随着移动互联网技术的发展，各式各样的网络服务在极大方便人们日常生活（如在线社交、网络支付和线上游戏等）的同时，也滋生了新型的网络犯罪。基于身份盗用的网络欺诈给网络空间安全带来巨大威胁。身份认证是应对身份盗用最有效的手段。现有的主流身份认证方法多是基于访问控制的门禁式方法，是侵入式的，而且难以满足高并发、强实时等应用需求。基于用户行为的身份认证能够弥补这些缺陷，但现有的行为建模方法依赖于高质量的行为数据，且缺乏相应的模型性能分析理论。本项目拟采用“多角度分解”和“多方位合成”的建模思路，分别从行为特征空间划分、模型行为主体分类和认证模型功能等多角度研究高效行为建模方法，并在此基础上形成基于特征层与模型层、联合与融合方法相结合的智能合成策略，以填补低质量数据和高性能模型之间的鸿沟。与此同时，拟针对复杂的行为数据分布，研究面向身份认证估计的实用型极限分析理论，为构建高效行为模型提供理论基础。

With the rapid development of the mobile Internet technology, a variety of Internet services have greatly facilitated people’s daily lives (e.g. online social networking services, online payment, and online games), and at the same time brought new kinds of cyber crimes. Among all of them, Internet fraud, which relies on identity theft, greatly threatens cyberspace security. Identity authentication is the most effective means to battle identity theft. For this purpose, current implementations mainly depend on access control, which works in an intrusive way and often fails to meet the requirements of high concurrence and hard real-time systems. Although using user behaviors for identity authentication is capable of combating these defects, effective behavior models are actually hard to develop due to the low quality of behavior data and the lack of model performance analysis theory. In order to bridge the gap between low-quality behavior data and high-quality behavior models, our project will follow two stages of multi-perspective-decomposition and multi-aspect-synthesis. Firstly, we will take the perspectives of behavior feature space partition, model behavioral agent granularity and authentication model function category to design effective behavior models respectively. Afterwards, we will work out an intelligent synthesis strategy by an integrated utilization of feature-level-synthesis and model-level-synthesis in a fused or joint manner. Meanwhile, in the purpose of providing a theoretical fundamental basis for building high quality behavior models, we will focus on the complicated distribution of behavior data to study the fundamental limits of identity authentication for a practical evaluation theory.