面向复杂数据的互联网金融征信建模技术研究

As the rapid development of the national economy and the Internet business, there appear a large variety of Internet financial modes and an increasing demand for the credit investigation into Internet finance entities. It is fairly important to improve the national financial supervision system and prevent systematic financial risks by assessing the credit of Internet financial entities and improving the credit investigation system. The system has faced several problems such as data heterogeneity, data multi-source, high data dimensionality, data sparsity, massive data missing, interaction complexity, and model inefficiency. We aim to study the following four problems: 1. avoiding the negative effect of massive missing data, 2. conducting the feature analysis on multi-source heterogeneous high-dimensional sparse data, 3. exploring the feature and interaction learning, and 4. improving the evaluation effect of the credit investigation models with the help of deep learning models. This project will be expected to provide both engineering and theoretical breakthroughs.

随着国家经济的不断发展和互联网业务的蓬勃发展，种类繁多的互联网金融模式大量涌现，对互联网金融主体征信评估的需求显著增长。构建和完善互联网金融主体征信体系，是健全金融监管体系，预防系统性金融风险的重要措施。本项目着力解决互联网金融征信建模中面临的诸多问题：1. 缺失数据较多； 2. 互联网金融数据异构多源高维稀疏特性明显，迥异于传统金融数据的特性；3. 数据特征间相关性和交互性复杂；4. 征信评估模型效果差。更进一步地，针对各研究内容，本项目分别提炼出相应的科学问题：1. 恢复缺失元素的最优化数学模型；2. 多源异构高维稀疏数据的特性分析；3. 特征学习算法设计等科学问题；4. 设计更有效的基于深度神经网络的互联网金融征信评估模型。本项目将在工程技术和学术理论两方面分别突破现有技术，实现创新。

随着国家经济的不断发展和互联网业务的蓬勃发展，种类繁多的互联网金融模式大量涌现，对互联网金融主体征信评估的需求显著增长。构建和完善互联网金融主体征信体系，是健全金融监管体系，预防系统性金融风险的重要措施。本项目着力解决互联网金融征信建模中面临的诸多问题：1. 缺失数据较多； 2. 互联网金融数据异构多源高维稀疏特性明显，迥异于传统金融数据的特性；3. 数据特征间相关性和交互性复杂；4. 征信评估模型效果差。针对上述问题，本项目取得以下方面成果：1. 数据不平衡下的金融主体征信研究；2. 多源异构高维稀疏数据的特性分析；3. 设计更有效的基于深度神经网络的互联网金融征信评估模型。