Credit risk assessment method based on Transformer

Abstract

The invention discloses a credit risk assessment method based on Transformer, which comprises a Transformer network, the Transformer network is divided into three parts, i.e., an Embedding part, an Encoder part and a classifier part, the behavior log input Embedding is composed of three parts: (1) current operation of a user, (2) the time of the current operation of the user, and (3) the credit granting and bill state during user operation, wherein the Encoder part is formed by stacking six Encoder Blocks, meanwhile, the Encoder Block splices the input and the result of the previous layer together at the same time to serve as the input of the current layer; and the classifier is used for finally outputting whether the user is a black sample. According to the invention, information in a complex data structure can be better extracted based on a Transformer network, and the model performance is better; and meanwhile, a multi-spatial-dimension feature simultaneous training mode is provided.

Description

technical field [0001] The invention relates to the field of Internet financial credit assessment, in particular to a Transformer-based credit risk assessment method. Background technique [0002] Personal credit risk assessment refers to the use of credit scoring models by credit assessment agencies to quantitatively analyze consumers' personal credit information, and the results show the level of personal credit in the form of scores. The key to improving the effect of the credit scoring model is to extract more credit-related information from more data dimensions and represent this information in a way suitable for model learning. [0003] At present, the most mainstream is the scorecard model, which mainly uses logistic regression. It has the advantages of simplicity, stability, strong interpretability, and regulatory approval. In addition, there are also integrated learning models such as XGBoost and LightGBM in the industry. Compared with the scorecard, this type of...

AI Technical Summary

Problems solved by technology

Compared with the scorecard, this kind of integrated learning model has better model performance, but has the disadvantages of complex model, poor interpretability, and possibly poor stability.

[0004] However, in the face of complex data structures, such as each application’s own user behavior data, the above models need to construct a large number of artificial features, and may only achieve suboptimal solutions without domain experts.

Images