位置社交网络的服务推荐与隐私保护研究

发布时间：2018-09-07 18:49

【摘要】：近年来,大量传感器嵌入的智能移动设备的广泛应用,促进了位置社交网络的快速发展。用户能够通过移动终端在任何时间、任何地点访问位置社交网络,以获取位置相关的服务。并且,位置社交网络中的用户利用现有的定位技术(例如:GPS、Wi-Fi或者RFID等),可以互相分享感兴趣的带地理位置信息的用户社交媒体。随着大数据时代的到来,海量的服务被发送到服务代理,以提供给用户不同功能或者性能的服务选项。由于传统服务检索机制以及通信模式的限制,一方面导致服务请求者无法快速、准确地从这些海量服务中挑选出所需要的服务;另一方面,对于一些新的服务,因为缺乏一定的服务描述信息,所以无法被合适的用户获取。服务推荐系统能够根据用户的历史访问记录,为用户主动推送服务。但是,现存的位置社交网络下的服务推荐方法缺乏对用户生活习惯以及行为偏好的考虑,无法将满足其个性化需求的服务推荐给用户。此外,用户在享受位置社交网络带来的巨大便利的同时,不可避免地会遭受个人隐私泄露的风险。位置社交网络中,用户需要将真实的位置信息上传到位置社交网络服务器,以获取个性化服务体验。但是,攻击者能够利用位置社交网络服务器所具有的“诚实而好奇”特点,窃取到受害者真实的位置数据,并通过背景知识信息,推理出受害者的生活模式、行为偏好以及未来将要访问的位置等,从而对用户的人身安全造成威胁。本文重点围绕位置社交网络中如何为用户推荐满足其需求的个性化服务、如何保护用户的个人隐私以及如何在隐私保护的情况下实现个性化服务推荐展开深入研究。主要工作体现在服务推荐机制与隐私保护两个方面,包括:(1)位置社交网络的相似用户发现;(2)位置社交网络的兴趣点服务推荐;(3)位置社交网络的轨迹隐私保护以及(4)隐私保护的个性化服务推荐四个研究点,形成了一套服务推荐与隐私保护基础研究体系。具体研究工作如下:(1)在位置社交网络的相似用户发现方面,针对现有的相似用户查找方法缺乏对用户偏好的考虑,研究了一种基于移动轨迹模式的潜在好友发现方法。通过分析原始轨迹数据的特点以及分布情况,设计了两种聚类算法。同时,提出了一种基于TF-IDF的位置分类方法,生成语义信息,构建用户的移动轨迹模式。通过考虑活动序列以及类型流行度,发现潜在的相似用户。这部分内容有效解决了数据稀疏性问题,同时提高用户的服务体验质量,为未来位置社交网络中个性化服务推荐提供必要的支撑技术。(2)在位置社交网络的兴趣点服务推荐方面,针对现有的兴趣点服务推荐方法中存在的位置有限性与数据稀疏性问题,研究了一种模式与偏好感知的兴趣点服务推荐方法。利用用户与兴趣点之间的关系,将地理位置信息转换为语义位置信息,通过考虑位置流行度与用户熟悉度,构建用户偏好模型。然后,提出了一种模式提取算法,有效提取出用户的移动行为模式,匹配每个用户的移动行为模式,为目标用户挖掘出符合其偏好的候选服务。最终,设计了一种打分机制,从而为目标用户推荐前k个候选服务。这部分内容利用移动轨迹描述方法有效地反映了用户的兴趣及偏好,提高了服务系统的可扩展性,为未来位置社交网络中个性化兴趣点服务推荐提供有益的解决思路。(3)在位置社交网络的轨迹隐私保护方面,针对传统轨迹隐私保护方法存在的用户敏感信息泄露、数据可用性低以及缺乏自适应等问题,研究了一种偏好感知的轨迹隐私保护方法。将用户移动轨迹中的原始点重构为停留区域与位置区域,构建位置匿名空间。设计了一种隐私风险评级方法,根据用户对位置的偏好不同,采取不同的位置匿名机制,将匿名后的位置连接起来,以生成匿名的轨迹序列。这部分内容能够在保护用户个人隐私不被泄露的情况下,提高数据的可用性,为未来位置社交网络中自适应隐私保护提供有效的解决方案。(4)在隐私保护的个性化服务推荐方面,针对现有的轨迹隐私保护方法缺乏考虑个性化服务推荐与隐私保护之间的均衡问题,研究了一种差分隐私保护的潜在轨迹社区发现方法。利用轨迹分段技术,将原始的轨迹序列分为若干个不同的轨迹段。同时,设计了位置泛化矩阵以及轨迹序列函数,以泛化原始的位置点与轨迹段。在位置泛化矩阵生成与轨迹序列函数生成阶段,分别将拉普拉斯分布的噪声与指数分布的噪声添加到输出结果中,以使提出的最优泛化轨迹序列选择算法满足∈-差分隐私。位置社交网络服务器接收到泛化的轨迹序列,通过考虑地理距离与语义距离,将轨迹序列聚类为社区,发现具有相似行为偏好的用户,以完成个性化服务推荐。这部分内容有效地平衡了个性化服务推荐与隐私保护,为未来位置社交网络中隐私保护下的服务推荐提供可行的技术方案。本文以位置社交网络的服务推荐与隐私保护为核心,从服务推荐、隐私保护以及支持隐私保护的服务推荐三方面展开系统研究,并有机地联系起来,以构成一整套研究体系,输出一些基础研究成果,具有一定的创新性。在研究方法上,本文采用相关工作调研、数学建模、算法设计以及实验分析等一系列研究方法。在数据分析上,本文基于真实数据集,考虑了位置社交网络下服务推荐与隐私保护的实际性能指标,验证所提方法的优越性。本文取得的研究成果对未来位置社交网络的研究与发展具有一定的借鉴意义。
[Abstract]:In recent years, the widespread use of smart mobile devices embedded with a large number of sensors has promoted the rapid development of location-based social networks. Users can access location-based social networks through mobile terminals at any time and anywhere to obtain location-related services. Moreover, users in location-based social networks make use of existing location-based technologies (e.g. GP) S, Wi-Fi, RFID, etc.) can share interesting social media with geographic location information. With the advent of the big data era, a large number of services are sent to service agents to provide users with different functions or performance of service options. Service requesters can not quickly and accurately select the required services from these massive services; on the other hand, for some new services, because of the lack of certain service description information, it can not be accessed by the appropriate users. The existing service recommendation methods in location-based social networks lack the consideration of users'living habits and behavior preferences, so they can not recommend services that meet their personalized needs to users. In a network, users need to upload real location information to a location-based social network server for personalized service experience. However, attackers can take advantage of the "honest and curious" characteristics of a location-based social network server to steal the victim's real location data and infer the victim's identity through background information. This paper focuses on how to recommend personalized services to meet users'needs in location-based social networks, how to protect users' privacy and how to implement personalized service recommendation under privacy protection. The main work includes: (1) similarity user discovery in location-based social networks; (2) interest point service recommendation in location-based social networks; (3) trajectory privacy protection in location-based social networks; and (4) personalized service recommendation in privacy protection. A basic research system of service recommendation and privacy protection is proposed. The main research work is as follows: (1) In the aspect of similarity user discovery in location-based social networks, a potential friend discovery method based on mobile trajectory pattern is studied for the lack of consideration of user preference in existing similar user search methods. At the same time, a location classification method based on TF-IDF is proposed to generate semantic information and construct the user's trajectory pattern. Potential similar users are found by considering the activity sequence and type popularity. This part of content effectively solves the problem of data sparsity and improves the performance. The quality of user service experience provides the necessary support technology for personalized service recommendation in future location-based social networks. (2) In the aspect of interest point service recommendation in location-based social networks, aiming at the problems of location limitation and data sparsity in the existing interest point service recommendation methods, this paper studies the development of a pattern and preference perception. User preference model is constructed by transforming geographic location information into semantic location information and considering location popularity and user familiarity. Then, a pattern extraction algorithm is proposed to effectively extract user's mobile behavior patterns and match each user's mobile line. Finally, a scoring mechanism is designed to recommend the first k candidate services for the target users. This part effectively reflects the interests and preferences of the users, improves the scalability of the service system, and provides a future location-based social network. (3) For the trajectory privacy protection of location-based social networks, a preference-aware trajectory privacy protection method is proposed to solve the problems of user-sensitive information leakage, low data availability and lack of self-adaptation in traditional trajectory privacy protection methods. The original point in the trajectory is reconstructed into the residence area and the location area to construct the location anonymity space. A privacy risk rating method is designed. According to the different preferences of the users, different location anonymity mechanisms are adopted to connect the anonymous locations to generate the anonymous trajectory sequence. (4) In the aspect of personalized service recommendation for privacy protection, the existing trajectory privacy protection methods do not consider the balance between personalized service recommendation and privacy protection. A potential trajectory community discovery method for differential privacy preservation is studied. The original trajectory sequence is divided into several different trajectory segments by using the trajectory segmentation technique. At the same time, the position generalization matrix and the trajectory sequence function are designed to generalize the original position points and trajectory segments. In the generation phase, the Laplace distribution noise and the exponential distribution noise are added to the output results respectively to satisfy the < - differential privacy of the proposed optimal generalized trajectory sequence selection algorithm. This part effectively balances personalized service recommendation and privacy protection, and provides feasible technical solutions for service recommendation under privacy protection in future location-based social networks. This paper focuses on service recommendation and privacy protection in location-based social networks. In order to form a complete research system and output some basic research results, this paper has a certain degree of innovation. In the research method, this paper uses related research, mathematical modeling, algorithm design and experimental points. Based on the real data set, this paper considers the actual performance indicators of service recommendation and privacy protection in location-based social networks, and verifies the superiority of the proposed method.
【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.3;TP309

【参考文献】