CN-122022987-A - Method, device, equipment and medium for processing credit asset expected credit loss

CN122022987ACN 122022987 ACN122022987 ACN 122022987ACN-122022987-A

Abstract

The application relates to a processing method, a device, equipment and a medium for expected credit loss of a credit asset, wherein the processing method for the expected credit loss of the credit asset comprises the steps of obtaining multi-modal data of a target user on the aspect of the credit asset, inputting the multi-modal data into a preset expected credit loss prediction model, carrying out fusion processing and prediction on the multi-modal data by utilizing the expected credit loss prediction model to obtain a prediction result, determining an expected credit loss measurement result based on the prediction result, and carrying out accounting processing based on the expected credit loss measurement result to obtain a target audit trail record corresponding to the target user. The application can improve the accuracy of risk prediction, thereby improving the accuracy of credit loss meter.

Inventors

LIU WEI

Assignees

上海信小飞数字科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260126
Priority Date: 20260108

Claims (10)

1. A method of processing a credit asset expected credit loss, the method comprising: Acquiring multi-modal data of a target user in terms of credit assets, wherein the multi-modal data comprises structured data, time sequence data, text data and graph data; inputting the multi-mode data into a preset expected credit loss prediction model; carrying out fusion processing and prediction on the multi-mode data by utilizing the expected credit loss prediction model to obtain a prediction result, wherein the prediction result comprises a default probability, a default loss rate and a default risk value; determining an expected credit loss measurement based on the prediction; And accounting processing is carried out based on the expected credit loss measurement result, and a target audit trail record corresponding to the target user is obtained.
2. The method of claim 1, wherein the expected credit loss prediction model comprises a feature extraction layer, a feature fusion layer, and a prediction layer; The fusion processing and prediction of the multi-modal data by using the expected credit loss prediction model comprises the following steps: preprocessing the multi-modal data to obtain preprocessed multi-modal data; inputting the preprocessed multi-modal data into the feature extraction layer; extracting features from the preprocessed multi-modal data by using the feature extraction layer to obtain multi-modal features; carrying out multi-mode coding processing on the multi-mode characteristics to obtain multi-mode characteristic coding vectors; inputting the multi-mode feature coding vector into a feature fusion layer for fusion treatment to obtain fusion features; And inputting the fusion characteristics into the prediction layer to obtain a prediction result.
3. The method according to claim 2, wherein the multi-modal features include basic features, statistical features, temporal features, text features, and cross features, the feature extraction from the preprocessed multi-modal data using the feature extraction layer, and the method comprises: Extracting basic features from the structured data; extracting statistical features and time features from the time sequence data; Extracting text features from the text data; cross features are extracted from the map data.
4. The method according to claim 2, wherein the step of inputting the multi-modal feature encoding vector into a feature fusion layer to perform fusion processing to obtain a fusion feature includes: Inputting the multi-mode feature coding vector into a feature fusion layer, calculating the attention weights among the features in the same mode by using self-attention, calculating the cross attention weights among different modes by using cross attention, and acquiring the feature relations of different layers by using multi-head attention; and carrying out fusion processing on the multi-mode feature coding vector based on the attention weights among the features in the same mode, the cross attention weights among different modes and the feature relation to obtain fusion features.
5. The method of claim 1, wherein the accounting based on the expected credit loss measurement comprises: Performing accounting entry template matching on the expected credit loss measurement result to obtain a matching result; performing entry parameter automatic filling based on the matching result to obtain a filling result; verifying the filling result by adopting preset accounting rules; And if the verification is passed, performing automatic posting processing, updating the decrement value and the financial statement to obtain a target audit trail record corresponding to the target user.
6. The method of claim 5, wherein prior to accounting entry template matching the expected credit loss measurement, the method further comprises: Carrying out credit risk stage identification on the expected credit loss measurement result; If the expected credit loss measurement result is identified to belong to a low risk stage, accounting and lifting processing is carried out according to the expected credit loss of 12 months; If the expected credit loss measurement result is identified to belong to the medium risk stage, accounting and lifting processing is carried out according to the life expected credit loss; if the expected credit loss measurement is identified as belonging to a high risk stage, accounting is carried out according to the life expected credit loss and interest is suspended.
7. The method of claim 5, wherein after the determining the expected credit loss measurement, the method further comprises: determining macroscopic scene information based on a preset economic index and scene probability distribution value; Determining prospective adjustment parameters based on preset adjustment factors, preset adjustment amplitude and preset adjustment rules, wherein the preset adjustment factors comprise macroscopic factors, industry factors, regional factors, product factors and time factors; And dynamically adjusting the expected credit loss metering result by adopting a preset dynamic updating mechanism based on the macroscopic scene information and the prospective adjusting parameter.
8. A processing device for expected credit loss of a credit asset, characterized by comprising means for performing the method of any of claims 1-7.
9. A computer device, characterized in that it comprises a memory on which a computer program is stored and a processor which, when executing the computer program, implements the method according to any of claims 1-7.
10. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method according to any of claims 1-7.

Description

Method, device, equipment and medium for processing credit asset expected credit loss Technical Field The application relates to the technical fields of financial science and technology and intelligent risk management, in particular to a method, a device, equipment and a medium for processing expected credit loss of credit assets. Background Currently, financial institutions rely primarily on traditional ETL (Transform, load) based data integration schemes when implementing the international financial reporting guidelines No. 9 (IFRS 9) for Expected Credit Loss (ECL) metering and accounting processes. The traditional ETL data integration scheme mainly performs ECL metering by periodically extracting, converting and loading structured data of each system through ETL tools such as DATASTAGE, INFORMATICA. However, the conventional ETL tool mainly performs data conversion operations, such as field mapping, type conversion, aggregation calculation, and the like, depending on preset rules, and cannot perform deep semantic analysis on data content, so that key risk information is ignored, and since the conventional technology generally only processes structured data, risk feature extraction is insufficient, and finally, accuracy of credit loss meter extraction is low. Disclosure of Invention The embodiment of the application provides a method, a device, equipment and a medium for processing expected credit loss of credit assets, which aim to solve the problem that the accuracy of credit loss metering is lower due to the fact that structured data of all systems are regularly extracted, converted and loaded by an ETL tool for ECL metering. In a first aspect, an embodiment of the present application provides a method for processing expected credit loss of a credit asset, where the method for processing expected credit loss of the credit asset includes: Acquiring multi-modal data of a target user in terms of credit assets, wherein the multi-modal data comprises structured data, time sequence data, text data and graph data; inputting the multi-mode data into a preset expected credit loss prediction model; carrying out fusion processing and prediction on the multi-mode data by utilizing the expected credit loss prediction model to obtain a prediction result, wherein the prediction result comprises a default probability, a default loss rate and a default risk value; determining an expected credit loss measurement based on the prediction; And accounting processing is carried out based on the expected credit loss measurement result, and a target audit trail record corresponding to the target user is obtained. The expected credit loss prediction model comprises a feature extraction layer, a feature fusion layer and a prediction layer; The fusion processing and prediction of the multi-modal data by using the expected credit loss prediction model comprises the following steps: preprocessing the multi-modal data to obtain preprocessed multi-modal data; inputting the preprocessed multi-modal data into the feature extraction layer; extracting features from the preprocessed multi-modal data by using the feature extraction layer to obtain multi-modal features; carrying out multi-mode coding processing on the multi-mode characteristics to obtain multi-mode characteristic coding vectors; inputting the multi-mode feature coding vector into a feature fusion layer for fusion treatment to obtain fusion features; And inputting the fusion characteristics into the prediction layer to obtain a prediction result. The method further comprises the steps that the multi-modal characteristics comprise basic characteristics, statistical characteristics, time characteristics, text characteristics and cross characteristics, the characteristics are extracted from the preprocessed multi-modal data by the characteristic extraction layer, and the multi-modal characteristics are obtained, and comprise the following steps: Extracting basic features from the structured data; extracting statistical features and time features from the time sequence data; Extracting text features from the text data; cross features are extracted from the map data. The further technical scheme is that the inputting the multi-mode feature coding vector into the feature fusion layer for fusion processing to obtain fusion features comprises the following steps: Inputting the multi-mode feature coding vector into a feature fusion layer, calculating the attention weights among the features in the same mode by using self-attention, calculating the cross attention weights among different modes by using cross attention, and acquiring the feature relations of different layers by using multi-head attention; and carrying out fusion processing on the multi-mode feature coding vector based on the attention weights among the features in the same mode, the cross attention weights among different modes and the feature relation to obtain fusion features. The further technical scheme is that the