EAGER: TWC: Collaborative: iPrivacy: Automatic Recommendation of Personalized Privacy Settings for Image Sharing

这个项目的目标是研究一个全面的图像隐私推荐系统，称为iPrivacy(图像隐私)，它可以高效地自动为新分享的照片生成适当的隐私设置，并考虑到同一照片中出现的多方共识。随着智能手机和其他移动设备的日益普及，照片分享已经变得非常流行。然而，许多人，尤其是社交网络的年轻用户，经常分享自己和朋友的私人照片，而没有意识到不想要的披露和侵犯隐私行为对他们未来生活的潜在影响。虽然一些照片分享平台开始提供隐私配置功能，但这种手动过程对用户来说可能非常繁琐，而且容易出错，因为没有多少用户对隐私有足够的背景知识。该项目将解决社交网站中照片分享日益加剧的隐私问题，并使数十亿社交网络用户受益。该项目的更广泛影响将通过教育和研究的整合而进一步增强。将开展一系列教育活动，包括课程开发、学生专业培训和K-12教师网络安全营，重点是代表不足的群体。该项目将无缝地整合图像理解和隐私管理两个不同领域的专业知识，从而形成首批全面和自动的政策建议系统之一。本项目包含以下创新性研究内容。首先，将开发一种多方隐私敏感对象识别算法，该算法将能够自动生成照片中每个人类主体的身份，从而自动执行后续的隐私协调过程。其次，将设计一种独特的隐私协调方法，该方法将进行分层隐私策略挖掘，以了解社区中不同级别的隐私关切，推荐有效协调同一照片中多个人的隐私偏好的策略，并适应人们隐私偏好的演变。建议的iPrivacy系统不仅将充分释放用户侧的隐私配置负担，还将促进基于从大规模历史和社会信息中学习的知识的更好的隐私实践。

The objective of this project is to investigate a comprehensive image privacy recommendation system, called iPrivacy (image Privacy), which can efficiently and automatically generate proper privacy settings for newly shared photos that also considers consensus of multiple parties appearing in the same photo. Photo sharing has become very popular with the growing ubiquity of smartphones and other mobile devices. However, many people especially young users of social networks often share private photos about themselves and their friends without being aware of the potential impact on their future lives caused by unwanted disclosure and privacy violations. Although some photo sharing platforms start to offering functions of privacy configuration, such manual process could be very tedious for users and also error-prone since not many users have sufficient background knowledge about privacy. This project will address these rising privacy concerns of photo sharing in social sites and benefit billions of social network users. The broader impact of this project will be further enhanced by the integration of education and research. A range of educational activities will be carried out including curriculum development, professional training for students and cybersecurity camp for K-12 teachers, with emphasis to under-represented groups.This project will seamlessly integrate expertise from two different domains: image understanding and privacy management, leading to one of the first comprehensive and automatic policy recommendation systems. The proposed project contains the following innovative researches. First, a multi-party privacy-sensitive object identification algorithm will be developed which will be capable of automatically generating the identity of each human subject in a photo so as to automate the subsequent privacy harmonization process. Second, a unique privacy harmonization approach will be designed, which will conduct hierarchical privacy policy mining to understand different levels of privacy concerns in communities, recommend policies that effectively harmonize privacy preferences of multiple people appearing in the same photo and also adapt to the evolution of people's privacy preferences. The proposed iPrivacy system will not only fully release the burden of privacy configuration at users' side, but will also promote better privacy practice based on knowledge learned from large-scale historical and societal information.