Predicting User Privacy Preferences based on Dynamic Interpersonal Relationships and Content Sensitivity Analysis

越来越多的用户在在线社交媒体上注册，并贡献了大量的个人用户生成内容。然而，分享内容可能会危及用户的隐私，如果这些媒体被不适当的受众访问，则会产生严重后果。例子包括解雇或失去健康保险福利。为了保护他们的隐私，用户可以使用现有的解决方案手动控制内容的发布和访问。然而，目前的技术水平一再被证明在适当地支持用户完成这项任务方面效率低下。由于用户通讯录数量庞大，共享频率高，手动选择访问控制规则的过程繁琐且耗时。由于现有解决方案缺乏直观性，用户通常不修改默认的共享设置，并且随着时间的推移放弃更新它们。这通常会导致与不合适的受众分享更多的内容，从而将用户的隐私置于危险之中。相比之下，其他用户由于担心现有解决方案的复杂性而选择不合适的受众，因此不愿在网上分享内容。因此，这可能会减少用户从他们的在线社交网络中获得的好处。因此，该提案的目标是允许用户在利用在线社交网络的同时更好地保护他们的隐私。我们的目标是解决当前部署的解决方案的缺点，并为一种新型的访问控制机制提供基础。我们将在本项目中探索的一种很有前途的方法是动态地建议为用户量身定制的共享设置。我们将研究如何根据用户之间的人际关系特征和他们的社交联系人以及将要发布的内容对用户隐私的敏感性来生成准确的建议。为了满足我们方法的可用性，我们将遵循以用户为中心的方法，开发一个非专业用户易于使用和理解的解决方案，以促进其被大规模接受。因此，我们将研究和开发创新的方法，分别从用户的传播模式和基于其特征的内容敏感程度来自主推断用户社会关系的性质和强度。开发的方法将尊重用户的隐私，更倾向于在本地处理用户设备上随时可用的个人数据。因此，在我们的项目中，我们将采用挖掘用户数据的新方法，通过为他们提供更细粒度的访问控制机制，同时减少相关的配置负担，帮助他们更好地保护自己的隐私。

An increasing number of users are registered in online social media and contribute a tremendous amount of personal user-generated content. Sharing content can however endanger users' privacy and have serious consequences if such media are accessible to an inappropriate audience. Examples include job dismissals or loss of health-insurance benefits. To protect their privacy, users can manually control the release and access of their content using existing solutions. However, the current state-of-art has repeatedly been demonstrated to be inefficient in appropriately supporting users in this task. Both the commonly observed substantial number of users' contacts and the high sharing frequency make the manual selection of access control rules a cumbersome and time-consuming process. Due to the lack of intuitiveness of existing solutions, users often do not modify default sharing settings and renounce in updating them over time. This regularly results in sharing more content than desired with an unsuited audience, hence putting the users' privacy at stake. In contrast, other users refrain from sharing content online by fear of selecting an inappropriate audience due to the complexity of existing solutions. It hence potentially reduces the benefits drawn by the users from their online social network. The goal of this proposal is therefore to allow users to simultaneously take advantage of online social networks while better protecting their privacy. We aim at addressing the shortcomings of currently deployed solutions and providing the foundations for a novel form of access control mechanisms. A promising approach that we will explore in this project is to dynamically suggest sharing settings tailored to the users. We will examine how accurate suggestions can be generated based on features of interpersonal relationships between users and their social contacts and the sensitivity of the content to be published with regards to the users' privacy. To cater for the usability of our approach, we will follow a user-centric approach and develop a solution that is easy to use and comprehend by non-expert users in order to foster its acceptance at large scale. We will hence investigate and develop innovative methods to autonomously infer both the nature and the strengths of the users' social relationships from their communication patterns and the degree of content sensitivity based on its characteristics, respectively. The developed methods will respect users' privacy by preferring local processing of readily available personal data on their devices. In our project, we will hence adopt the novel approach of mining users' data to help them in better protecting their privacy, by providing them more fine-granular access control mechanisms while simultaneously reducing the associated configuration burden.