荐,从而降低恶意攻击对系统的影响。通过在真实数据集上实验表明该算法在提高推荐系统的准确性和脆弱性上有较好的表现。Undue reliance on user＇s preference information results in the recommendation systems being vulnerable to malicious attacks,therefore impacts the quality of system recommendation. In this paper we propose an anti-attack recommendation algorithm which fuses the information entropy and the trust mechanism. Based on considering the difference of rating variation range between the shilling attacks and normal users,we present the similarity improvement algorithm with the information entropy fused,at the time,the trust update mechanism is introduced as well. In recommendation process the trust degrees between users are combined in an organic way with the similarity,and by the means of recommending the neighbouring users with higher weights through screening the algorithm obtains reliable recommendations so as to reduce the impact of the malicious attacks on the system. It is revealed by the experiments on real data sets that the proposed algorithm has better performance in improving the accuracy and vulnerability of the recommendation system.

关 键 词：协同过滤 恶意攻击 评分变化幅度 复合权重 脆弱性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]