Search

CN-121998031-A - Transfer learning method of credit assessment model and credit assessment method

CN121998031ACN 121998031 ACN121998031 ACN 121998031ACN-121998031-A

Abstract

The specification provides a transfer learning method and a credit assessment method of a credit assessment model. The method comprises the steps of obtaining sample data, mapping source domain data and target domain data with common characteristics to the same characteristic dimension, determining mapping relations between the source domain data and the target domain data and the characteristic dimension respectively, constructing a characteristic space for a newly added industry based on the determined mapping relations, inputting any sample data into a characteristic extraction network of a credit assessment model obtained by training the source domain data to determine credit characteristics of a training sample under the characteristic space through the characteristic extraction network, determining prediction domain classification corresponding to the sample data according to the credit characteristics, adjusting network parameters of the characteristic extraction network according to deviation between the prediction domain classification and actual domain classification of the training sample, and determining the credit assessment model after migration based on the adjusted characteristic extraction network.

Inventors

  • HU QUN

Assignees

  • 钱塘征信有限公司

Dates

Publication Date
20260508
Application Date
20260401

Claims (12)

  1. 1. A transfer learning method of a credit assessment model comprises the following steps: Acquiring sample data, wherein the sample data is credit investigation data, and comprises source domain data corresponding to the existing industry and target domain data corresponding to the newly added industry; Mapping source domain data and target domain data with common characteristics to the same characteristic dimension, determining the mapping relation between the source domain data and the target domain data and the characteristic dimension respectively, and constructing a characteristic space for the newly added industry based on the determined mapping relation; inputting any sample data into a feature extraction network of a credit assessment model trained by the source domain data so as to determine credit characteristics of the training sample under the feature space through the feature extraction network; According to the credit sign characteristics, determining the prediction domain classification corresponding to the sample data, adjusting network parameters of the characteristic extraction network according to the deviation between the prediction domain classification and the actual domain classification of the training sample, and determining a transferred credit sign evaluation model based on the adjusted characteristic extraction network, wherein the transferred credit sign evaluation model is applied to credit sign evaluation of the new industry.
  2. 2. The method of claim 1, wherein the adjusting the model parameters of the feature extraction network based on the deviation between the predicted domain classification and the actual domain classification of the training sample comprises: Inputting the target credit sign characteristics into a temporarily set domain discrimination network to obtain a predicted domain classification output by the domain discrimination network; determining a first loss value according to the deviation between the predicted domain classification and the actual domain classification of the training sample, wherein the first loss value is inversely related to the deviation; And adjusting model parameters of the feature extraction network by taking the minimized first loss value as an optimization target.
  3. 3. The method according to claim 2, wherein the adjusting the network parameters of the feature extraction network with the objective of minimizing the first loss value specifically comprises: and adjusting network parameters of the feature extraction network by taking the minimized first loss value as an optimization target, and adjusting network parameters of the domain discrimination network by taking the maximized first loss value as the optimization target.
  4. 4. The method of claim 1, wherein the credit assessment model comprises a credit assessment network; The method further comprises the steps of: determining a sample label corresponding to the target domain data, wherein the sample label comprises an actual credit assessment result aiming at the target domain data; inputting the target domain data into the credit assessment network to obtain a predicted credit assessment result output by the credit assessment network; Determining a second loss value according to the deviation between the sample label and the prediction credit assessment result, wherein the second loss value is positively correlated with the deviation; and adjusting network parameters of the credit assessment network by taking the minimized second loss value as an optimization target.
  5. 5. The method of claim 1, wherein the credit assessment model comprises a credit assessment network: The method further comprises the steps of: Inputting the data of each target domain into the credit assessment model to obtain each credit assessment result and each corresponding confidence coefficient output by the credit assessment network; determining a target credit assessment result with the confidence coefficient higher than a preset value from the credit assessment results, and constructing a subsampled according to the target credit assessment result and corresponding target domain data thereof; inputting the subsamples into the credit assessment network to obtain a predicted credit assessment result output by the credit assessment network; Determining a third loss value according to the deviation between the target credit assessment result and the predicted credit assessment result of the corresponding target domain data, wherein the third loss value is positively correlated with the deviation; And adjusting the network parameters of the credit assessment network by taking the minimized third loss value as an optimization target.
  6. 6. The method of claim 1, prior to constructing a feature space for the newly added industry based on the mapping relationship, the method further comprising: Determining industry configuration data of the existing industry closest to the industry scene information of the newly added industry in a preset industry configuration library according to the industry scene information of the newly added industry, wherein the industry configuration information comprises characteristic space configuration of the existing industry and model data of a credit assessment model; constructing a feature space for the newly added industry based on the mapping relation, specifically comprising: Based on the mapping relation, adjusting the feature space configuration information of the credit assessment model corresponding to the existing industry to construct a feature space for the newly added industry; After adjusting the model parameters of the credit assessment model, the method further comprises: And storing the model data of the adjusted credit assessment model and the characteristic space configuration thereof as industry configuration data of the newly added industry in the industry configuration library.
  7. 7. The method according to claim 1, mapping source domain data and target domain data with common characteristics to the same characteristic dimension, specifically comprising: determining a plurality of credit assessment dimensions affecting credit assessment results; Determining, for each sample data, a degree of association between the sample data and each credit assessment dimension; and determining source domain data and target domain data with commonality according to the association degree between each sample data and each credit assessment dimension, and shooting the source domain data and the target domain data to the same characteristic dimension.
  8. 8. A credit assessment method, comprising: Acquiring credit investigation data of a target industry; Inputting the credit information data into a pre-trained credit information evaluation model to obtain a credit information evaluation result output by the credit information evaluation model; wherein the credit assessment model is trained by the method of any one of claims 1-7.
  9. 9. A training device of a credit assessment model, comprising: The system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring sample data, wherein the sample data is credit investigation data and comprises source domain data corresponding to the existing industry and target domain data corresponding to the newly added industry; The construction module is used for mapping the source domain data and the target domain data with the common characteristics to the same characteristic dimension, determining the mapping relation between the source domain data and the target domain data and the characteristic dimension respectively, and constructing a characteristic space for the newly added industry based on the determined mapping relation; The input module is used for inputting any sample data into a feature extraction network of a credit assessment model obtained by training the source domain data so as to determine the credit characteristics of the training sample under the feature space through the feature extraction network; The migration module is used for determining the prediction field classification corresponding to the sample data according to the credit investigation characteristics, adjusting the network parameters of the characteristic extraction network according to the deviation between the prediction field classification and the actual field classification of the training sample so as to determine a migrated credit investigation evaluation model based on the adjusted characteristic extraction network, wherein the migrated credit investigation evaluation model is applied to credit investigation evaluation of the new industry.
  10. 10. An electronic device comprising a processor, a memory for storing processor-executable instructions, wherein the processor is configured to implement the steps of the method of any one of claims 1-8 by executing the executable instructions.
  11. 11. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any of claims 1-8.
  12. 12. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of any of claims 1-8.

Description

Transfer learning method of credit assessment model and credit assessment method Technical Field One or more embodiments of the present disclosure relate to the field of computer technologies, and in particular, to a method for learning and evaluating credit by migration of a credit evaluation model. Background When a new industry appears, a credit evaluation model needs to be reconstructed aiming at the industry, and due to the fact that the new industry generally lacks a mature modeling foundation and lacks an explicit credit label in the traditional credit industry, effective training of the model is difficult to achieve, and moreover, the mode of modeling and manually processing related data features is repeated, so that overall evaluation efficiency is further low, and the requirements of the new industry are difficult to adapt. Disclosure of Invention In view of this, one or more embodiments of the present disclosure provide the following technical solutions: According to a first aspect of one or more embodiments of the present disclosure, a method for learning to migrate a credit assessment model is provided, including: Acquiring sample data, wherein the sample data is credit investigation data, and comprises source domain data corresponding to the existing industry and target domain data corresponding to the newly added industry; Mapping source domain data and target domain data with common characteristics to the same characteristic dimension, determining the mapping relation between the source domain data and the target domain data and the characteristic dimension respectively, and constructing a characteristic space for the newly added industry based on the determined mapping relation; inputting any sample data into a feature extraction network of a credit assessment model trained by the source domain data so as to determine credit characteristics of the training sample under the feature space through the feature extraction network; According to the credit sign characteristics, determining the prediction domain classification corresponding to the sample data, adjusting network parameters of the characteristic extraction network according to the deviation between the prediction domain classification and the actual domain classification of the training sample, and determining a transferred credit sign evaluation model based on the adjusted characteristic extraction network, wherein the transferred credit sign evaluation model is applied to credit sign evaluation of the new industry. According to a second aspect of one or more embodiments of the present disclosure, there is provided a credit assessment method, including: Acquiring credit investigation data of a target industry; Inputting the credit information data into a pre-trained credit information evaluation model to obtain a credit information evaluation result output by the credit information evaluation model; The credit assessment model is obtained by training the transfer learning method of the credit assessment model. According to a third aspect of one or more embodiments of the present specification, there is provided a training apparatus of a credit assessment model, comprising: The system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring sample data, wherein the sample data is credit investigation data and comprises source domain data corresponding to the existing industry and target domain data corresponding to the newly added industry; The construction module is used for mapping the source domain data and the target domain data with the common characteristics to the same characteristic dimension, determining the mapping relation between the source domain data and the target domain data and the characteristic dimension respectively, and constructing a characteristic space for the newly added industry based on the determined mapping relation; The input module is used for inputting any sample data into a feature extraction network of a credit assessment model obtained by training the source domain data so as to determine the credit characteristics of the training sample under the feature space through the feature extraction network; The migration module is used for determining the prediction field classification corresponding to the sample data according to the credit investigation characteristics, adjusting the network parameters of the characteristic extraction network according to the deviation between the prediction field classification and the actual field classification of the training sample so as to determine a migrated credit investigation evaluation model based on the adjusted characteristic extraction network, wherein the migrated credit investigation evaluation model is applied to credit investigation evaluation of the new industry. According to a fourth aspect of one or more embodiments of the present specification, there is provided a credit ass