该项目聚焦大数据背景下小微企业信用风险评估，针对小微企业财务数据不健全、关联风险较隐蔽、信用评估模型不完善等痛点，提出“复杂网络-特征挖掘-半监督学习-集成预测”的一体化研究思路。项目深入分析小微企业与其他微观主体之间的交互行为和关联方式，通过构建小微企业关联风险复杂网络，对小微企业关联风险特征进行深度挖掘，力图从数据特征层面弱化借贷双方信息不对称程度。将模型算法与业务场景深度融合，突破传统监督学习的局限，构建拒绝推断和代价敏感相结合的半监督模型，兼顾分类精度与误分代价。为应对复杂网络视角下小微企业“大样本”和“高维稀疏特征”，提出自适应随机子空间学习的集成评估模型，进一步提升模型的稳定性和泛化性。该项目将理论研究与实证分析有机结合，为缓解小微企业融资约束、强化金融机构风险管控、防范信用风险、甄别信用欺诈提供科学决策依据，也为监管部门制定相关政策提供决策参考，具有重要的理论意义和应用价值。

The project will focus on the credit risk assessment of the small and micro enterprises under the background of big data. The innovative and systemic research idea “complex network-feature mining-semi-supervised learning-ensemble prediction” is proposed，with consideration of problems of incomplete financial data, hard-to-find correlated risk, and imperfect credit evaluation models in terms of the small and micro enterprises. This project deeply analyzes the interaction behavior and the association way between small and micro enterprises and other subjects. By building a complex network of small and micro enterprises' associated risks, this study explores the characteristics of small and micro enterprises' associated risks and weakens the degree of information asymmetry between small and micro enterprises and financial institutions. Combining the model with the financial business scenario, a semi-supervised evaluation model combining reject inference and cost sensitivity is proposed, which takes into account the classification accuracy and misclassification cost. From the perspective of complex networks, small and micro enterprises have the characteristics of "large sample" and "high-dimensional sparse" . The evaluation model of adaptive random subspace learning is proposed to further improve the stability and generalization of model evaluation, making it more suitable for small and micro enterprise credit business scenarios. With the organic combination of the theoretical research and empirical analysis, the project will provide the scientific basis in decision making for commercial banks and other financial institutions in strengthening loan risk control, credit risks prevention and identifying credit fraud etc. which also have significant practical application value.