基于社交多媒体行为的物理世界用户属性研究

Web user profiling and understanding plays an important role in digitalized information management and customized information services. It is desirable to exploit social multimedia for user profiling because of the adequate user data, rich multimedia sensory simulation and low privacy invasion. Current user profiling research mainly focus on the online perspective, while little work has studied the topic of physical user attribute modeling. The challenges lie in lackage of ground-truth, noisy and multi-modal user data, and complex correlation between user physical attribute and online data. This project propose s to exploit the users' social multimedia behaviors, to study and infer their physical attributes like demographics, personality and physical social networks. The research outcome will provide the fundamental theories and key technique support for physical user attribute modeling, and promote the applications in social media-based multimedia information awareness, service and regulatory.

理解互联网用户属性在数字化信息管理和个性化信息服务方面具有重大的社会和商业价值。基于社交多媒体行为进行用户属性研究具有用户行为数据充足、多媒体信息丰富和对用户隐私侵犯小等优势。目前相关研究主要面向用户的网络空间属性，基于社交多媒体行为的物理世界用户属性研究尚处于起步阶段。针对物理世界用户属性研究中真值数据缺乏、用户行为的噪声多模态、物理世界属性与网络行为关联复杂等挑战，本项目提出基于用户在社交网络和多媒体分享网站的行为，分析和推断人口属性、性格属性、真实社会关系属性等物理世界的用户属性。研究成果将为物理世界用户属性建模提供理论和技术基础，推动基于社交多媒体的信息感知、服务和监管等应用的发展。