Parameter adjustment method and device of credit risk model, equipment and storage medium
By dynamically adjusting the configuration parameters of the credit risk model using a multi-granularity hybrid expert model, the problem of insufficient flexibility in traditional credit risk assessment methods is solved, enabling more efficient credit risk identification and decision-making.
Patent Information
- Application Number
- CN202510999478.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-11-04
AI Technical Summary
Traditional credit risk assessment methods lack flexibility in the internet finance environment, leading to inaccurate model performance evaluation, increasing the risk of misjudgment and omission, and affecting decision-making effectiveness.
By dynamically adjusting the configuration parameters of the credit risk model using a multi-granularity hybrid expert model, the diversity and flexibility of the parameters are improved, and the optimal configuration parameters are determined using a pre-trained multi-granularity hybrid expert model.
It improves the performance and identification capabilities of the credit risk model, reduces the risk of misjudgment and omission, and enhances the accuracy and decision-making efficiency of the model.
Smart Images
Figure CN120893503A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computer and the technical field of finance, and in particular, relates to a parameter adjustment method and device of a credit risk model, equipment and a storage medium. BACKGROUND
[0002] With the rapid development of financial technology, the field of credit risk control is facing unprecedented challenges. In particular, in the Internet financial environment, credit institutions need to respond quickly to changing customer behavior and market dynamics. Traditional credit risk assessment methods have been difficult to adapt to increasingly complex risk situations, especially when facing different credit models, how to effectively compare their performance has become a key problem.
[0003] In related technologies, the performance of credit risk models can be compared through a champion challenge. The idea of the champion challenge is to define the model with the best current performance as the champion group and the other models as the challenge group. During the model iteration and data source update process, the performance between the champion group and the challenge group is constantly compared, and the latest optimal model is set as the champion group. Parameter configuration is performed during the champion challenge process to more effectively compare models.
[0004] However, in the existing champion challenge process, some configuration parameters, such as traffic configuration parameters and threshold setting parameters, are usually configured in a static manner. This configuration method lacks flexibility, affecting the accuracy of credit risk model performance evaluation in the champion challenge, and also causing resource waste. In addition, this implementation cannot obtain a credit risk model with optimal credit risk identification ability, thereby increasing the risk of misjudgment and omission of the credit risk model, and thus affecting the decision-making effect of the credit institution. SUMMARY
[0005] The credit risk model parameter adjustment method, device, equipment and storage medium provided by the embodiments of the present application are used to achieve the effect of improving the accuracy of the credit risk control model by flexibly adjusting the model parameters, and thus making the credit risk model in the champion challenge provide more accurate credit risk identification results.
[0006] In a first aspect, the embodiments of the present application provide a credit risk model parameter adjustment method, comprising:
[0007] obtaining a plurality of initial configuration parameters determined by a credit risk model in an initial champion challenge; wherein the initial configuration parameters indicate the configuration parameters applied by the credit risk model in the initial champion challenge;
[0008] After the plurality of initial configuration parameters are divided into a plurality of groups, the plurality of initial configuration parameters are input into a pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model;
[0009] The configuration parameters of the credit risk model are updated according to the new configuration parameters to obtain a new credit risk model, and a new champion challenge is performed according to the new credit risk model.
[0010] In a second aspect, an embodiment of the present application provides a parameter adjustment device of a credit risk model, including:
[0011] An obtaining unit is configured to obtain a plurality of initial configuration parameters determined by a credit risk model in an initial champion challenge; wherein the initial configuration parameters indicate configuration parameters applied by the credit risk model in the initial champion challenge;
[0012] A processing unit is configured to divide the plurality of initial configuration parameters into a plurality of groups, and then input the plurality of initial configuration parameters into a pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model;
[0013] An updating unit is configured to update the configuration parameters of the credit risk model according to the new configuration parameters to obtain a new credit risk model, and perform a new champion challenge according to the new credit risk model.
[0014] In a third aspect, an embodiment of the present application provides a computer device, including a memory and a processor.
[0015] The memory stores computer execution instructions.
[0016] The processor executes the computer execution instructions stored in the memory, so that the processor executes the embodiments of the first aspect.
[0017] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer execution instructions, and the computer execution instructions are executed by a processor to implement the embodiments of the first aspect.
[0018] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the embodiments of the first aspect.
[0019] The parameter adjustment method, apparatus, device, and storage medium for the credit risk model provided in this application can acquire multiple initial configuration parameters determined by the credit risk model in the initial champion challenge. This improves the diversity and flexibility of the initial configuration parameters, avoiding the limitations and inaccuracies of obtaining initial configuration parameters solely through static methods. Subsequently, the multiple initial configuration parameters can be divided into multiple groups and input into a pre-trained multi-granularity hybrid expert model for processing, resulting in new configuration parameters corresponding to the credit risk model. This implementation method can determine the new configuration parameters corresponding to the credit risk model through a pre-trained multi-granularity hybrid expert model, thereby improving the accuracy of the credit risk model's configuration parameters. The configuration parameters of the credit risk model can then be updated based on the new parameters, resulting in a new credit risk model with more accurate configuration parameters and thus a higher-performing credit risk model, thereby improving the performance of the credit risk model. Subsequently, when conducting a new champion challenge based on the new credit risk model, not only is the speed improved, but the ability of the credit risk model to identify credit risks is also enhanced. Attached Figure Description
[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0021] Figure 1 A flowchart illustrating a parameter adjustment method for a credit risk model provided in this application embodiment;
[0022] Figure 2 A schematic diagram illustrating the implementation process of a parameter adjustment method for a credit risk model provided in this application embodiment;
[0023] Figure 3 A flowchart illustrating another parameter adjustment method for a credit risk model provided in this application embodiment;
[0024] Figure 4 A schematic diagram illustrating the implementation process of another parameter adjustment method for a credit risk model provided in this application embodiment;
[0025] Figure 5 A schematic diagram of the structure of a parameter adjustment device for a credit risk model provided in an embodiment of this application;
[0026] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application.
[0027] The specific embodiments of the application have been shown by way of example in the above figures, and will be described in more detail hereafter. These figures and this written description are not intended to limit the scope of the inventive concept in any way, but rather to illustrate the inventive concept to one of ordinary skill in the art by reference to specific embodiments. DETAILED DESCRIPTION
[0028] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to any embodiment of the application, unless specified otherwise. Accordingly, when the description of the exemplary embodiments contains language that can imply limitations on the scope of the application, such limitations are not intended to apply to any specific embodiment unless specifically stated otherwise. For example, the words "comprising," "including," "containing," and "having" are intended to be open and permit the inclusion of any additional feature, step, or component without excluding any other features, steps, or components. Moreover, the use of the term "about" is intended to allow for variations by including "about" a stated value or range, for example, to account for experimental error, variations in manufacture, differences in measurements, and the like.
[0029] First, the terms related to the present application are explained:
[0030] Credit risk model: A credit risk model is a statistical model or algorithm used to assess and manage lending risk, aiming to help financial institutions (such as banks, credit companies, etc.) decide whether to approve loan applications, set loan limits, set interest rates, and monitor the creditworthiness of borrowers, etc.
[0031] Champion challenge: A competitive testing method used to evaluate and optimize models or strategies, widely applied in decision-making systems in finance, marketing, and other industries; its core idea is to compare the existing best model (champion) with the newly proposed model (challenger) to determine which model performs better in a specific task;
[0032] Mixture of experts: Mixture of experts, MoE, is a regulatory mechanism for building a system with multiple expert networks; each expert network is responsible for processing a different subset of training samples and focuses on a specific area of the input space; a gating network is used to select which expert network to process a specific input;
[0033] Multi-granularity: Multi-granularity data refers to data obtained by dividing different granularities.
[0034] With the rapid development of financial technology, the field of credit risk management is facing unprecedented challenges. In particular, in the Internet financial environment, credit institutions need to respond quickly to changing customer behavior and market dynamics. Traditional credit risk assessment methods have been difficult to adapt to the increasingly complex risk situation, especially when facing different credit models, how to effectively compare their performance has become a key problem.
[0035] In the related art, the credit risk model performance can be compared through the champion challenge. The idea of the champion challenge is to define the model with the best current performance as the champion group and define other models as the challenge group. In the model iteration and data source updating process, the performance between the champion group and the challenge group is compared, the latest optimal model is set as the champion group, and the parameter configuration is performed in the process of the champion challenge to more effectively compare the models.
[0036] However, in the existing champion challenge process, some configuration parameters, such as traffic configuration parameters and threshold setting parameters, are usually configured in a static manner. This configuration method relies on fixed parameters and rules and lacks flexibility, making it difficult to adapt to the rapid changes in the types of credit risk models and the configuration parameters of the credit risk models. This results in a certain lag between the set configuration parameters and the actual situation, which affects the accuracy of the credit risk model performance evaluation in the champion challenge, and also causes resource waste. In addition, this implementation cannot obtain the credit risk model with the optimal credit risk identification ability, thereby increasing the risk of misjudgment and omission of the credit risk model, and thus affecting the decision-making effect of the credit institution.
[0037] The present application provides a credit risk model parameter adjustment method, which dynamically adjusts the basic parameter information of the credit risk model through a multi-granularity mixed expert model, thereby promoting the comparison of the credit risk model under the same conditions, promoting the benign competition between models, and improving the performance of the credit risk model.
[0038] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.
[0039] Figure 1 A flowchart of a credit risk model parameter adjustment method provided by an embodiment of the present application is shown in FIG. 1, and the method comprises the following steps. Figure 1
[0040] S101, obtaining a plurality of initial configuration parameters determined by a credit risk model in an initial champion challenge.
[0041] The initial configuration parameters indicate the configuration parameters applied by the credit risk model in the initial champion challenge.
[0042] In one example, the initial configuration parameters can be understood as configuration parameters associated with the credit risk identification capability of the credit risk model. For example, the initial configuration parameters can include, but are not limited to, a plurality of configuration parameters of the traffic type and a plurality of configuration parameters of the threshold type applied by the credit risk model in the initial champion challenge.
[0043] For example, the initial configuration parameters can include, but are not limited to, business tracking variables, positive and negative sample division methods, traffic allocation methods, model scenarios, customer types, model update frequencies, and the like.
[0044] S102, after the plurality of initial configuration parameters are divided into a plurality of groups, the plurality of initial configuration parameters are input into the pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model.
[0045] In one example, the new configuration parameters can be understood as the optimal configuration parameters in the plurality of initial configuration parameters. At this time, the score of each initial configuration parameter can be determined by the pre-trained multi-granularity hybrid expert model, and the optimal configuration parameters in the plurality of initial configuration parameters can be determined according to the score of each initial configuration parameter.
[0046] S103, the configuration parameters of the credit risk model are updated according to the new configuration parameters to obtain a new credit risk model, and a new champion challenge is performed according to the new credit risk model.
[0047] As described above, the embodiments of the present application can obtain a plurality of initial configuration parameters determined by the credit risk model in the initial champion challenge, thereby improving the diversity and flexibility of the initial configuration parameters, avoiding obtaining the initial configuration parameters only according to the static method, thereby making the initial configuration parameters have greater limitations, and thereby affecting the accuracy of the initial configuration parameters. Then, after the plurality of initial configuration parameters are divided into a plurality of groups, the plurality of initial configuration parameters are input into the pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model. This implementation can determine the new configuration parameters corresponding to the credit risk model by the pre-trained multi-granularity hybrid expert model, thereby improving the accuracy of the configuration parameters of the credit risk model. At this time, the configuration parameters of the credit risk model can be updated according to the new configuration parameters to obtain a new credit risk model with more accurate configuration parameters, thereby obtaining a credit risk model with better performance, and thereby improving the performance of the credit risk model. When a new champion challenge is performed according to the new credit risk model, not only the speed can be improved, but also the ability of the credit risk model to identify credit risks can be improved.
[0048] In Figure 1Based on the embodiments, the implementation process of the parameter adjustment method of the credit risk model is described in detail. Specifically, refer to Figure 2 , Figure 2 The implementation process diagram of the parameter adjustment method of the credit risk model provided by the embodiments of the present application is shown in Figure 2 , a plurality of initial configuration parameters can be obtained first, and each initial configuration parameter is determined as an information vector, obtaining a plurality of information vectors.
[0049] Then, the plurality of information vectors can be input into the pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters. At this time, the configuration parameters of the credit risk model can be updated according to the new configuration parameters to obtain a new credit risk model. Then, a new champion is determined according to the new credit risk model, so as to obtain a champion model for credit risk identification.
[0050] Figure 3 The flowchart of another parameter adjustment method of the credit risk model provided by the embodiments of the present application is shown in Figure 3 , based on the embodiments of Figure 1 and Figure 2 , the parameter adjustment method of the credit risk model is described in detail. The method comprises:
[0051] S301, obtaining a plurality of initial configuration parameters determined by the credit risk model in the initial champion challenge.
[0052] Among them, the initial configuration parameter indicates the configuration parameter applied by the credit risk model in the initial champion challenge.
[0053] In one example, this step can refer to the content described in S101 above, which will not be described in detail here.
[0054] In one possible implementation, each initial configuration parameter can include a plurality of sub-parameters. At this time, after the plurality of initial configuration parameters are determined, the initial configuration parameters can be grouped according to the respective sub-parameters included in each initial configuration parameter. Refer to the content described in S302 below.
[0055] S302, comparing the plurality of sub-parameters included in each initial configuration parameter, and dividing the initial configuration parameters including at least partially same sub-parameters into a group to obtain a plurality of groups of initial configuration parameters.
[0056] For example, assume that there are two initial configuration parameters, where the initial configuration parameter 1 is "scene: application scene; business tracking variable: default of corporate customers; traffic allocation mode: independent operation; model update frequency: daily frequency", and the initial configuration parameter 2 is "scene: application scene; business tracking variable: default of customers including overdue and interest arrears; traffic allocation mode: shunting operation; model update frequency: monthly frequency". At this time, since the initial configuration parameter 1 and the initial configuration parameter 2 have the same sub-parameter "scene: application scene", the initial configuration parameter 1 and the initial configuration parameter 2 can be divided into a group.
[0057] This implementation can divide a group of initial configuration parameters as a coarse granularity, and divide different sub-parameters included in each initial configuration parameter as a fine granularity, thereby constructing input data at different granularities. Meanwhile, this implementation can divide initial configuration parameters that do not have the same sub-parameters into different groups, thereby reducing the processing complexity of the model (i.e., the pre-trained multi-granularity hybrid expert model), and further improving the quality and efficiency of model processing.
[0058] In a possible implementation, after the plurality of initial configuration parameters are divided into a plurality of groups, the plurality of groups of initial configuration parameters can be input into the pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model. At this time, in the case where the pre-trained multi-granularity hybrid expert model includes a plurality of expert models and a gating network model, reference can be made to the content described below in S303.
[0059] S303, input each group of initial configuration parameters into the corresponding expert model to obtain the parameter score output by each expert model; and input the plurality of groups of initial configuration parameters into the gating network model to obtain the configuration parameter weight.
[0060] In one example, the parameter score output by each expert model can indicate the parameter score corresponding to each initial configuration parameter in each group of initial configuration parameters.
[0061] In one example, the configuration parameter weight output by the gating network model can include the configuration parameter weight corresponding to each expert model, that is, one or more initial configuration parameters in a group of initial configuration parameters share the configuration parameter weight.
[0062] At this time, the configuration parameter weight output by the gating network model can be used to maximize the use of relevant knowledge between different expert models and form a complementary effect, thereby improving the generalization ability of the pre-trained multi-granularity hybrid expert model. Meanwhile, the configuration parameter weight corresponding to each expert model can be adjusted in time in the case of updating the pre-trained multi-granularity hybrid expert model, thereby improving the performance of the optimized configuration parameters.
[0063] In a possible implementation, the expert model is a residual network model, and the residual network model includes a plurality of network layers; each network layer is a bottleneck structure.
[0064] At this time, the residual network model can be determined as the expert model, thereby being able to have the feature extraction capability of the traditional convolutional neural network model while solving the problem of small gradient, solving the problem of erosion, and adopting a skip connection to add the input feature and the output feature, thereby better preserving the original feature.
[0065] In a possible implementation, the gating network model includes a weight selection layer, a fully connected layer, and a normalization layer. At this time, the corresponding weight determination manner can be determined through the weight selection layer, for example, the weight selection layer can determine the weight corresponding to each expert model through linear transformation, and the weight selection layer is not limited here as long as it can be implemented.
[0066] S304, after the parameter score is weighted and calculated according to the configuration parameter weight, a weighted parameter score is obtained.
[0067] At this time, the weighted parameter score corresponding to each initial configuration parameter can be obtained.
[0068] S305, a new configuration parameter is determined according to the weighted parameter score.
[0069] In an example, the initial configuration parameter corresponding to the weighted parameter score with the highest value can be determined as the new configuration parameter.
[0070] S306, the configuration parameter of the credit risk model is updated according to the new configuration parameter, a new credit risk model is obtained, and a new champion challenge is performed according to the new credit risk model.
[0071] In Figure 3 Based on the embodiments, the implementation process of the parameter adjustment method of the credit risk model is further described in detail, specifically, refer to Figure 4 , Figure 4 Another implementation process of the parameter adjustment method of the credit risk model provided by the embodiments of the present application is shown in the following Figure 4As shown, after obtaining the plurality of initial configuration parameters, each initial configuration parameter can be determined as an information vector, obtaining a plurality of information vectors (for example, Figure 4 As shown, Vector 1-Vector 9), and then, the plurality of information vectors are divided into a plurality of groups (for example, Figure 4 As shown, Vector Group 1-Vector Group m).
[0072] After that, each vector group can be input into the corresponding expert model to obtain the parameter score output by each expert model, and the plurality of initial configuration parameters are input into the gating network model to obtain the configuration parameter weight.
[0073] At this time, according to the configuration parameter weight, each parameter score can be weighted and calculated to obtain a weighted parameter score, and according to the weighted parameter score, a new configuration parameter is determined, that is, a new champion challenge configuration parameter is determined.
[0074] In one possible implementation, the multi-granularity hybrid expert model can be trained through the following process.
[0075] First, according to the historical configuration parameters of the credit risk model and the model processing scores under each historical configuration parameter, a training data set is constructed. In one example, the historical configuration parameters can be divided into a plurality of groups of historical configuration parameters according to the grouping method described above, and each group of historical configuration parameters is used to train an expert model in the multi-granularity hybrid expert model. In one example, the model processing score under each historical configuration parameter can be determined according to the credit risk identification effect of the credit risk model under the corresponding historical configuration parameter.
[0076] Then, the multi-granularity hybrid expert model is trained according to the training data set to obtain a pre-trained multi-granularity hybrid expert model.
[0077] In one possible implementation, each expert model included in the multi-granularity hybrid expert model can be trained according to a cross-entropy loss function to obtain a trained expert model. Then, the gating network model included in the multi-granularity hybrid expert model is trained according to the cross-entropy loss function to obtain a trained gating network model. Then, the pre-trained multi-granularity hybrid expert model can be determined according to the trained expert model and the trained gating network model.
[0078] In one example, the cross-entropy loss function corresponding to the expert model can be represented by the following formula (1).
[0079]
[0080] wherein, N represents the number of training samples corresponding to the expert model, c represents the number of information vectors included in each training sample, y ic represents the model processing score, h c represents the parameter score of the information vector output by the expert model.
[0081] In one example, after training the expert model, the gating network model is further trained. In actual implementation, the output results of all expert models are weighted to obtain weighted output results, and the gating network model is trained according to the cross-entropy loss function shown in formula (1) based on the weighted output results, to obtain the trained gating network model.
[0082] In one possible implementation, when it is determined that the number of groups corresponding to the plurality of initial configuration parameters is different from the number of expert models in the pre-trained multi-granularity hybrid expert model, the number of expert models included in the pre-trained multi-granularity hybrid expert model is updated.
[0083] For example, when new configuration parameters appear and the configuration parameter groups are added, new expert models can be added and trained separately, so that the expert models can be updated without affecting the trained expert models.
[0084] Figure 5 A structural schematic diagram of a parameter adjustment device of a credit risk model provided by an embodiment of the present application is shown in FIG. 1. Figure 5 As shown in FIG. 1, the parameter adjustment device 50 of the credit risk model provided by the embodiment includes:
[0085] An acquisition unit 501 is configured to acquire a plurality of initial configuration parameters determined by a credit risk model in an initial champion challenge; wherein the initial configuration parameters indicate configuration parameters applied by the credit risk model in the initial champion challenge.
[0086] A processing unit 502 is configured to input the plurality of initial configuration parameters into a pre-trained multi-granularity hybrid expert model for processing after the plurality of initial configuration parameters are divided into a plurality of groups, to obtain new configuration parameters corresponding to the credit risk model.
[0087] An updating unit 503 is configured to update configuration parameters of the credit risk model according to the new configuration parameters, to obtain a new credit risk model, and to perform a new champion challenge according to the new credit risk model.
[0088] In one possible implementation, the initial configuration parameters include a plurality of sub-parameters; at this time, the processing unit 502 is configured to:
[0089] The plurality of sub-parameters included in each of the initial configuration parameters are compared, and initial configuration parameters including at least partially same sub-parameters are divided into a group, to obtain a plurality of groups of initial configuration parameters.
[0090] In a possible implementation, the pre-trained multi-granularity hybrid expert model includes a plurality of expert models and a gating network model; at this time, the processing unit 502 is configured to:
[0091] Each group of initial configuration parameters is input into a corresponding expert model to obtain parameter scores output by each expert model; and the plurality of groups of initial configuration parameters are input into the gating network model to obtain configuration parameter weights;
[0092] The parameter scores are weighted and calculated according to the configuration parameter weights to obtain weighted parameter scores;
[0093] The new configuration parameters are determined according to the weighted parameter scores.
[0094] In a possible implementation, the expert model is a residual network model, and the residual network model includes a plurality of network layers; each network layer is a bottleneck structure.
[0095] The gating network model includes a weight selection layer, a fully connected layer, and a normalization layer.
[0096] In a possible implementation, the apparatus is further configured to:
[0097] The training data set is constructed according to the historical configuration parameters of the credit risk model and the model processing scores under each historical configuration parameter;
[0098] The pre-trained multi-granularity hybrid expert model is obtained by training and processing the multi-granularity hybrid expert model according to the training data set.
[0099] In a possible implementation, the apparatus is further configured to:
[0100] Each expert model included in the multi-granularity hybrid expert model is trained and processed according to a cross-entropy loss function to obtain a trained expert model;
[0101] The gating network model included in the multi-granularity hybrid expert model is trained and processed according to a cross-entropy loss function to obtain a trained gating network model.
[0102] The pre-trained multi-granularity hybrid expert model is determined according to the trained expert model and the trained gating network model.
[0103] In a possible implementation, the apparatus is further configured to:
[0104] In a case where it is determined that the number of configuration parameter groups included in the training data set is different from the number of expert models in the multi-granularity hybrid expert model, the number of expert models included in the multi-granularity hybrid expert model is updated.
[0105] The parameter adjustment apparatus of the credit risk model provided in the embodiment can execute the method provided in the method embodiment, and has similar implementation principles and technical effects. Details are not described herein again.
[0106] Figure 6 A structural schematic diagram of a computer device is provided in the embodiment. As shown in the figure, Figure 6 The computer device 60 provided in the embodiment includes at least one processor 601 and a memory 602. Optionally, the computer device 60 further includes a communication component 603. The processor 601, the memory 602 and the communication component 603 are connected through a bus 604.
[0107] In the implementation process, the at least one processor 601 executes the computer execution instructions stored in the memory 602, so that the at least one processor 601 executes the method described above.
[0108] The specific implementation process of the processor 601 can refer to the method embodiment described above, and has similar implementation principles and technical effects. Details are not described herein again.
[0109] In the above embodiment, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC) and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like. The steps of the method disclosed in the application can be directly embodied as execution completed by a hardware processor, or executed by a combination of hardware and software modules in the processor.
[0110] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), for example, at least one disk memory.
[0111] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.
[0112] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described above.
[0113] The present application also provides a computer readable storage medium, which stores computer execution instructions, and when a processor executes the computer execution instructions, the method described above is implemented.
[0114] The readable storage medium described above can be realized by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special purpose computer.
[0115] An exemplary readable storage medium is coupled to the processor, so that the processor can read information from the readable storage medium, and can write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.
[0116] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0117] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.
[0118] In addition, each functional unit in various embodiments of the application can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.
[0119] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the various embodiment methods of the application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.
[0120] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The aforementioned program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, and various media that can store program codes.
[0121] Finally, it should be noted that those skilled in the art, after considering the specification and practicing the application disclosed herein, will easily think of other embodiments of the application. The application is intended to cover any variations, uses, or adaptations of the application that follow the general principles of the application and include common knowledge or conventional technical means in the art that are not disclosed by the application, and is not limited to the precise structure described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the application is only limited by the appended claims.
Claims
1. A method for adjusting parameters in a credit risk model, characterized in that, include: Obtain multiple initial configuration parameters determined by the credit risk model in the initial champion challenge; wherein, the initial configuration parameters indicate the configuration parameters applied by the credit risk model when conducting the initial champion challenge; After dividing the initial configuration parameters into multiple groups, they are input into a pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model. The configuration parameters of the credit risk model are updated according to the new configuration parameters to obtain a new credit risk model, and a new champion challenge is conducted based on the new credit risk model.
2. The method according to claim 1, characterized in that, The initial configuration parameters include multiple sub-parameters; these multiple initial configuration parameters are divided into multiple groups, including: The multiple sub-parameters included in each of the initial configuration parameters are compared, and the initial configuration parameters that include at least some of the same sub-parameters are grouped together to obtain multiple groups of initial configuration parameters.
3. The method according to claim 2, characterized in that, The pre-trained multi-granularity hybrid expert model includes multiple expert models and a gated network model; after dividing the multiple initial configuration parameters into multiple groups, they are input into the pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model, including: Each set of initial configuration parameters is input into the corresponding expert model to obtain the parameter score output by each expert model; and the multiple sets of initial configuration parameters are input into the gated network model to obtain the configuration parameter weights. Based on the configuration parameter weights, the parameter scores are weighted and calculated to obtain the weighted parameter scores; The new configuration parameters are determined based on the weighted parameter scores.
4. The method according to claim 3, characterized in that, The expert model is a residual network model, which includes multiple network layers; each network layer is a bottleneck structure. The gated network model includes a weight selection layer, a fully connected layer, and a normalization layer.
5. The method according to any one of claims 1-4, characterized in that, The method further includes: A training dataset is constructed based on the historical configuration parameters of the credit risk model and the model processing score under each historical configuration parameter. The multi-granularity hybrid expert model is trained based on the training dataset to obtain the pre-trained multi-granularity hybrid expert model.
6. The method according to claim 5, characterized in that, The multi-granularity hybrid expert model is trained based on the training dataset to obtain the pre-trained multi-granularity hybrid expert model, including: The multi-granularity hybrid expert model is trained using the cross-entropy loss function to obtain a trained expert model. The gating network model included in the multi-granularity hybrid expert model is trained according to the cross-entropy loss function to obtain the trained gating network model. Based on the trained expert model and the trained gating network model, the pre-trained multi-granularity hybrid expert model is determined.
7. The method according to claim 5, characterized in that, The method further includes: If it is determined that the grouping of configuration parameters included in the training dataset is different from the number of expert models in the multi-granularity hybrid expert model, the number of expert models included in the multi-granularity hybrid expert model is updated.
8. A parameter adjustment device for a credit risk model, characterized in that, include: An acquisition unit is used to acquire multiple initial configuration parameters determined by the credit risk model in the initial champion challenge; wherein the initial configuration parameters indicate the configuration parameters applied by the credit risk model when conducting the initial champion challenge; The processing unit is used to divide the multiple initial configuration parameters into multiple groups and input them into a pre-trained multi-granularity hybrid expert model for processing to obtain new configuration parameters corresponding to the credit risk model. The update unit is used to update the configuration parameters of the credit risk model according to the new configuration parameters to obtain a new credit risk model, and to conduct a new champion challenge based on the new credit risk model.
9. A computer device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.