Federal learning-based personal credit evaluation model training method and evaluation method

Abstract

The invention relates to a federal learning-based personal credit evaluation model training method and evaluation method. The method comprises the steps of obtaining heterogeneous data of multiple parties about a target user; performing feature processing on multi-party heterogeneous data; generating a heterogeneous information network diagram of each party according to the type and features of the data after feature processing; inputting the heterogeneous information network diagrams of all parties into a pre-trained personal credit evaluation model to obtain credit calculation results of all parties output by the personal credit evaluation model; and according to the credit calculation result of each party, based on a preset formula, performing comprehensive calculation on the personal credit evaluation value to obtain the personal credit evaluation value. According to the invention, the security feature of federal learning and the rich expression feature of the meta-path of the heterogeneous information network are utilized, unified expression of multi-party heterogeneous data is realized, various individual credit risks can be evaluated more directly and objectively, and the method has better effects in the aspects of prediction accuracy, robustness and the like of the individual credit risks.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a training method and an evaluation method for a personal credit evaluation model based on federated learning. Background technique [0002] With the development of the big data era, more and more attention is paid to data security, and regulations are constantly improving. Federated learning is an emerging technology based on machine learning, which has received extensive attention from all walks of life in recent years. The so-called federated learning refers to the joint training of machine learning models by multiple participants without exposing local data, and during the entire learning process, the client's data does not expose local data to other parties, which can ensure data privacy and Security, thus ensuring data privacy and security while solving the problem of data islands. Because federated learning technology can guarantee data privacy and security, it ...

AI Technical Summary

Problems solved by technology

[0007] (1) With the development of society, personal credit risk also presents the characteristics of diversification and multi-source. Many unstructured data or semi-structured data or information need to be introduced into the personal risk evaluation model, while the traditional personal credit evaluation The method can only calculate the quantitative risks of financial institutions, and cannot calculate many non-quantitative risks faced by financial institutions. Therefore, the traditional personal credit evaluation method is not suitable for the multi-dimensional characterization of credit counterparties in modern finance, and can only be calculated from a single Quantitative dimensions are described;

[0008] (2) When traditional personal credit evaluation needs to use multi-party data, it generally relies on standard API interfaces or XML files for data interaction. This method will cause leakage of private data during transmission or calculation, and cannot ensure the evaluation process. Data privacy and security, and with the soundness of the law, this data transfer method of directly interacting with personal data does not comply with legal regulations and requirements, and cannot meet the requirements of financial institutions for personal credit evaluation

[0009] (3) In traditional risk analysis, the scope of risk connotation is vaguely defined, and there is no uniform rule for risk judgment

Due to the different understanding of risk connotation and different standards for risk judgment, the analysis results will have a strong personal color, and the risk results judged by different analysts for the same subject matter will also show differences

Images