Method and device for determining whether to open credit function, and electronic device

Abstract

The present application discloses a method, device and electronic device for determining whether to open a credit function. The method includes: obtaining relevant information of an object to be detected, wherein the relevant information includes: static data for describing personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; using a credit risk prediction model set to process and analyze the relevant information of the object to be detected, and obtaining a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, and the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by multiple credit risk prediction models. The present application solves the technical problem of inaccurate evaluation results caused by the fact that in the related art, when evaluating the credit risk of a user, only static characteristics describing the personal characteristics of the user are considered.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for determining whether to open a credit function, and an electronic device. Background Art

[0002] With the development of Internet technology, the Internet finance market is also growing. The consumer finance market has shown explosive growth. This change has not only changed people's consumption habits, but also made personal lending the choice of more and more people. The evaluation of user credit is crucial for financial institutions. When facing a large number of loan applicants, accurate evaluation of user credit can help financial institutions better identify risks. In related technologies, when evaluating the credit of users, only the user's personal characteristics, such as age, gender, occupation, education level, etc., are used as evaluation factors; therefore, there is a technical problem of inaccurate evaluation results.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method and device, and an electronic device for determining whether to open a credit function, so as to at least solve the technical problem of inaccurate evaluation results caused by only considering static features that describe the user's personal characteristics when evaluating the credit risk of the user in the related technology.

[0005] According to one aspect of an embodiment of the present application, a method for determining whether to open a credit function is provided, comprising: obtaining relevant information of an object to be detected, wherein the relevant information comprises: static data for describing personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; using a credit risk prediction model set to process and analyze the relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, the credit risk prediction model set comprises multiple credit risk prediction models, the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models, and the multiple credit risk prediction models are neural network models trained using training data, and the training data comprises: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each object.

[0006] Optionally, the static data and the dynamic data include: first feature data of text type, second feature data of numerical type; using a credit risk prediction model to process and analyze relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, including: converting the first feature data into a target format that can be recognized by the credit risk prediction model, and mapping the numerical value of each second feature data to a preset numerical range to obtain multiple feature data, and generating a feature matrix based on the multiple feature data, wherein the multiple feature data include: the mapping results of the converted first feature data and the second feature data; using each credit risk prediction model in the credit risk prediction model set to process the feature matrix respectively to obtain a prediction value output by each credit risk prediction model, wherein the prediction value is used to indicate the probability of credit default behavior of the object to be detected; obtaining the preset weight corresponding to each credit risk prediction model, and determining the credit risk prediction value based on multiple preset weights and multiple prediction values, wherein the preset weight is used to indicate the accuracy of the credit risk prediction model.

[0007] Optionally, the preset weight corresponding to each credit risk prediction model is obtained by the following method: generating multiple training sets based on the training data, wherein each training set contains an equal amount of training data; using multiple training sets to perform multiple rounds of verification on multiple credit risk prediction models to obtain multiple accuracy rates corresponding to each credit risk prediction model, wherein in each round of verification, different training sets are used to determine the accuracy rates of different credit risk prediction models; for each credit risk prediction model, determining the average value of the multiple accuracy rates corresponding to the credit risk prediction model, and determining the average value as the preset weight of the credit risk prediction model.

[0008] Optionally, the multiple credit risk prediction models included in the credit risk prediction model set include: a target credit risk prediction model trained using a particle swarm optimization algorithm and training data, wherein the target credit risk prediction model is trained by the following method: multiple model parameter sets of the target credit risk prediction model are determined as multiple particles to be optimized, wherein each model parameter set is a particle to be optimized, and each model parameter set contains multiple model parameters of the target credit risk prediction model, and the model parameters include: a learning rate of the target credit risk prediction model, a regularization parameter of the target credit risk prediction model, and the depth of each decision tree in the multiple decision trees that constitute the target credit risk prediction model; and an adaptive particle swarm optimization algorithm is used to iteratively optimize the particles to be optimized until the number of iterations reaches a preset number, and the training is determined to be completed.

[0009] Optionally, an adaptive particle swarm optimization algorithm is used to iteratively optimize the particles to be optimized, including: after each iterative optimization, obtaining an iteratively optimized target credit risk prediction model; determining a first evaluation result of the initial target credit risk prediction model, and determining a second evaluation result of the iteratively optimized target credit risk prediction model, wherein the initial target credit risk prediction model is a target credit risk prediction model whose model parameters are the particles to be optimized, the first evaluation result is used to indicate the accuracy of the initial target credit risk prediction model, and the second evaluation result is used to indicate the accuracy of the iteratively optimized target credit risk prediction model; comparing the first evaluation result and the second evaluation result to obtain a comparison result, and when the comparison result indicates that the accuracy indicated by the first evaluation result is higher, performing the next iterative optimization on the initial target credit risk prediction model; when the comparison result indicates that the accuracy indicated by the second evaluation result is higher, performing the next iterative optimization on the iteratively optimized target credit risk prediction model.

[0010] Optionally, the particles to be optimized are iteratively optimized using an adaptive particle swarm optimization algorithm, and further include: determining a third evaluation result of each particle to be optimized each time the iterative optimization is performed; determining an inertia weight used in this iterative optimization based on multiple third evaluation results, wherein the inertia weight is used to adjust the value of the particle to be optimized, the adjustment direction of the value, and the adjustment amplitude of the value, and the adjustment direction includes: increase and decrease.

[0011] Optionally, the training data also includes: the actual credit risk value of each object, determining the third evaluation result of each particle to be optimized, including: adjusting each model parameter of the target credit risk prediction model to the model parameter contained in the particle to be optimized, and obtaining the adjusted target credit risk prediction model; using the adjusted target credit risk prediction model to process and analyze the static data of multiple objects and the historical dynamic data related to the historical transaction behavior of each object, and obtain multiple prediction results corresponding to the multiple objects; determining the third evaluation result based on the multiple prediction results and the multiple actual credit risk values.

[0012] According to another aspect of an embodiment of the present application, there is also provided a device for determining whether to open a credit function, including: an acquisition module, used to acquire relevant information of an object to be detected, wherein the relevant information includes: static data used to describe personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; an analysis module, using a credit risk prediction model set to process and analyze the relevant information of the object to be detected, and obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, and the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by multiple credit risk prediction models, and the multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each object.

[0013] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above-mentioned method for determining whether to open a credit function through the computer program.

[0014] According to another aspect of the embodiment of the present application, a computer program product is also provided, including computer instructions, which implement the above-mentioned method for determining whether to open a credit function when the computer instructions are executed by a processor.

[0015] In an embodiment of the present application, relevant information of the object to be detected is obtained, wherein the relevant information includes: static data for describing personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; a credit risk prediction model set is used to process and analyze the relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, and the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, historical transactions with each object, and dynamic transactions with each object. The method of historical dynamic data related to user's behavior, by simultaneously taking static data used to describe the user's personal characteristics and dynamic data related to the user's transaction behavior as evaluation factors, is used to evaluate the user's credit risk, thereby improving the comprehensiveness of the evaluation factors; wherein, a set of trained neural network models is used to evaluate the user's credit risk based on the above evaluation factors, and the user's credit risk is jointly determined according to the prediction results of each neural network model in the neural network model set, thereby achieving the purpose of improving the accuracy of the evaluation results, thereby achieving the technical effect of reducing credit risk, and further solving the technical problem of inaccurate evaluation results caused by only considering static features describing the user's personal characteristics when evaluating the user's credit risk in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a hardware structure block diagram of a computer terminal for implementing a method for determining whether to open a credit function according to an embodiment of the present application;

[0018] Figure 2 is a flowchart of a method for determining whether to open a credit function according to an embodiment of the present application;

[0019] Figure 3 is a record table of relevant information of an object to be detected according to an embodiment of the present application;

[0020] Figure 4 It is a structural diagram of a device for determining whether to open a credit function according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In the related art, the credit risk of a user is evaluated based on static data such as the user's age, gender, occupation, and education level as evaluation factors. Therefore, there is a problem of inaccurate evaluation results caused by a single evaluation factor. In order to solve this problem, a relevant solution is provided in the embodiments of the present application, which is described in detail below.

[0024] According to an embodiment of the present application, a method embodiment for determining whether to open a credit function is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0025] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing a method for determining whether to open a credit function is shown in FIG. Figure 1As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0026] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0027] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for determining whether to open the credit function in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned method for determining whether to open the credit function is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0028] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0029] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0030] The embodiment of the present application provides a method for determining whether to open a credit function that can be run in the above-mentioned operating environment. Figure 2 is a flowchart of the steps of the method for determining whether to open the credit function, such as Figure 2 As shown, the method comprises the following steps:

[0031] Step S202, obtaining relevant information of the object to be detected, wherein the relevant information includes: static data for describing personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected.

[0032] The method provided in the embodiment of the present application uses static data used to describe the user's personal characteristics and dynamic data related to the user's transaction behavior as evaluation factors to evaluate the user's credit risk, wherein the user's personal characteristics include age, gender, occupation, etc., then the data used to describe the user's personal characteristics include: data describing the user's age, data describing the user's gender, data describing the user's occupation; the user's transaction behavior includes: consumption, repayment, income, etc., for example, recent income, recent credit card bills, recent repayment status, etc., these data can reflect the user's repayment ability, therefore, in step S202, before determining whether to open the credit function to the object to be detected, firstly, relevant information of the object to be detected is required, Figure 3 It is a record table of relevant information of the object to be detected, such as Figure 3As shown, the relevant data of the object to be detected include: static data describing the personal characteristics of the object to be detected, such as age X5, gender X2, occupation X7-X8, education level X3, and other personal characteristics, such as static data of marital status X4, etc.; and dynamic data related to the transaction behavior of the object to be detected, such as recent (such as within one year) income amount X6, recent repayment situation: such as monthly repayment situation X9-X14, recent consumption amount: such as monthly credit card consumption amount X15-X20, monthly total payment amount X21-X26; and other information related to the credit behavior of the object to be detected: such as credit limit X1, whether there is any default behavior Y.

[0033] It should be noted that the static data and dynamic data of the object to be detected obtained in step S202 are information and data authorized by the object to be detected or "fully authorized" by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding "operation entrances" for users (objects to be detected) to authorize or refuse.

[0034] Optionally, the static data and the dynamic data include: first feature data of text type, second feature data of numerical type; using a credit risk prediction model to process and analyze relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, including: converting the first feature data into a target format that can be recognized by the credit risk prediction model, and mapping the numerical value of each second feature data to a preset numerical range to obtain multiple feature data, and generating a feature matrix based on the multiple feature data, wherein the multiple feature data include: the mapping results of the converted first feature data and the second feature data; using each credit risk prediction model in the credit risk prediction model set to process the feature matrix respectively to obtain a prediction value output by each credit risk prediction model, wherein the prediction value is used to indicate the probability of credit default behavior of the object to be detected; obtaining the preset weight corresponding to each credit risk prediction model, and determining the credit risk prediction value based on multiple preset weights and multiple prediction values, wherein the preset weight is used to indicate the accuracy of the credit risk prediction model.

[0035] The relevant information of the object to be detected includes both text-type data describing the gender, education level, occupation, etc. of the object to be detected (i.e., the first characteristic data), and numerical data such as the recent consumption amount of the object to be detected (i.e., the second characteristic data). Figure 3As shown, in this embodiment, when the credit risk prediction model set is used to analyze and process the above data, first, the static data of the above text type is converted into a format that can be recognized by each credit risk prediction model (i.e., the target format), for example, Figure 3 The numeric format shown; and, mapping dynamic data of numeric type to numeric values ​​within the same numeric range, for example, Figure 3 In the method, the repayment situation of the object to be detected is classified into different values ​​according to the difference between the actual repayment time and the prescribed repayment time in the actual repayment situation of the object to be detected. Further, the text type data converted into a numerical type (i.e., the first feature data after conversion) and the numerical type data mapped to the same numerical range (i.e., the mapping result of the second feature data) are used as elements of the matrix to generate a feature matrix; when the generated feature matrix is ​​processed and analyzed using a set of credit risk prediction models, each credit risk prediction model in the set processes and analyzes the feature matrix once, and outputs a result (i.e., a prediction value) for indicating the probability of credit default behavior of the object to be detected, and the result finally determined according to the weight corresponding to each credit risk prediction model (i.e., the preset weight) and the prediction value output by each credit risk prediction model is finally used as the credit risk prediction value of the output result; the preset weight used in the process of determining the credit risk prediction value as the final output result is determined according to the accuracy of each credit risk prediction model during the training process, and can be updated during the application process.

[0036] According to some optional embodiments of the present application, the preset weight corresponding to each credit risk prediction model is obtained by the following method: generating multiple training sets based on the training data, wherein each training set contains an equal amount of training data; using multiple training sets to perform multiple rounds of verification on multiple credit risk prediction models to obtain multiple accuracy rates corresponding to each credit risk prediction model, wherein in each round of verification, different training sets are used to determine the accuracy rates of different credit risk prediction models; for each credit risk prediction model, determining the average value of the multiple accuracy rates corresponding to the credit risk prediction model, and determining the average value as the preset weight of the credit risk prediction model.

[0037] As mentioned in the previous embodiment, the weight (i.e., the preset weight) of each credit risk prediction model is determined during the training process. Specifically, it can be determined by the following method provided in this embodiment. During the training process, the training data is divided into multiple training data sets (i.e., training sets) according to the number of credit risk prediction models contained in the credit risk prediction model set, ensuring that multiple training sets contain equal amounts of training data; in the process of iterative training, each credit risk prediction model is used to process and analyze its corresponding training data to verify the accuracy of the credit risk prediction model. For example, the ratio of the prediction result output by the credit risk prediction model for processing and analyzing its corresponding training set to the actual credit risk value of the object corresponding to the training set contained in the training data is determined as the accuracy of the credit risk prediction model; after multiple rounds of training, multiple accuracy rates corresponding to each credit risk prediction model can be obtained. In this embodiment, the average value of the multiple accuracy rates corresponding to each credit risk prediction model is determined as the weight (i.e., the preset weight) of the credit risk prediction model.

[0038] Step S204, using a credit risk prediction model set to process and analyze relevant information of the object to be detected, and obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, and the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each object.

[0039] After the above-mentioned static data and dynamic data (i.e., relevant information of the object to be detected) used as evaluation factors are obtained in step S202, in step S204, each information risk prediction model in the credit risk prediction model set is used to process and analyze the static data and dynamic data of the object to be detected, each credit risk prediction model outputs a prediction value for indicating the credit risk of the object to be detected, and a credit risk prediction value for ultimately indicating the credit risk of the object to be detected is determined based on the multiple prediction values ​​output by the multiple credit risk prediction models, and the final credit risk prediction value is output as an output result, and it is determined whether to open a credit function to the object to be detected based on the output result; wherein, the larger the numerical value of the credit risk prediction value, the lower the credit of the object to be detected and the higher the credit risk, therefore, if the output result is greater than the preset credit risk value, the credit function is refused to be opened to the object to be detected, otherwise, if the output result is less than or equal to the preset credit risk value, the credit function is allowed to be opened to the object to be detected. Each credit risk prediction model in the credit risk prediction model set adopted in step S204 is a neural network model trained with static data of multiple users (i.e., multiple objects) and historical dynamic data of these multiple users as training data, wherein the historical dynamic data of each user is generated based on the historical transaction behavior of each user before the current moment, and is used to describe the user's known historical transaction behavior.

[0040] In step S204, the credit risk prediction model set may be loaded into the memory, for example, the raw data of the credit risk prediction model set may be loaded from the non-volatile memory into the volatile memory, so that the processor runs the credit risk prediction model set. The raw data of the credit risk prediction model set refers to unprocessed data, and generally includes parameters and structural data of the credit risk prediction model set. The structural data may be a calculation relationship based on the parameters, such as a forward propagation calculation relationship between intermediate layers and neurons. Specifically, the structural data may include code related to the structure of the credit risk prediction model set, such as code for performing related calculations between intermediate layers and neurons.

[0041] In one embodiment, an area for loading a set of credit risk prediction models can be divided in the memory, which may include a structure data storage area and a parameter storage area. The structure data storage area is used to store structure-related codes, and the parameters referenced by it can point to the address of a specific parameter in the parameter storage area through a pointer. During the training process of the credit risk prediction model set, it may be necessary to frequently update the parameters, and the parameter values ​​in the parameter storage area can be updated.

[0042] Optionally, the multiple credit risk prediction models included in the credit risk prediction model set include: a target credit risk prediction model trained using a particle swarm optimization algorithm and training data, wherein the target credit risk prediction model is trained by the following method: multiple model parameter sets of the target credit risk prediction model are determined as multiple particles to be optimized, wherein each model parameter set is a particle to be optimized, and each model parameter set contains multiple model parameters of the target credit risk prediction model, and the model parameters include: a learning rate of the target credit risk prediction model, a regularization parameter of the target credit risk prediction model, and the depth of each decision tree in the multiple decision trees that constitute the target credit risk prediction model; and an adaptive particle swarm optimization algorithm is used to iteratively optimize the particles to be optimized until the number of iterations reaches a preset number, and the training is determined to be completed.

[0043] The credit risk prediction model set used in the embodiment of the present application includes a neural network model (i.e., the target credit risk prediction model) trained by the particle swarm optimization algorithm. In this embodiment, when the particle swarm optimization algorithm is used to train the (target) credit risk prediction model, a set of model parameters of the target credit risk prediction model to be trained is used as a particle to be optimized, that is, each particle to be optimized is a set of model parameters, wherein the model parameters included in the model parameter set are the learning rate of the target credit risk prediction model, the regularization parameter of the target credit risk prediction model, the depth of each decision tree in the multiple decision trees constituting the target credit risk prediction model, and other hyperparameters of the target credit risk prediction model, wherein at least one model parameter with different values ​​exists in the two model parameter sets corresponding to any two particles to be optimized. During the training process, the particle to be optimized is iteratively optimized until the number of iterations reaches the preset number, the optimization result of the particle to be optimized outputted by the last iteration is obtained, and the numerical value of the model parameter of the target credit risk prediction model is correspondingly adjusted to the model parameter numerical value included in the optimization result, and the trained target credit risk prediction model is obtained.

[0044] According to some other optional embodiments of the present application, an adaptive particle swarm optimization algorithm is used to iteratively optimize the particles to be optimized, including: after each iterative optimization, obtaining an iteratively optimized target credit risk prediction model; determining a first evaluation result of the initial target credit risk prediction model, and determining a second evaluation result of the iteratively optimized target credit risk prediction model, wherein the initial target credit risk prediction model is a target credit risk prediction model whose model parameters are the particles to be optimized, the first evaluation result is used to indicate the accuracy of the initial target credit risk prediction model, and the second evaluation result is used to indicate the accuracy of the iteratively optimized target credit risk prediction model; comparing the first evaluation result and the second evaluation result to obtain a comparison result, and when the comparison result indicates that the accuracy indicated by the first evaluation result is higher, performing the next iterative optimization on the initial target credit risk prediction model; when the comparison result indicates that the accuracy indicated by the second evaluation result is higher, performing the next iterative optimization on the iteratively optimized target credit risk prediction model.

[0045] As mentioned in the previous embodiment, the target credit risk prediction model uses a particle swarm algorithm to iteratively optimize the particles to be optimized, wherein a new particle to be optimized will be output after each iterative optimization. Compared with the particle to be optimized before the iterative optimization, the values ​​of one or more model parameters in the new particle to be optimized will be changed; however, whether the new particle to be optimized is a better particle to be optimized needs to be determined through evaluation, that is, whether the result output by the iterative optimization is an optimization result needs to be determined by evaluating the new particle to be optimized and the particle to be optimized before the iterative optimization. In this embodiment, the new particles to be optimized and the particles to be optimized before iterative optimization are evaluated specifically according to the prediction accuracy; during the evaluation, the numerical values ​​of the model parameters of the target credit risk prediction model are respectively adjusted to the numerical values ​​of the model parameters contained in the particles to be optimized before iterative optimization, and the accuracy of the target credit risk prediction model (i.e., the first evaluation result) is determined using the training data; the numerical values ​​of the model parameters of the target credit risk prediction model are correspondingly adjusted to the numerical values ​​of the model parameters contained in the new particles to be optimized, and the accuracy of this target risk prediction model (i.e., the second evaluation result) is determined using the training data; by comparing the two evaluation results (i.e., the first evaluation result and the second evaluation result), the target credit risk prediction model with higher accuracy is screened out, the particles to be optimized to which the model parameters of the target credit risk prediction model with higher accuracy belong are determined as the optimization results, and the optimization results are iteratively optimized in the next iteration.

[0046] Optionally, the particles to be optimized are iteratively optimized using an adaptive particle swarm optimization algorithm, and further include: determining a third evaluation result of each particle to be optimized each time the iterative optimization is performed; determining an inertia weight used in this iterative optimization based on multiple third evaluation results, wherein the inertia weight is used to adjust the value of the particle to be optimized, the adjustment direction of the value, and the adjustment amplitude of the value, and the adjustment direction includes: increase and decrease.

[0047] In the embodiment of the present application, after each iterative optimization is completed and the object of the next iterative optimization is determined, before starting the next iterative update, the adjustment method of the particle to be optimized is determined. Specifically, the adjustment method of the particle to be optimized is changed by adjusting the inertia weight, wherein the adjustment method is used to indicate how to adjust the position (the value of the model parameter) and speed (the adjustment amplitude and adjustment direction of the value of the model parameter) of the particle to be optimized, the adjustment amplitude is used to indicate the amount of change of the value, and the adjustment direction is used to indicate whether the value increases or decreases); the inertia weight can be adjusted by the following formula: Among them, ω min is a predetermined minimum inertia coefficient (for example, 0.4), ω max is a predetermined maximum inertia coefficient (for example, 0.9), is the average value of the evaluation results (i.e., the third evaluation results) obtained by evaluating all particles to be optimized during the d-th iteration; It is the evaluation result with the lowest accuracy among the multiple evaluation results (i.e., the third evaluation result) obtained by evaluating all the particles to be optimized in the dth iteration process; represents the (third) evaluation result of the particle i to be optimized at the d-th iteration; through the method provided in this embodiment, the inertia weight ω is adjusted according to the evaluation result of the particle to be optimized, thereby adjusting the update method of the particle, thereby improving the speed of finding the optimal solution.

[0048] The iterative optimization of the particles to be optimized in the embodiment of the present application can be summarized as the following steps af: a. Update particle speed and position: update the speed and position of each particle to be optimized according to the update rule of the particle swarm optimization algorithm, wherein, when updating the speed and position, the current position of the particle to be optimized, the speed of the particle to be optimized, the individual optimal position of the particle to be optimized, and the global optimal position in the particle swarm to be optimized are considered as factors affecting the update of the position and speed, wherein the current position of the particle to be optimized indicates the value of each model parameter contained in the current particle to be optimized; the speed of the particle to be optimized indicates the change direction of the value of each model parameter randomly determined or preset when the speed is first updated (index value becomes larger or smaller) and amplitude (the amount of change in the index value); the individual optimal position of the particle to be optimized is the optimal position that each particle to be optimized has experienced since initialization, that is, the set of model parameters corresponding to each particle to be optimized is the set of model parameters that maximizes the accuracy of the target credit risk prediction model among the multiple sets of model parameters corresponding to it since initialization; the global optimal position in the particle group to be optimized is the individual optimal position selected from the individual optimal position of each particle to be optimized. When the numerical value of the model parameter of the target credit risk model is adjusted to the numerical value of the model parameter corresponding to the global optimal position, the accuracy of the target credit risk model reaches the maximum value since initialization to the current moment. b. Boundary processing: ensure that the updated hyperparameter value is within a reasonable range; if a hyperparameter (i.e., model parameter) exceeds the predefined search space (i.e., the numerical value of the model parameter exceeds the predicted model parameter numerical range), it can be set to a boundary value or re-initialized randomly. c. Train and evaluate the target credit risk prediction model: Use the current position (hyperparameter) of each particle to configure the target credit risk prediction model and determine the evaluation result; In this embodiment, the target credit risk prediction model can be a decision tree model (XGBoost model) or other neural network models for processing classification and regression problems; When evaluating the target credit risk prediction model, the root mean square error (RMSE) can be selected as the performance metric. The root mean square error (RMSE) is obtained by calculating the square root of the sum of the squares of the difference between the predicted value and the true value, and is used to measure the degree of deviation between the predicted value and the true value. d. Update the individual optimum and the global optimum: Compare the evaluation results before optimization and the evaluation results after optimization of the same particle to be optimized, and determine the particle to be optimized that has the highest accuracy of the target credit risk prediction model. The particle to be optimized is the updated individual optimal position; Then compare the evaluation results before optimization and the evaluation results after optimization of multiple different particles to be optimized, and determine a particle to be optimized that indicates the evaluation result with the highest accuracy of the target credit risk prediction model. The particle to be optimized is the updated global optimal position; The individual optimal position and the global optimal position may correspond to the same particle to be optimized.e. Record the best solution: After each iteration, record the current global optimal solution (best hyperparameter combination) and its corresponding evaluation result. f. When the maximum number of iterations is reached or other stopping conditions are met (such as the evaluation result of the particle to be optimized no longer changes, or the change is less than the preset change), stop the iteration; output the final global optimal solution, that is, the best hyperparameter combination.

[0049] According to some optional embodiments of the present application, the training data also includes: the actual credit risk value of each object, determining the third evaluation result of each particle to be optimized, including: adjusting each model parameter of the target credit risk prediction model to the model parameter contained in the particle to be optimized, and obtaining the adjusted target credit risk prediction model; using the adjusted target credit risk prediction model to process and analyze the static data of multiple objects and the historical dynamic data related to the historical transaction behavior of each object, and obtain multiple prediction results corresponding to the multiple objects; determining the third evaluation result based on the multiple prediction results and the multiple actual credit risk values.

[0050] In the embodiment of the present application, an evaluation method is adopted throughout the iterative optimization process of the target credit risk prediction model. Therefore, whether it is used to evaluate the particles to be optimized before the iterative optimization when determining the particles to be optimized as the object of the next iterative optimization, to evaluate the new particles to be optimized after the iterative optimization is completed, or to evaluate the particles to be optimized each time the inertia weight is adjusted, the same evaluation method is adopted; that is, in the embodiment of the present application, the first evaluation result, the second evaluation result and the third evaluation result are all calculated using the same evaluation method. In the embodiment of the present application, the root mean square error (RMSE) is calculated as the evaluation result; the root mean square error (RMSE) is obtained by calculating the square root of the sum of the squares of the differences between the predicted value and the true value. Therefore, before adjusting the inertia weight, the (third) evaluation result obtained by evaluating the particle to be optimized is also the root mean square error (RMSE) corresponding to the particle to be optimized. When determining the root mean square error (RMSE) of any particle to be optimized, first, the numerical values ​​of the model parameters of the target credit risk prediction model are adjusted to the numerical values ​​of the model parameters contained in the particle to be optimized, and the target credit risk prediction model after adjusting the model parameters is used to process and analyze the training data. Since the training data contains relevant information of multiple objects, multiple predicted values ​​corresponding to multiple objects can be obtained; in addition, to determine the root mean square error (RMSE), not only the predicted value output by the model is required, but also the corresponding true value is required. In this embodiment, the true value corresponding to the predicted value output by the target credit risk prediction model is the historical actual credit risk probability (i.e., the actual credit risk value) of each object corresponding to the training data. Since the training data is historical data, these actual credit risk values ​​are known and are also included in the training data; therefore, the actual credit risk value of each object can be extracted from the training data, and the accuracy of the target credit risk prediction model can be determined based on the actual credit risk value and predicted value corresponding to the same object; and the root mean square error (RMSE) of the target credit risk prediction model at this time is further determined based on the accuracy of the target credit risk prediction model for each object.

[0051] Through the above steps, it is possible to build a comprehensive credit assessment system that includes static data and dynamic data, thereby improving the comprehensiveness of assessment factors; combining multiple credit risk prediction models to assess the credit risk of the object to be tested from multiple dimensions, thereby improving the accuracy of the assessment results.

[0052] Figure 4 is a structural diagram of a device for determining whether to open a credit function according to an embodiment of the present application, such as Figure 4As shown, the device includes: an acquisition module 40, which is used to obtain relevant information of the object to be detected, wherein the relevant information includes: static data used to describe the personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; an analysis module 42, which uses a credit risk prediction model set to process and analyze the relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, and the credit risk prediction model set contains multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each object.

[0053] When the device for determining whether to open the credit function is used to determine the credit risk of a certain object to be detected, the device uses the relevant information of the object to be detected through the acquisition module 40 to construct a complete and comprehensive credit evaluation system, and the credit evaluation system is jointly constructed by the static data and dynamic data of the object to be detected, wherein the static data of the object to be detected is data used to describe the personal characteristics of the object to be detected, including age, gender, occupation, education level, etc.; the dynamic data of the object to be detected is data related to the transaction behavior of the object to be detected, such as the recent (such as within one year) income of the object to be detected, the recent (such as within one year) repayment situation, the recent (such as within one year) debt situation, etc.; the acquisition module 40 transmits the above data obtained to the analysis module 42, and the analysis module 42 uses multiple credit risk prediction models in the credit risk prediction model set to process and analyze the data transmitted by the acquisition module 40, each credit risk model outputs a prediction value for indicating the credit risk probability of the object to be detected, and the analysis module 42 determines a final credit risk prediction value for determining the object to be detected based on the multiple prediction values, and determines whether to open the credit function to the object to be detected based on this credit risk prediction value. Each credit risk prediction model in the credit risk prediction set adopted by the analysis module 42 is a neural network model trained with static data and dynamic data of multiple objects as training data. Each credit risk prediction model is a neural network model used to process classification and regression problems, such as: XGBoost model, adaptive boosting (AdaBoost) model, random forest (RF) model, etc.; and, in order to improve the accuracy of the evaluation results, at least one credit risk prediction model in the credit risk prediction model set is trained using a particle swarm optimization algorithm.

[0054] It should be noted that Figure 4 The preferred implementation of the illustrated embodiment can be found in Figure 2 The relevant description of the illustrated embodiment will not be repeated here.

[0055] An embodiment of the present application also provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above method of determining whether to open a credit function through the computer program.

[0056] The processor in the above-mentioned electronic device is used to run a program that performs the following functions: obtaining relevant information of the object to be detected, wherein the relevant information includes: static data used to describe the personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; using a credit risk prediction model set to process and analyze the relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each object.

[0057] The embodiment of the present application also provides a computer program product, including computer instructions, which implement the above method of determining whether to open a credit function when the computer instructions are executed by a processor.

[0058] It should be noted that the various modules in the above-mentioned device for determining whether to open the credit function can be a program module (for example, a set of program instructions that implement a certain specific function) or a hardware module. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0059] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0060] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0062] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0063] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0064] If the integrated unit is implemented in the form of a software functional 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 solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0065] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining whether to open a credit function, characterized in that: include: Acquire relevant information of the object to be detected, wherein the relevant information includes: static data for describing personal characteristics of the object to be detected, and dynamic data related to transaction behavior of the object to be detected; A credit risk prediction model set is used to process and analyze relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects, and historical dynamic data related to the historical transaction behavior of each of the objects.

2. The method according to claim 1, characterized in that The static data and the dynamic data include: first characteristic data of text type and second characteristic data of numerical type; The credit risk prediction model is used to process and analyze the relevant information of the object to be detected to obtain the credit risk prediction value of the object to be detected, including: Converting the first feature data into a target format recognizable by the credit risk prediction model, and mapping the value of each second feature data to a preset value range to obtain a plurality of feature data, and generating a feature matrix according to the plurality of feature data, wherein the plurality of feature data includes: the first feature data after conversion, and the mapping result of the second feature data; Using each of the credit risk prediction models in the credit risk prediction model set to process the feature matrix, respectively, to obtain the prediction value output by each of the credit risk prediction models, wherein the prediction value is used to indicate the probability of credit default behavior of the object to be detected; Obtain a preset weight corresponding to each of the credit risk prediction models, and determine the credit risk prediction value based on a plurality of the preset weights and a plurality of the prediction values, wherein the preset weight is used to indicate the accuracy of the credit risk prediction model.

3. The method according to claim 2, characterized in that The preset weight corresponding to each of the credit risk prediction models is obtained by the following method: Generate multiple training sets according to the training data, wherein each training set contains an equal amount of training data; Using the multiple training sets to perform multiple rounds of verification on the multiple credit risk prediction models, and obtaining multiple accuracy rates corresponding to each credit risk prediction model, wherein in each round of verification, different training sets are used to determine the accuracy rates of different credit risk prediction models; For each of the credit risk prediction models, an average value of the multiple accuracy rates corresponding to the credit risk prediction model is determined, and the average value is determined as a preset weight of the credit risk prediction model.

4. The method according to claim 1, characterized in that The multiple credit risk prediction models included in the credit risk prediction model set include: a target credit risk prediction model trained using a particle swarm optimization algorithm and the training data, wherein the target credit risk prediction model is trained using the following method: Determine multiple model parameter sets of the target credit risk prediction model as multiple particles to be optimized, wherein each of the model parameter sets is one particle to be optimized, and each of the model parameter sets contains multiple model parameters of the target credit risk prediction model, and the model parameters include: a learning rate of the target credit risk prediction model, a regularization parameter of the target credit risk prediction model, and a depth of each of the multiple decision trees constituting the target credit risk prediction model; The particles to be optimized are iteratively optimized using an adaptive particle swarm optimization algorithm until the number of iterations reaches a preset number, and the training is determined to be completed.

5. The method according to claim 4, characterized in that Adopting an adaptive particle swarm optimization algorithm to iteratively optimize the particles to be optimized, including: After each iteration of the optimization, an iteratively optimized target credit risk prediction model is obtained; Determine a first evaluation result of an initial target credit risk prediction model, and determine a second evaluation result of the iteratively optimized target credit risk prediction model, wherein the initial target credit risk prediction model is a target credit risk prediction model whose model parameters are the particles to be optimized, the first evaluation result is used to indicate the accuracy of the initial target credit risk prediction model, and the second evaluation result is used to indicate the accuracy of the iteratively optimized target credit risk prediction model; Comparing the first evaluation result with the second evaluation result to obtain a comparison result, and when the comparison result indicates that the accuracy indicated by the first evaluation result is higher, performing the next iterative optimization on the initial target credit risk prediction model; When the comparison result indicates that the second evaluation result has a higher accuracy, the next iterative optimization is performed on the iteratively optimized target credit risk prediction model.

6. The method according to claim 4, characterized in that Adopting an adaptive particle swarm optimization algorithm to iteratively optimize the particles to be optimized also includes: Each time the iterative optimization is performed, determining a third evaluation result of each particle to be optimized; The inertia weight used in this iterative optimization is determined according to the plurality of third evaluation results, wherein the inertia weight is used to adjust the value of the particle to be optimized, the adjustment direction of the value, and the adjustment amplitude of the value, and the adjustment direction includes: increase and decrease.

7. The method according to claim 6, characterized in that The training data also includes: an actual credit risk value of each of the objects; determining a third evaluation result of each particle to be optimized includes: Adjusting each model parameter of the target credit risk prediction model to the model parameter included in the particle to be optimized to obtain an adjusted target credit risk prediction model; Using the adjusted target credit risk prediction model to process and analyze the static data of the multiple objects and the historical dynamic data related to the historical transaction behavior of each of the objects, to obtain multiple prediction results corresponding to the multiple objects; The third evaluation result is determined according to the plurality of prediction results and the plurality of actual credit risk values.

8. A device for determining whether to open a credit function, characterized in that: include: An acquisition module, used to acquire relevant information of the object to be detected, wherein the relevant information includes: static data used to describe the personal characteristics of the object to be detected, and dynamic data related to the transaction behavior of the object to be detected; An analysis module uses a credit risk prediction model set to process and analyze relevant information of the object to be detected to obtain a credit risk prediction value of the object to be detected, wherein the credit risk prediction value is used to determine whether to open a credit function to the object to be detected, the credit risk prediction model set includes multiple credit risk prediction models, and the credit risk prediction value is determined based on multiple prediction values ​​output by the multiple credit risk prediction models. The multiple credit risk prediction models are neural network models trained using training data, and the training data includes: static data of multiple objects and historical dynamic data related to the historical transaction behavior of each of the objects.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to execute the method for determining whether to open a credit function according to any one of claims 1 to 7 through the computer program.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the method for determining whether to open a credit function as described in any one of claims 1 to 7 is implemented.

Images