协同过滤推荐系统概貌注入式攻击攻击特征提取研究

发布时间：2018-03-02 23:17

本文选题：协同过滤推荐　切入点：攻击概貌检测　出处：《燕山大学》2014年硕士论文　论文类型：学位论文

【摘要】：协同过滤推荐技术作为应用最广的个性化推荐技术之一，被认为是解决信息爆炸时代信息过载问题的有效方法。但是由于协同过滤推荐系统的开放性和用户参与性，系统存在严重的安全隐患。一些恶意用户出于商业的目的，通过向系统中注入大量的虚假用户概貌来使得推荐系统产生有利于他们自己的推荐结果。因此，如何保障协同过滤推荐系统的安全成为了协同过滤推荐技术研究中的一个热点。本文在对国内外研究现状深入分析的基础上，进一步对协同过滤推荐系统的攻击特征提取进行了研究。针对已有的攻击特征种类还不够丰富，，攻击检测能力不强的问题，从丰富攻击特征的角度上。首先，在深入分析攻击概貌特征的基础上，基于传统的攻击检测特征只关注攻击评分值的分布特征，忽视了攻击概貌在选择填充项目时随机选择的这个特点，提出了攻击概貌的关联规则特征。从关联规则挖掘的角度，找到项目中存在的强关联规则，利用攻击用户比正常用户满足关联规则的概率低这个特点，来对攻击概貌进行检测，提高了对攻击用户的检测能力。其次，攻击概貌和正常用户概貌在评分分布上的区别使得攻击概貌的目标项目在受攻击前后评分分布有着巨大的变化，而已有的攻击特征没能很好的反映攻击概貌的这个特点。针对这个问题，提出攻击概貌的目标项目特征，能够描述项目在受攻击前后平均评分值的变化，选择其中大于一个阈值的项目作为攻击概貌的目标项目。本文的研究能够有效的丰富攻击特征，提高攻击概貌检测能力。最后，对本文提出的两种攻击特征进行了实验验证与分析，验证其有效性，并且对今后的研究工作进行了展望。
[Abstract]:As one of the most widely used personalized recommendation technologies, collaborative filtering recommendation technology is considered to be an effective method to solve the problem of information overload in the era of information explosion, but because of the openness and user participation of collaborative filtering recommendation system. There are serious security risks in the system. For commercial purposes, some malicious users make the recommendation system produce recommendation results in their own interests by injecting a large number of false user profiles into the system. How to ensure the security of collaborative filtering recommendation system has become a hot spot in the research of collaborative filtering recommendation technology. Furthermore, the attack feature extraction of collaborative filtering recommendation system is studied. In view of the problem that the existing attack features are not rich enough and the ability of attack detection is not strong, from the point of view of enriching attack features, firstly, on the basis of in-depth analysis of attack general features, Based on the fact that the traditional attack detection features only focus on the distribution of attack score, and neglects the random selection of attack profile when selecting filling items, the association rule feature of attack profile is proposed. Find the strong association rules existing in the project, take advantage of the fact that the probability of attacking users meeting the association rules is lower than that of normal users, to detect the attack profile, and improve the detection ability of attack users. Secondly, The difference between the attack profile and the normal user profile results in a huge change in the score distribution of the target item before and after the attack. However, the existing attack features can not well reflect the characteristics of the attack profile. In view of this problem, the target item feature of the attack profile can describe the change of the average score of the item before and after the attack. The research in this paper can effectively enrich the attack characteristics and improve the ability of attack profile detection. Finally, the two attack features proposed in this paper are tested and analyzed to verify their effectiveness, and the future research work is prospected.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.3;TP393.08

【参考文献】