基于虚假概貌协同作用的托攻击检测算法研究

发布时间：2018-05-04 00:10

本文选题：推荐系统 + 攻击检测　；参考：《燕山大学》2016年硕士论文

【摘要】：信息时代的到来,方便了我们的生活,拓展了我们的眼界。与此同时出现的问题是,信息的过载给人们带来麻烦,想要快速找到所需的信息所付出的代价越来越高。搜索引擎和协同过滤技术是当前解决这一问题的两种主流手段。值得一提的是协同过滤技术在个性化定制方面对用户具有很强的吸引力。伴随电子商务的发展,协同过滤技术正融入到其中,成为推荐系统的核心部分,推荐系统的出现大大提升了用户的购物、听音乐等使用体验。然而由于推荐系统本身的开放性,使得其容易遭受恶意用户的攻击,直接影响到了用户的使用体验,间接影响了电子商务的生存。本文首先分析了目前对于这一问题研究的主要解决技术和国内外现状,同时对于托攻击、托攻击模型、托攻击特征等进行了详细的描述。针对托攻击通过协同作用影响推荐系统这一问题,本文主要围绕托攻击的特点以及作用的方式来进行了思考与研究。对于识别单个用户概貌的托攻击检测算法,推荐系统中的“专家型”用户往往被错误标记,它们所展现的“与众不同”的特征与虚假概貌会很相似。首先,基于信号理论当中的去噪原理,本文基于主成分分析方法对其进行了改进,结合了逻辑斯蒂回归进行有监督分类,该算法能够有效地去除对攻击强度这一先验知识的依赖,并且在准确率这一评价指标上有较好的表现,该算法具有较好的实际应用价值。然后,对于多种混合攻击类型,先前所提出的算法效果较差。对于此问题,本文提出了结合信息熵和主题模型的托攻击检测算法,使用主题模型得到用户的主题分布,托攻击概貌的主题集中,即对应的信息熵较小;相反地,正常用户的通常含有多个主题,即对应的信息熵较大。最后,对前面所提出的算法进行了实验验证,将两个算法在两个不同的数据集上进行对比实验和结果分析。结果表明,本文提出的改进后的算法相较于原始算法,大大提高了预测的准确率。
[Abstract]:The arrival of the information age has facilitated our lives and broadened our horizons. At the same time, the problem is that the overload of information brings trouble to people, and the cost of finding the information quickly is becoming higher and higher. Search engine and collaborative filtering technology are two main methods to solve this problem. It is worth mentioning that collaborative filtering technology has a strong appeal to users in personalized customization. With the development of electronic commerce, collaborative filtering technology is becoming the core part of the recommendation system. The appearance of the recommendation system greatly improves the user's experience of shopping, listening to music and so on. However, because of the openness of recommendation system, it is vulnerable to malicious user attacks, which directly affects the user's experience and indirectly affects the survival of e-commerce. This paper first analyzes the main research technologies and the current situation at home and abroad for this problem, and at the same time, describes in detail the supporting attack, the model of the supporting attack, the characteristics of the supporting attack, and so on. Aiming at the problem that the depot attack affects the recommendation system through synergy, this paper mainly focuses on the characteristics of the depot attack and the way in which it works. For the trust attack detection algorithm for recognizing the profile of a single user, the "expert" users in the recommendation system are often wrongly marked, and their "distinctive" features will be very similar to the false profile. Firstly, based on the principle of de-noising in signal theory, this paper improves it based on principal component analysis (PCA). The algorithm can effectively remove the dependence on the prior knowledge of attack intensity and has a good performance on the evaluation index of accuracy. The algorithm has good practical application value. Then, for various types of mixed attacks, the proposed algorithm has a poor effect. In order to solve this problem, this paper proposes an algorithm combining information entropy and topic model, which can get the user's topic distribution, the theme set of the general profile of the support attack, that is, the corresponding information entropy is small; on the contrary, Normal users usually contain more than one topic, that is, the corresponding information entropy is larger. Finally, the proposed algorithm is verified by experiments, and the two algorithms are compared with each other on two different data sets. The results show that the improved algorithm greatly improves the prediction accuracy compared with the original algorithm.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】