基于大数据分析框架的在线信用评级关键技术研究

While Internet Finance can considerably boost the economy of China, it also causes various financial risks and practical problems. The proposed research project aims to develop a novel online credit scoring methodology and the related techniques for effectively assessing the credit risks of individual Internet Finance users, and hence contribute to mitigate the financial risks in general and credit risks in particular of Internet Finance services. With the rapid proliferation of the Social Web, it is likely for individuals to leave traces of credit risks (e.g., recurring debts) on online social networks. However, traditional credit scoring models fail to leverage these rich credit assessment features left behind on online social networks for enhancing credit scoring performance. The proposed research project is just able to fill such a research gap. The main challenge of the proposed research work is to develop novel computational methods that can effectively and efficiently mine latent credit assessment features from big social network data for supporting near real-time credit scoring on the Internet. To tackle such a challenge, we will first propose a novel big data analytics framework to facilitate online credit scoring. Under such a framework, our main research work includes: (1) the design and development of a parallel topic modeling based latent behavioral feature mining method for credit risk assessment; (2) the design and development of a dynamic and parallel community search method for mining community-based credit assessment features; (3) the design and development of a novel workload balancing scheme for big data analytics. The main contributions of the proposed research project are as follows: (1) from a theoretical perspective, we contribute to design a novel big data analytics framework to facilitate Internet Finance as a whole, and develop the corresponding methods and techniques to facilitate big social media analytics; (2) from a practical standpoint, we contribute to mitigate the financial risks of Internet Finance by prototyping a robust online credit scoring service for individuals.

互联网金融在极大的促进我国经济发展的同时也导致了各种风险。本课题针对个人信用风险评估，探讨在线信用评分关键技术，从而对互联网金融的风险控制作出贡献。随着社交网络的普及，个人用户常在社交网络中遗留与信用风险相关的特征。然而，传统的信用评分模型没有充分利用这些社交网络特征。本课题的研究可弥补现存信用评级模型的不足，其难点在于从社交网络大数据里快速高效地挖掘隐藏的信用风险特征，从而支持近实时信用评分。围绕该核心问题，课题将提出一个大数据分析框架，并在此基础上重点研究以下内容：（1） 基于并行主题模型的个人信用风险行为特征挖掘；（2）基于动态社区搜索的社区风险特征挖掘；（3）大数据分析平台上计算任务的均衡分配。课题研究意义在于：1）理论上，发展面向互联网金融的大数据分析框架，丰富和扩充社交网络大数据挖掘的理论与技术；2）应用中，课题的研究成果可直接为互联网金融参与者提供针对个人的在线信用评分服务。

互联网金融在极大的促进我国经济发展的同时也导致了各种风险。本项目针对个人信用风险评估，探讨在线信用评分关键技术，从而对互联网金融的风险控制作出贡献。随着社交网络的普及，个人用户常在社交网络中遗留与信用风险相关的特征。然而，传统的信用评分模型没有充分利用这些社交网络特征。本项目的研究弥补了现存信用评级模型的不足，其主要贡献在于从社交网络大数据里快速高效地挖掘隐藏的信用风险特征，从而支持近实时信用评分。围绕该核心问题，本项目已经开发了一个新颖的大数据分析框架，并在此基础上成功完成了以下重点内容：..（1） 开发了一个新颖的基于并行主题模型的个人信用风险行为特征挖掘模型和算法；..（2）开发了一个新颖的基于动态社区搜索的社区风险特征挖掘技术；..（3）开发了一个新颖的大数据分析平台上计算任务的均衡分配模型和算法。..项目的主要科学意义在于：1）从理论上，发展面向互联网金融的大数据分析框架，丰富和扩充社交网络大数据挖掘的理论与技术；2）关于实际应用方面，本项目的研.究成果已经直接为互联网金融参与者提供了一个新颖的基于社交网络大数据的在线信用评分服务系统。