基于匿名隐私保护模型的上下文推荐系统研究

发布时间：2018-01-04 13:36

本文关键词：基于匿名隐私保护模型的上下文推荐系统研究　出处：《新疆大学》2017年硕士论文　论文类型：学位论文

【摘要】：在当前的互联网时代电子商务、社交网络等网络应用已经成为日常生活中不可分割的一部分,而各种应用针对不同用户的个性化信息推荐已经作为用户获取相关信息的重要途径。然而在推荐系统中为了计算出准确的推荐结果,针对不同用户的个性化推荐系统中为了获取更加精确的推荐需要更广泛的收集用户上下文信息。上下文感知的推荐系统中通过引入上下环境信息来为用户做出推荐。上下文感知推荐系统相比传统推荐系统更具有准确性,同时又兼具普适计算及个性化等优势,可进一步提高推荐系统的准确性和用户的满意度。在上下文感知推荐系统中,搜集了大量的包含各种与个体相关的上下文信息。这些信息的搜集和发布,使得推荐系统具有双面性。一方面,这些信息的搜集和应用将会获取更加准确的推荐结果,但与此同时也会带来另一个反面,将会有更多的用户隐私信息暴露。搜集和发布这些数据前如果没有考虑到隐私相关问题,用户的敏感信息往往能够被攻击者间接或直接的获取到。若不重视用户隐私保护的相关问题很可能会导致用户个人信息被非法利用,甚至造成大量的经济或精神损失。所以对这些信息的隐私保护已经成为互联网隐私安全的重中之重。隐私安全在推荐系统中也是广受关注的问题,也是当前研究热点之一。传统的k-匿名隐私保护是一种安全有效的隐私保护算法,但不能很好抵御一些同质性或背景知识性攻击。针对在推荐系统中用户信息的隐私保护问题本文进行了如下研究:(1)分析和概述了推荐系统相关研究,以及在推荐系统中隐私安全问题的相关技术,以及对最常用的k-匿名隐私保护算法进行了分析。(2)针对传统k-匿名隐私保护算法的缺点本文提出了两种改进的算法,通过满足不同的匿名保护条件对用户隐私进行保护。第一种是根据用户敏感属性值的隐私保护程度的不同对用户信息的敏感属性进行分组,并给不同的分组设置隐私保护程度,从而对多样性的敏感信息进行有效的隐私保护;第二种是借助微聚集隐私保护的算法对用户属性进行保护,对用户的准标识符和敏感属性的没有具体要求。(3)实验验证两个保护算法均能实现在保护用户隐私信息的前提下为用户提供准确的个性化推荐。两种算法所针对的是不同的类型的数据集进行不同的聚类和隐私保护处理并对其进行泛化。
[Abstract]:In the current era of the Internet e-commerce, social networking and other network applications have become an integral part of daily life, and an important way for various applications of personalized information recommendation of different users as the user has access to relevant information. However in recommender systems in order to calculate the recommended results accurately, according to the system in order to get more accurate recommendation needs to collect user context information more widely different users personalized recommendation. Context aware recommendation system by introducing the context information to make recommendations for users. Context aware recommender system compared with the traditional recommendation system is more accurate, and both pervasive computing and personalized and other advantages, can further improve the recommendation system the accuracy and user satisfaction. In the context aware recommendation system, the collection contains a variety of large The context information associated with the individual. The information collection and dissemination, the recommendation system has two sides. On the one hand, the collection and application of this information will get recommended more accurate results, but at the same time will bring another side, there will be more exposed. User privacy information collection and dissemination of these data if without considering the relevant issues of privacy, the user's sensitive information often can be direct or indirect access to the attacker. If we do not pay attention to issues related to user privacy protection is likely to lead to users of illegal use of personal information, and even caused a lot of economic or spiritual loss. So the information privacy protection has become a priority among priorities Internet privacy. Privacy and security in recommender systems is also concern, is also one of the research hotspots. The traditional k- anonymous privacy protection Nursing is a safe and effective privacy protection algorithm, but it is not good to resist some homogeneity or background knowledge attack. For this problem the protection of user privacy information in recommendation systems of the research are as follows: (1) analysis and overview of related research on recommender system, related technology and privacy in Recommendation System Security issues k-, and the anonymous privacy protection algorithm is the most commonly used are analyzed. (2) aiming at the disadvantages of the traditional k- anonymous privacy protection algorithm and two improved algorithms are put forward, through to meet the different conditions of anonymity to protect user privacy. The first one is the sensitive attributes were grouped according to the degree of privacy sensitive user attributes the value of different user information, and give different packet privacy settings, and sensitive information on the diversity of effective privacy protection; the second is to borrow To help micro aggregation privacy protection algorithm for the protection of the user attributes, there is no specific requirements for the user's Quasi identifiers and sensitive attributes. (3) experimental validation of two protection algorithm can achieve accurate personalized recommendation for users on the premise of protecting user privacy information. Two algorithms are for different types the data sets of different clustering and privacy protection processing and generalization on it.

【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309;TP391.3

【参考文献】