Data processing method and apparatus, electronic device, and program product

Abstract

This application discloses a data processing method, apparatus, electronic device, and program product, relating to the fields of artificial intelligence and fintech. The method includes: collecting target customer information from multiple channels after a preset time interval, and extracting target credit characteristics from the target customer information; determining a target state space based on the target credit information and target credit characteristics in the target customer information; inputting the target state space into a target decision model to obtain the decision actions output by the target decision model and the evaluation results of the decision actions; and determining target adjustment actions for credit services to the target customer based on the evaluation results and decision actions. This application solves the problem in related technologies where financial institutions typically rely on static data to provide services to customers, making it difficult to capture dynamic customer information in a timely manner and respond promptly, resulting in a relatively lagging response mechanism for financial institutions.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and fintech, and more specifically, to a data processing method, apparatus, electronic device, and program product. Background Technology

[0002] In the banking and other financial services sectors, financial institutions typically build service response models based on historically entered static information when providing credit services or related financial products to customers. This static information only reflects the customer's inherent state at the time of data collection and cannot dynamically reflect the real-time evolution of subsequent customer behavior, transaction patterns, or system interaction states. When customers exhibit new operational behaviors in subsequent stages (such as changes in account login frequency, changes in transaction terminals, abnormal fund flows, or a sudden increase in the frequency of system interface calls), the existing system still uses the initially entered data for service decisions, failing to incorporate the newly generated interaction data into the real-time processing flow. Due to the lack of a continuous perception and closed-loop feedback mechanism for dynamic behavioral data, the service response logic lags behind actual changes in customer behavior, failing to trigger automatic verification or process adjustments in the early stages of behavioral anomalies. Therefore, when facing real-time migration of customer behavior patterns, financial institutions generally suffer from technical delays in their service response mechanisms, such as data collection delays, delayed decision updates, and the inability to automatically trigger intervention actions, affecting the adaptive capabilities of the service system and the timeliness of the operational loop.

[0003] There is currently no effective solution to the problem that financial institutions, which typically rely on static data to provide services to customers, are unable to capture dynamic customer information in a timely manner and respond to it promptly, resulting in a relatively lagging response mechanism. Summary of the Invention

[0004] The main purpose of this application is to provide a data processing method, apparatus, electronic device, and program product to solve the problem that financial institutions, which typically rely on static data to provide services to customers, are unable to capture dynamic customer information in a timely manner and respond to such information promptly, resulting in a relatively lagging response mechanism for financial institutions.

[0005] To achieve the above objectives, according to one aspect of this application, a data processing method is provided, comprising: upon detecting that a target customer is using credit services and with the target customer's authorization, collecting target customer information from multiple channels after a preset time interval, and extracting target credit features from the target customer information; determining a target state space based on the target credit information and target credit features in the target customer information, wherein the target state space describes the set of all credit features of the target customer during the use of credit services; inputting the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action, wherein the target decision model is obtained by training a neural network using historical customer information and historical credit features based on a deep deterministic policy gradient algorithm; and determining a target adjustment action for the credit service for the target customer based on the evaluation result and the decision action.

[0006] Furthermore, before inputting the target state space into the target decision model, the above method also includes: acquiring historical customer information of historical customers and extracting historical credit features from the historical customer information, wherein historical customers refer to customers who have applied for credit services within a historical time period; generating predictive adjustment actions based on the historical state space and action space through a first neural network, wherein the historical state space is used to describe the set of all credit features of historical customers during the use of credit services, and the action space is used to describe the set of various adjustment actions taken by financial institutions for credit services based on the historical state space; inputting the predictive adjustment actions into multiple agents to obtain the first observation results determined by the multiple agents according to the reward function, and updating the historical state space according to the first observation results, wherein the multiple agents are determined according to the various credit types of credit services, and the reward function is determined according to the repayment behavior of historical customers; randomly sampling sample data multiple times from the updated historical state space, inputting the sample data extracted each time into a second neural network, and iteratively training the first neural network according to the value evaluation results output by the second neural network to obtain the target decision model.

[0007] Furthermore, the first neural network generates predictive adjustment actions based on the historical state space and action space, including: extracting historical credit information related to the credit services of historical customers from historical customer information, and determining the historical state space based on the historical credit information and historical credit characteristics, wherein the historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information; obtaining multiple credit types of credit services from the credit business system of financial institutions, and determining the action space based on the adjustment actions corresponding to the multiple credit types, wherein the action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan recall, and increasing the minimum collateral amount; inputting the historical state space into the first neural network, calculating the probability distribution information of each adjustment action in the action space through the first neural network, and determining the predictive adjustment action based on the probability distribution information.

[0008] Furthermore, the first neural network is iteratively trained based on the value assessment results output by the second neural network to obtain the target decision model. This includes: inputting randomly selected arbitrary sample data into the second neural network to obtain the value assessment results output by the second neural network; updating the model parameters of the second neural network based on the loss function according to the preset value target and the value assessment results to obtain the updated second neural network; determining gradient information based on the updated second neural network using the policy gradient method; updating the model parameters of the first neural network based on the gradient information, and repeatedly extracting sample data, iteratively updating the model parameters of the first neural network based on each extracted sample data until the number of iterations reaches the preset number to obtain the target decision model.

[0009] Furthermore, after determining the target adjustment action for credit services to the target customer based on the evaluation results and decision action, the above method further includes: inputting the target adjustment action into multiple agents to obtain the second observation results determined by the multiple agents according to the reward function; collecting the repayment behavior of the target customer from multiple channels after a preset time interval; updating the target state space based on the second observation results, the target adjustment action, and the repayment behavior; and updating the model parameters of the second neural network and the target decision model based on the updated target state space to obtain the updated second neural network and the updated target decision model.

[0010] Furthermore, extracting target credit features from target customer information includes: preprocessing the target customer information to obtain first customer information, wherein the preprocessing operations include at least one of the following: data cleaning, data integration, data transformation, and noise reduction; filling in missing values ​​in the first customer information and modifying outliers in the first customer information to obtain second customer information; extracting the original credit features of the target customer from the second customer information through feature engineering, wherein the original credit features include at least: asset features, credit score features, repayment stability features, and financial environment features; and normalizing the original credit features to obtain target credit features.

[0011] Furthermore, target customer information is collected from multiple channels, including: obtaining target customer credit service information from the credit business systems of financial institutions; obtaining target customer credit record information from third-party credit reporting agencies and obtaining financial environment information through the Internet; obtaining target customer profile information, transaction behavior information and asset information from the customer information systems of financial institutions; and determining target customer information based on credit service information, credit record information, financial environment information, customer profile information, transaction behavior information and asset information.

[0012] To achieve the above objectives, according to another aspect of this application, a data processing apparatus is provided, comprising: a first acquisition unit, configured to, upon detecting that a target customer is applying for credit services and with the authorization of the target customer, acquire target customer information from multiple channels after a preset time interval, and extract target credit features from the target customer information; a first determination unit, configured to determine a target state space based on the target credit information and target credit features in the target customer information, wherein the target state space describes the set of all credit features of the target customer during the use of credit services; a decision unit, configured to input the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action, wherein the target decision model is obtained by training a neural network based on a deep deterministic policy gradient algorithm using historical customer information and historical credit features; and a second determination unit, configured to determine a target adjustment action for the credit service for the target customer based on the evaluation result of the decision action and the decision action.

[0013] Furthermore, the aforementioned apparatus further includes: an extraction unit, configured to acquire historical customer information of historical customers and extract historical credit features from the historical customer information before inputting the target state space into the target decision model, wherein historical customers refer to customers who have applied for credit services within a historical time period; a generation unit, configured to generate predictive adjustment actions based on the historical state space and action space through a first neural network, wherein the historical state space describes the set of all credit features of historical customers during the use of credit services, and the action space describes the set of various adjustment actions taken by financial institutions for credit services based on the historical state space; a third determination unit, configured to input the predictive adjustment actions into multiple agents, obtain a first observation result determined by the multiple agents according to the reward function, and update the historical state space according to the first observation result, wherein the multiple agents are determined based on various credit types of credit services, and the reward function is determined based on the repayment behavior of historical customers; and a training unit, configured to randomly extract sample data multiple times from the updated historical state space, input the sample data extracted each time into a second neural network, and iteratively train the first neural network according to the value evaluation result output by the second neural network to obtain the target decision model.

[0014] Furthermore, the generation unit includes: a first determining subunit, used to extract historical credit information related to credit services for historical customers from historical customer information, and determine a historical state space based on the historical credit information and historical credit characteristics, wherein the historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information; a second determining subunit, used to obtain multiple credit types of credit services from the credit business system of financial institutions, and determine an action space based on the adjustment actions corresponding to the multiple credit types, wherein the action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan recall, and increasing the minimum collateral amount; and a third determining subunit, used to input the historical state space into a first neural network, calculate the probability distribution information of each adjustment action in the action space through the first neural network, and determine the predicted adjustment action based on the probability distribution information.

[0015] Furthermore, the training unit includes: a first processing subunit, used to input randomly selected arbitrary sample data into the second neural network to obtain the value evaluation result output by the second neural network; a first updating subunit, used to update the model parameters of the second neural network based on the loss function according to the preset value target and value evaluation result, to obtain the updated second neural network; a fourth determining subunit, used to determine gradient information based on the updated second neural network according to the policy gradient method; and a second updating subunit, used to update the model parameters of the first neural network according to the gradient information, and repeatedly extract sample data, iteratively updating the model parameters of the first neural network according to each extracted sample data, until the number of iterations reaches the preset number, to obtain the target decision model.

[0016] Furthermore, the aforementioned apparatus further includes: a fourth determining unit, used to input the target adjustment action into multiple agents after determining the target adjustment action for credit services to the target customer based on the evaluation result of the decision action and the decision action, to obtain the second observation result determined by the multiple agents based on the reward function; a second acquisition unit, used to collect the repayment behavior of the target customer from multiple channels every preset time interval; a first updating unit, used to update the target state space based on the second observation result, the target adjustment action, and the repayment behavior; and a second updating unit, used to update the model parameters of the second neural network and the model parameters of the target decision model based on the updated target state space, to obtain the updated second neural network and the updated target decision model.

[0017] Further, the first acquisition unit includes: a second processing subunit, used to preprocess the target customer information to obtain first customer information, wherein the preprocessing operation includes at least one of the following operations: data cleaning, data integration, data transformation, and noise reduction; a third processing subunit, used to fill in missing values ​​in the first customer information and modify outliers in the first customer information to obtain second customer information; an extraction subunit, used to extract the original credit features of the target customer from the second customer information through feature engineering, wherein the original credit features include at least: asset features, credit score features, repayment stability features, and financial environment features; and a fourth processing subunit, used to normalize the original credit features to obtain target credit features.

[0018] Furthermore, the first acquisition unit includes: a first acquisition subunit, used to acquire credit service information of target customers from the credit business system of financial institutions; a second acquisition subunit, used to acquire credit record information of target customers from third-party credit reporting agencies and acquire financial environment information through the Internet; a third acquisition subunit, used to acquire customer profile information, transaction behavior information and asset information of target customers from the customer information system of financial institutions; and a fifth determination subunit, used to determine target customer information based on credit service information, credit record information, financial environment information, customer profile information, transaction behavior information and asset information.

[0019] To achieve the above objectives, according to one aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the above-described data processing methods, and when executed by a processor, implements the steps of the data processing methods in various embodiments of this application.

[0020] To achieve the above objectives, according to one aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including stored computer instructions, wherein any of the above data processing methods are implemented when the computer instructions are executed by a processor.

[0021] To achieve the above objectives, according to one aspect of this application, an electronic device is provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors cause the one or more processors to implement any of the above data processing methods.

[0022] In this embodiment, by detecting that a target customer is applying for credit services and with the target customer's authorization, target customer information is collected from multiple channels after a preset time interval, and target credit features are extracted from the target customer information. A target state space is determined based on the target credit information and target credit features in the target customer information, whereby the target state space describes the set of all credit features of the target customer during the use of credit services. The target state space is input into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action. The target decision model is obtained by training a neural network using historical customer information and historical credit features based on a deep deterministic policy gradient algorithm. Based on the evaluation result and the decision action, a target adjustment action for credit services to the target customer is determined. This solves the technical problem that financial institutions, which typically rely on static data to provide services to customers, cannot timely capture dynamic customer information and respond to it, resulting in a relatively lagging response mechanism.

[0023] By regularly updating target customer information and extracting credit characteristics from multiple channels, a target state space is determined, enabling timely capture of dynamic customer information. This addresses, to some extent, the issue of financial institutions relying on static data and enhances their responsiveness. By inputting the target state space into a target decision-making model trained using a deep deterministic strategy gradient algorithm, the model can output decision actions and their evaluation results, improving the accuracy and speed of decision-making and reducing reliance on traditional static data evaluation. Furthermore, by using the evaluation results and decision actions to determine specific adjustments to credit services, the model dynamically optimizes credit service strategies, further achieving the technical effects of reducing credit risk and improving service efficiency for financial institutions. Attached Figure Description

[0024] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0025] Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a data processing method according to Embodiment 1 of this application;

[0026] Figure 2 This is a flowchart of an optional data processing method provided according to Embodiment 1 of this application;

[0027] Figure 3 This is a flowchart of an optional credit service system provided according to Embodiment 1 of this application;

[0028] Figure 4 This is a schematic diagram of a data processing apparatus provided according to Embodiment 2 of this application;

[0029] Figure 5 This is a schematic diagram of a data processing electronic device provided according to Embodiment 3 of this application. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] It should be noted that the processing methods, apparatus, storage media, and electronic devices specified in this application can be used in the financial technology field to improve the response timeliness of financial institutions in the process of providing credit services to customers. They can also be used in any field other than the financial technology field. The application fields of the processing methods, apparatus, storage media, and electronic devices specified in this application are not limited.

[0032] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, collected data, used data, generated data, processed data, etc.) and the data (including but not limited to data used for analysis, stored data, displayed data, collected information, used information, generated information, processed information, etc.) are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations, providing users with corresponding operation entry points for users to choose to agree to or refuse automated decision results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0033] Example 1

[0034] According to an embodiment of this application, a method embodiment for data processing is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0035] The method embodiment provided in Embodiment 1 of this application can be executed in a mobile terminal, computer terminal or similar computing device. Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a data processing method, according to Embodiment 1 of this application. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), 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 a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0036] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0037] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the data processing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the aforementioned data processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0038] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0039] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0040] Under the aforementioned operating environment, this application provides the following: Figure 2 The data processing method shown. Figure 2 This is a flowchart of an optional data processing method provided according to Embodiment 1 of this application.

[0041] Step S201: Upon detecting that a target customer is applying for credit services and with the target customer's authorization, target customer information is collected from multiple channels after a preset time interval, and target credit features are extracted from the target customer information.

[0042] The implementing entity in this embodiment 1 can be a credit service system. The specific function of this system is that it can proactively identify the behavior of target customers after applying for credit services (provided that explicit authorization is obtained from the customer), extract key credit features from this information using deep learning technology, and thus achieve dynamic adjustment of credit services.

[0043] In this embodiment 1, once it is identified that a target customer has already applied for credit services and has given prior consent to information collection, the credit service system will collect relevant customer information from the target customer at set time intervals. These sources may include, but are not limited to, internal bank databases, credit reports, public records, and other legitimate data sources. Subsequently, through a series of preset data processing algorithms, key credit characteristics, such as financial status and credit history, are extracted from the collected customer information for use in subsequent credit service decision-making processes. This mechanism ensures the timeliness and comprehensiveness of the credit decision-making basis.

[0044] Step S202: Determine the target state space based on the target credit information and target credit characteristics in the target customer information. The target state space is used to describe the set of all credit characteristics of the target customer during the use of credit services.

[0045] In this embodiment 1, the target state space is used to represent the set of credit characteristics of a target customer during the credit service period. Specifically, the credit service system will filter out credit-related data from the target customer information, and through data preprocessing, extract the features that influence credit service decisions, thus constructing the target state space, which represents the multi-dimensional state of the customer during the use of credit services. The construction of this space provides the necessary information foundation for the subsequent dynamic adjustment of credit services using reinforcement learning algorithms, ensuring that decisions are based on the customer's latest and most comprehensive credit behavior.

[0046] Step S203: Input the target state space into the target decision model to obtain the decision action and the evaluation result of the decision action output by the target decision model. The target decision model is obtained by training the neural network with historical customer information and historical credit features based on the deep deterministic policy gradient algorithm.

[0047] In this embodiment 1, the target decision model receives information from the target state space as input. The target decision model is based on the Deep Deterministic Policy Gradient (DDPG) algorithm, which has been previously trained using a large amount of historical customer data and corresponding credit characteristics. After the model operates, it outputs two parts: first, a decision action, such as adjusting credit limits, interest rates, or taking other credit service measures; and second, an evaluation result of the decision action, i.e., a prediction of the decision's effectiveness, measuring its contribution to reducing credit risk and improving credit service efficiency. This model design ensures that decision-making is both based on historical experience and adapted to the current situation, realizing the intelligent and dynamic nature of financial institutions' credit services.

[0048] Step S204: Based on the evaluation results of the decision-making actions and the decision-making actions, determine the target adjustment actions for credit services to the target customers.

[0049] In this embodiment 1, after receiving the output of the target decision model—namely, the decision action and its evaluation results—the credit service system determines specific target adjustment actions aimed at optimizing credit services for the target customer. These adjustments may include, but are not limited to, changing the loan interest rate, adjusting the loan amount, or updating collateral conditions. The entire process demonstrates the system's ability to make targeted adjustments based on the recommendations of the intelligent model, thereby achieving more refined credit service decisions.

[0050] Optionally, in the data processing method provided in Embodiment 1 of this application, before inputting the target state space into the target decision model, the method further includes: obtaining historical customer information of historical customers and extracting historical credit features from the historical customer information, wherein historical customers refer to customers who have applied for credit services within a historical time period; generating predictive adjustment actions based on the historical state space and action space through a first neural network, wherein the historical state space is used to describe the set of all credit features of historical customers during the use of credit services, and the action space is used to describe the set of various adjustment actions taken by financial institutions for credit services based on the historical state space; inputting the predictive adjustment actions into multiple agents to obtain a first observation result determined by the multiple agents according to the reward function, and updating the historical state space according to the first observation result, wherein the multiple agents are determined based on various credit types of credit services, and the reward function is determined based on the repayment behavior of historical customers; randomly sampling sample data multiple times from the updated historical state space, inputting the sample data extracted each time into a second neural network, and iteratively training the first neural network according to the value evaluation result output by the second neural network to obtain the target decision model.

[0051] In this embodiment 1, in order to construct a high-precision credit risk assessment and decision-making model, the credit risk management process is modeled as a Markov decision process. The reinforcement learning model adopts the deep deterministic policy gradient algorithm DDPG to reduce the bank's credit risk losses and improve risk management efficiency by adjusting the credit granting strategy.

[0052] First, relevant credit information of historical customers is retrieved from the database. These historical customers specifically refer to the group of customers who have received credit services within a specific period in the past. Then, historical credit characteristics are extracted from this information. The historical state space refers to the representation of all credit characteristics of historical customers within the credit cycle, while the action space contains multiple actions that financial institutions may take to adjust credit services, such as adjusting interest rates or changing loan amounts. Next, using a first neural network (or actor network), predictive adjustment actions for different credit states are generated based on the historical state space and action space. For example, predicting an interest rate adjustment to 5% or an increase in the loan amount to 1 million yuan. These predicted actions are assigned to multiple agents, each simulating a specific type of credit business. Based on a reward function set according to historical repayment behavior, the agents evaluate the impact of the predicted adjustment actions, forming initial observations. These observations update the historical state space, incorporating more real-world feedback and enhancing the model's learning ability. Finally, samples are extracted from the updated historical state space and input into the second neural network (or Critic network). This network is responsible for evaluating the value of the actions generated by the first neural network (i.e., the value evaluation results mentioned above). The value evaluation results are fed back to the first neural network to guide its iterative optimization until the model converges, resulting in the final target decision model.

[0053] Through the above steps, a deep learning model based on historical data and real-time feedback was constructed. This model possesses dynamic adaptability and self-optimization capabilities, enabling it to intelligently predict optimal credit service adjustments based on historical repayment behavior and changes in the current credit environment. Specifically, by training the target decision-making model, the decision-making quality and response speed of financial institutions in credit services are improved. Furthermore, by simulating different credit scenarios through intelligent agents, the model's ability to handle diverse credit demands is enhanced, achieving refined risk management. In addition, the feedback mechanism based on the reward function makes the model more closely aligned with actual business needs during training.

[0054] Optionally, in the data processing method provided in Embodiment 1 of this application, a predictive adjustment action is generated based on a historical state space and an action space using a first neural network. This includes: extracting historical credit information related to the credit services of historical customers from historical customer information, and determining a historical state space based on the historical credit information and historical credit characteristics. The historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information; obtaining multiple credit types of credit services from the credit business system of financial institutions, and determining an action space based on the adjustment actions corresponding to the multiple credit types. The action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan repayment, and increasing the minimum collateral amount; inputting the historical state space into the first neural network, calculating the probability distribution information of each adjustment action in the action space using the first neural network, and determining the predictive adjustment action based on the probability distribution information.

[0055] In this embodiment 1, to establish a decision-making model, key information related to credit services is first extracted from historical customer information, such as loan interest rates (e.g., 5%), loan amounts (e.g., 800,000 yuan), and minimum collateral requirements (e.g., 200,000 yuan). Then, based on the extracted historical credit information and characteristics, a historical state space is constructed as the basis for model learning. Next, an action space encompassing various possible adjustment strategies, such as adjusting loan interest rates, changing loan amounts, early loan recall, or increasing the minimum collateral requirement, is derived from the actual credit operations of financial institutions. Finally, the historical state space is input into a first neural network, which calculates the execution probability of each adjustment action in the action space, thereby determining the predicted adjustment action.

[0056] By quantifying historical credit characteristics and information, the model training is grounded in real business data, enhancing its practical applicability and predictive accuracy. Secondly, defining a detailed action space ensures the decision-making model covers all possible options for financial institutions in credit adjustments, improving the flexibility and comprehensiveness of decision-making. Thirdly, utilizing a first neural network to calculate action probability distribution information automates the process of automatically selecting high-value options from numerous potential adjustment strategies, reducing human interference and improving decision-making efficiency. In summary, these steps significantly enhance the intelligence level of financial institutions' decision-making in the credit service field.

[0057] Optionally, in the data processing method provided in Embodiment 1 of this application, the first neural network is iteratively trained based on the value assessment result output by the second neural network to obtain a target decision model. This includes: inputting randomly selected arbitrary sample data into the second neural network to obtain the value assessment result output by the second neural network; updating the model parameters of the second neural network based on the loss function according to the preset value target and the value assessment result to obtain the updated second neural network; determining gradient information based on the updated second neural network using the policy gradient method; updating the model parameters of the first neural network based on the gradient information, and repeatedly extracting sample data, iteratively updating the model parameters of the first neural network based on each extracted sample data, until the number of iterations reaches the preset number, thereby obtaining the target decision model.

[0058] In this embodiment 1, to optimize the decision-making model and predict credit business adjustment strategies, a set of sample data randomly selected from the historical state space is first input into a second neural network. This network is responsible for evaluating the credit business value corresponding to the adjustment actions proposed by the first neural network, where the value assessment result can be expressed as the expected reduction in credit risk after adjustment. Then, the preset value target, i.e., the maximum credit risk that the financial institution expects to achieve, is compared with the actual value assessment result output by the second neural network, and the difference is calculated. This difference is used as a feedback signal to adjust the model parameters of the second neural network according to the loss function, in order to reduce the deviation between the predicted value and the target value and improve the accuracy of value assessment. Next, using the strategy gradient method, the gradient information of the model parameters of the first neural network is calculated based on the adjusted second neural network, aiming to guide the optimization strategy of the first neural network to match the optimized value assessment standard. Finally, the model parameters of the first neural network are updated based on the calculated gradient information. This process is repeated, with each update based on randomly selected sample data, until the number of iterations of the model parameters reaches a preset upper limit, ensuring that the decision-making model fully learns and optimizes, reaches the expected level of credit business decision-making, and produces the target decision-making model.

[0059] Through the above steps, the following technical effects were achieved: First, through iterative training of deep neural networks, the model can understand and learn the actual impact of various adjustment actions on business objectives in a complex credit environment, thereby making more accurate predictions of credit strategy adjustments; Second, the adaptability of the decision-making model was improved. Through continuous analysis of historical data and dynamic adjustment of model parameters, the model can continuously optimize itself, reducing the risk of decision-making errors; Third, through dual-channel iteration of value assessment and strategy gradient optimization, not only was the effectiveness of model training ensured, but the stability of the model's prediction of future credit conditions was also strengthened; Fourth, the efficiency of credit services for financial institutions was further improved, the reliance on traditional static data analysis was reduced, and the reaction speed and decision-making quality in the face of dynamic information were enhanced.

[0060] Optionally, in the data processing method provided in Embodiment 1 of this application, after determining the target adjustment action for credit services to the target customer based on the evaluation result of the decision action and the decision action, the method further includes: inputting the target adjustment action into multiple agents to obtain the second observation result determined by the multiple agents according to the reward function; collecting the repayment behavior of the target customer from multiple channels after a preset time interval; updating the target state space based on the second observation result, the target adjustment action and the repayment behavior; and updating the model parameters of the second neural network and the model parameters of the target decision model based on the updated target state space to obtain the updated second neural network and the updated target decision model.

[0061] In this embodiment 1, to continuously optimize the credit decision-making model, the determined target adjustment action is first input into multiple agents. These agents are determined based on different types of credit business. They return a second observation of the effect of the target adjustment action based on a reward function, which considers loan recovery and default costs. For example, the second observation might reflect the percentage decrease in the default rate under a specific adjustment strategy. Then, after each pre-set time period, dynamically updated repayment behavior data of target customers is collected from multiple channels, including but not limited to: on-time repayment, deferred payment, or default records, reflecting real-time changes in the customer's credit status. Next, the target state space is iteratively updated by combining the second observation, the target adjustment action, and the latest collected repayment behavior information to ensure that the state space accurately reflects the true risk status of the current credit service. Finally, based on the updated target state space, the parameters of the second neural network and the target decision-making model are adjusted. Machine learning algorithms are used to continuously correct the model, making it more aligned with actual business needs and improving the accuracy and timeliness of decision-making.

[0062] By constructing a closed-loop learning and feedback mechanism, the model can continuously accumulate experience in actual operation, optimize credit service adjustments, and improve the model's adaptability and decision quality. Through agent simulation and reward function settings, the model can automatically identify adjustment actions that meet the needs of credit business, avoiding the uncertainty and reduced processing efficiency caused by blind decision-making. Through the above steps, the dynamic adaptability and intelligent optimization characteristics of the credit decision-making model are significantly enhanced, realizing dynamic optimization of credit decisions. Each round of model updates is based on the latest customer repayment behavior, ensuring the timeliness and forward-looking nature of the decision-making model and reducing the lag in credit service response caused by information delays.

[0063] Optionally, in the data processing method provided in Embodiment 1 of this application, extracting target credit features from target customer information includes: performing preprocessing operations on the target customer information to obtain first customer information, wherein the preprocessing operations include at least one of the following operations: data cleaning, data integration, data transformation, and noise reduction; filling in missing values ​​in the first customer information and modifying outliers in the first customer information to obtain second customer information; extracting the original credit features of the target customer from the second customer information through feature engineering, wherein the original credit features include at least: asset features, credit score features, repayment stability features, and financial environment features; and normalizing the original credit features to obtain target credit features.

[0064] In this embodiment 1, to ensure the accuracy and reliability of credit risk assessment, a series of preprocessing operations are first performed on the collected target customer information, including data cleaning, integration, transformation, and noise reduction, to eliminate redundancy and irrelevant items and improve data quality. Then, for missing values ​​found in the processed first customer information, statistical methods or machine learning algorithms are used to fill them in appropriately, while outliers are corrected to obtain more complete and accurate second customer information. Next, feature engineering techniques are used to extract key features reflecting customer creditworthiness from the second customer information, such as asset characteristics, credit score characteristics, repayment stability characteristics (e.g., the percentage of monthly on-time repayments over the past few years), and financial environment characteristics. These features constitute the original credit features. Finally, the original credit features are normalized to ensure that all features are compared on the same scale, resulting in standardized target credit features, providing a data foundation for subsequent model training and risk assessment.

[0065] Preprocessing eliminated noise and inconsistencies in the original data, improving data quality and providing a solid foundation for subsequent analysis and modeling. Imputation of missing values ​​and correction of outliers enabled the model to be trained on a complete dataset, avoiding evaluation bias caused by incomplete data. Feature engineering successfully extracted features highly correlated with loan customer creditworthiness, providing multi-dimensional information for the deep learning model and helping it to more comprehensively understand multi-dimensional credit characteristics. Normalization of the target credit features resolved potential issues of inconsistent dimensions and large differences in numerical ranges in the original feature data, enabling the model to efficiently process and analyze these features, further improving the responsiveness of credit service decisions.

[0066] Optionally, in the data processing method provided in Embodiment 1 of this application, target customer information of target customers is collected from multiple channels, including: obtaining credit service information of target customers from the credit business system of financial institutions; obtaining credit record information of target customers from third-party credit reporting agencies and obtaining financial environment information through the Internet; obtaining customer profile information, transaction behavior information and asset information of target customers from the customer information system of financial institutions; and determining target customer information based on credit service information, credit record information, financial environment information, customer profile information, transaction behavior information and asset information.

[0067] In this Example 1, to comprehensively understand the risk profile of the target customer, the first step is to extract the target customer's credit service information from the financial institution's credit business system. This includes, but is not limited to, current loan balance, repayment history, and loan type, as this information directly relates to the target customer's creditworthiness and repayment ability. Next, credit record information, such as the number of overdue payments and historical default rates, is obtained from third-party credit reporting agencies. Simultaneously, the latest financial environment information is collected using internet platforms. Furthermore, the target customer's non-credit information is obtained from the financial institution's customer information system, including customer profiles (age, occupation), transaction behavior (transaction frequency, transaction amount), and asset status. Finally, by integrating the various information obtained from these multiple sources—including details of credit services, credit history, financial environment, personal background, transaction patterns, and asset details—a comprehensive and accurate set of target customer information is determined, providing a solid data foundation for subsequent credit risk assessment.

[0068] In this Example 1, the collection of all target customer information complies with financial industry data security standards and financial institution customer authorization management requirements. Credit service information obtained from the financial institution's credit business system is limited to data explicitly authorized by the customer during the transaction; credit record information retrieved from third-party credit reporting agencies is obtained through legal and compliant interface channels, and data sharing is explicitly agreed to by the customer in the authorization agreement; financial environment information obtained through the internet uses only publicly available, non-personally associated macro data sources, without involving the collection of individual privacy; profile information, transaction behavior, and asset information extracted from the customer information system are all based on the usage permissions granted by the customer in the service agreement, and the data extraction process implements anonymization processing and the principle of minimum necessity. All data sources have signed data usage agreements, the collection behavior is confirmed through the customer's informed consent mechanism, and the system records a complete authorization log to ensure that the entire information acquisition process is traceable, auditable, and compliant.

[0069] By constructing a multi-layered, multi-faceted target customer information system, comprehensive and accurate information input for the target decision-making model is ensured, avoiding biases in credit service decisions caused by incomplete or omitted information. Synchronous data acquisition from internal financial institution systems and external third-party platforms establishes a complementary and verification mechanism between internal and external information, improving the authenticity and reliability of the information. In-depth analysis of target customer information identifies dynamic interaction characteristics that are difficult for traditional models to capture, such as non-linear transaction patterns, abnormal operation sequences, and coupled changes in device and account behavior. This allows financial institutions to construct more comprehensive and granular customer behavior characteristics, aiding in credit service decisions.

[0070] Optionally, in this embodiment 1, Figure 3 This is a flowchart illustrating an optional credit service system provided according to Embodiment 1 of this application. Figure 3As shown, firstly, relevant information about target customers is collected from multiple channels, including internal systems of financial institutions, third-party credit reporting agencies, and internet data providers. This information covers the target customers' financial status, credit history, transaction behavior, and external environment, providing a rich and comprehensive data foundation for subsequent model building. Secondly, a reinforcement learning model is established, i.e., an intelligent agent is constructed. This agent, based on reinforcement learning, takes the real-time customer behavior flow as its state input and configurable service intervention actions as its spatial output. Its learning objective is to achieve rapid identification and automated response loop for changes in customer behavior patterns. Finally, the DDPG algorithm is used for model training because it is suitable for solving decision-making problems in a continuous action space. After the model training converges, the obtained target decision model is embedded into the financial institution's credit service system and runs as a real-time decision-making component. When a customer initiates a credit service request or when customer behavior data is updated, the model dynamically analyzes the latest behavioral characteristics and outputs corresponding credit service decision instructions. This effectively solves the technical defects of traditional service systems, such as response lag and slow intervention, caused by reliance on static data, until the service process terminates naturally or the customer's behavior returns to normal.

[0071] It should be noted that the steps shown in the flowchart in 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 may be executed in a different order than that shown here.

[0072] In summary, the data processing method provided in this application, upon detecting that a target customer is applying for credit services and with the target customer's authorization, collects target customer information from multiple channels after a preset time interval and extracts target credit features from the target customer information; determines a target state space based on the target credit information and target credit features in the target customer information, wherein the target state space describes the set of all credit features of the target customer during the use of credit services; inputs the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action, wherein the target decision model is obtained by training a neural network based on a deep deterministic policy gradient algorithm using historical customer information and historical credit features; and determines the target adjustment action for credit services for the target customer based on the evaluation result and the decision action. This solves the problem in related technologies where financial institutions typically rely on static data to provide services to customers, making it impossible to capture dynamic customer information in a timely manner and respond to such information promptly, resulting in a relatively lagging response mechanism for financial institutions.

[0073] By regularly updating target customer information and extracting credit characteristics from multiple channels, a target state space is determined, enabling timely capture of dynamic customer information. This addresses, to some extent, the issue of financial institutions relying on static data and enhances their responsiveness. By inputting the target state space into a target decision-making model trained using a deep deterministic strategy gradient algorithm, the model can output decision actions and their evaluation results, improving the accuracy and speed of decision-making and reducing reliance on traditional static data evaluation. Furthermore, by using the evaluation results and decision actions to determine specific adjustments to credit services, the model dynamically optimizes credit service strategies, further achieving the technical effects of reducing credit risk and improving service efficiency for financial institutions.

[0074] Example 2

[0075] This application also provides a data processing apparatus. It should be noted that the data processing apparatus of this application can be used to execute the data processing method provided in this application. The data processing apparatus provided in this application will be described below.

[0076] According to an embodiment of this application, an apparatus for implementing the above-described data processing method is also provided. Figure 4 This is a schematic diagram of a data processing apparatus provided according to Embodiment 2 of this application. Figure 4 As shown, the device includes: a first acquisition unit 401, a first determination unit 402, a decision unit 403, and a second determination unit 404.

[0077] Specifically, the first collection unit 401 is used to collect target customer information from multiple channels after a preset time interval when it is detected that the target customer is applying for credit services and with the authorization of the target customer, and to extract target credit features from the target customer information.

[0078] The first determining unit 402 is used to determine the target state space based on the target credit information and target credit characteristics in the target customer information, wherein the target state space is used to describe the set of all credit characteristics of the target customer during the use of credit services.

[0079] The decision unit 403 is used to input the target state space into the target decision model to obtain the decision action and the evaluation result of the decision action output by the target decision model. The target decision model is obtained by training the neural network with historical customer information and historical credit features based on the deep deterministic policy gradient algorithm.

[0080] The second determining unit 404 is used to determine the target adjustment action for credit services to the target customer based on the evaluation results and decision action.

[0081] The data processing apparatus provided in this application embodiment, through a first acquisition unit 401, upon detecting that a target customer is applying for credit services and with the authorization of the target customer, collects target customer information from multiple channels every preset time interval and extracts target credit features from the target customer information; a first determination unit 402 determines a target state space based on the target credit information and target credit features in the target customer information, wherein the target state space is used to describe the set of all credit features of the target customer during the use of credit services; a decision unit 403 inputs the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action, wherein the target decision model is obtained by training a neural network based on a deep deterministic strategy gradient algorithm using historical customer information and historical credit features; a second determination unit 404 determines the target adjustment action for the credit service for the target customer based on the evaluation result and the decision action, thereby solving the problem in related technologies where financial institutions usually rely on static data to provide services to customers, making it impossible to capture dynamic customer information in a timely manner and respond to such information in a timely manner, resulting in a relatively lagging response mechanism of financial institutions.

[0082] By regularly updating target customer information and extracting credit characteristics from multiple channels, a target state space is determined, enabling timely capture of dynamic customer information. This addresses, to some extent, the issue of financial institutions relying on static data and enhances their responsiveness. By inputting the target state space into a target decision-making model trained using a deep deterministic strategy gradient algorithm, the model can output decision actions and their evaluation results, improving the accuracy and speed of decision-making and reducing reliance on traditional static data evaluation. Furthermore, by using the evaluation results and decision actions to determine specific adjustments to credit services, the model dynamically optimizes credit service strategies, further achieving the technical effects of reducing credit risk and improving service efficiency for financial institutions.

[0083] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the apparatus further includes: an extraction unit, configured to acquire historical customer information of historical customers and extract historical credit features from the historical customer information before inputting the target state space into the target decision model, wherein historical customers refer to customers who have applied for credit services within a historical time period; a generation unit, configured to generate predictive adjustment actions based on the historical state space and action space through a first neural network, wherein the historical state space is used to describe the set of all credit features of historical customers during the use of credit services, and the action space is used to describe the set of various adjustment actions taken by financial institutions for credit services based on the historical state space; a third determination unit, configured to input the predictive adjustment actions into multiple agents to obtain a first observation result determined by the multiple agents according to the reward function, and update the historical state space according to the first observation result, wherein the multiple agents are determined based on various credit types of credit services, and the reward function is determined based on the repayment behavior of historical customers; and a training unit, configured to randomly extract sample data multiple times from the updated historical state space, input the sample data extracted each time into a second neural network, and iteratively train the first neural network according to the value evaluation result output by the second neural network to obtain the target decision model.

[0084] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the aforementioned generation unit includes: a first determining subunit, configured to extract historical credit information related to credit services for historical customers from historical customer information, and determine a historical state space based on the historical credit information and historical credit characteristics, wherein the historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information; a second determining subunit, configured to obtain multiple credit types of credit services from the credit business system of financial institutions, and determine an action space based on the adjustment actions corresponding to the multiple credit types, wherein the action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan repayment, and increasing the minimum collateral amount; and a third determining subunit, configured to input the historical state space into a first neural network, calculate the probability distribution information of each adjustment action in the action space through the first neural network, and determine the predicted adjustment action based on the probability distribution information.

[0085] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the training unit includes: a first processing subunit, used to input randomly selected arbitrary sample data into a second neural network to obtain the value assessment result output by the second neural network; a first updating subunit, used to update the model parameters of the second neural network based on a loss function according to a preset value target and the value assessment result to obtain an updated second neural network; a fourth determining subunit, used to determine gradient information based on a policy gradient method according to the updated second neural network; and a second updating subunit, used to update the model parameters of the first neural network according to the gradient information, and repeatedly extract sample data, iteratively updating the model parameters of the first neural network according to each extracted sample data until the number of iterations reaches a preset number to obtain a target decision model.

[0086] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the apparatus further includes: a fourth determining unit, configured to, after determining the target adjustment action for credit services to the target customer based on the evaluation result of the decision action and the decision action, input the target adjustment action into multiple agents to obtain a second observation result determined by the multiple agents based on the reward function; a second acquisition unit, configured to acquire the repayment behavior of the target customer from multiple channels every preset time interval; a first updating unit, configured to update the target state space based on the second observation result, the target adjustment action, and the repayment behavior; and a second updating unit, configured to update the model parameters of the second neural network and the model parameters of the target decision model based on the updated target state space to obtain the updated second neural network and the updated target decision model.

[0087] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the first acquisition unit 401 includes: a second processing subunit, used to perform preprocessing operations on the target customer information to obtain first customer information, wherein the preprocessing operations include at least one of the following operations: data cleaning, data integration, data conversion, and noise reduction; a third processing subunit, used to fill in missing values ​​in the first customer information and modify outliers in the first customer information to obtain second customer information; an extraction subunit, used to extract the original credit features of the target customer from the second customer information through feature engineering, wherein the original credit features include at least: asset features, credit score features, repayment stability features, and financial environment features; and a fourth processing subunit, used to normalize the original credit features to obtain target credit features.

[0088] Optionally, in the data processing apparatus provided in Embodiment 2 of this application, the first acquisition unit 401 includes: a first acquisition subunit, used to acquire credit service information of target customers from the credit business system of financial institutions; a second acquisition subunit, used to acquire credit record information of target customers from third-party credit reporting agencies and acquire financial environment information through the Internet; a third acquisition subunit, used to acquire customer profile information, transaction behavior information and asset information of target customers from the customer information system of financial institutions; and a fifth determination subunit, used to determine target customer information based on credit service information, credit record information, financial environment information, customer profile information, transaction behavior information and asset information.

[0089] It should be noted that the first acquisition unit 401, the first determination unit 402, the decision unit 403, and the second determination unit 404 mentioned above correspond to steps S201 to S204 in Embodiment 1. The two modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0090] Example 3

[0091] Embodiments of this application may provide an electronic device. Figure 5 This is a schematic diagram of a data processing electronic device according to Embodiment 3 of this application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0092] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] The processor can access information and applications stored in memory via a transmission device to execute the following steps: upon detecting that a target customer is applying for credit services and with the target customer's authorization, collect target customer information from multiple channels after a preset time interval, and extract target credit features from the target customer information; determine a target state space based on the target credit information and target credit features in the target customer information, wherein the target state space describes the set of all credit features of the target customer during the use of credit services; input the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action, wherein the target decision model is obtained by training a neural network based on a deep deterministic policy gradient algorithm using historical customer information and historical credit features; and determine the target adjustment action for the credit service for the target customer based on the evaluation result and the decision action.

[0094] The processor can invoke information and applications stored in the memory via a transmission device to execute the following steps: Before inputting the target state space into the target decision model, the above method further includes: obtaining historical customer information of historical customers and extracting historical credit features from the historical customer information, wherein historical customers refer to customers who have applied for credit services within a historical time period; generating predictive adjustment actions based on the historical state space and action space through a first neural network, wherein the historical state space is used to describe the set of all credit features of historical customers during the use of credit services, and the action space is used to describe the set of various adjustment actions taken by financial institutions for credit services based on the historical state space; inputting the predictive adjustment actions into multiple agents to obtain the first observation results determined by the multiple agents according to the reward function, and updating the historical state space according to the first observation results, wherein the multiple agents are determined according to the various credit types of credit services, and the reward function is determined according to the repayment behavior of historical customers; randomly sampling sample data multiple times from the updated historical state space, inputting the sample data of each sampling into a second neural network, and iteratively training the first neural network according to the value evaluation results output by the second neural network to obtain the target decision model.

[0095] The processor can invoke information and applications stored in the memory via a transmission device to execute the following steps: generating predictive adjustment actions based on a historical state space and an action space using a first neural network, including: extracting historical credit information related to credit services for historical customers from historical customer information, and determining a historical state space based on the historical credit information and historical credit characteristics, wherein the historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information; obtaining multiple credit types of credit services from the credit business system of financial institutions, and determining an action space based on the adjustment actions corresponding to the multiple credit types, wherein the action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan recall, and increasing the minimum collateral amount; inputting the historical state space into the first neural network, calculating the probability distribution information of each adjustment action in the action space through the first neural network, and determining the predictive adjustment action based on the probability distribution information.

[0096] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: iteratively training the first neural network based on the value assessment results output by the second neural network to obtain a target decision model, including: inputting randomly selected arbitrary sample data into the second neural network to obtain the value assessment results output by the second neural network; updating the model parameters of the second neural network based on the loss function according to the preset value target and the value assessment results to obtain the updated second neural network; determining gradient information based on the updated second neural network using the policy gradient method; updating the model parameters of the first neural network based on the gradient information, and repeatedly extracting sample data, iteratively updating the model parameters of the first neural network based on each extracted sample data until the number of iterations reaches the preset number, thereby obtaining the target decision model.

[0097] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: After determining the target adjustment action for credit services to the target customer based on the evaluation results and decision actions, the method further includes: inputting the target adjustment action into multiple agents to obtain second observation results determined by the agents based on a reward function; collecting repayment behavior of the target customer from multiple channels after a preset time interval; updating the target state space based on the second observation results, the target adjustment action, and the repayment behavior; and updating the model parameters of the second neural network and the target decision model based on the updated target state space to obtain the updated second neural network and the updated target decision model.

[0098] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: extracting target credit features from target customer information, including: performing preprocessing operations on the target customer information to obtain first customer information, wherein the preprocessing operations include at least one of the following operations: data cleaning, data integration, data transformation, and noise reduction; filling in missing values ​​in the first customer information and modifying outliers in the first customer information to obtain second customer information; extracting the original credit features of the target customer from the second customer information through feature engineering, wherein the original credit features include at least: asset features, credit score features, repayment stability features, and financial environment features; and normalizing the original credit features to obtain target credit features.

[0099] The processor can access information and applications stored in memory via a transmission device to perform the following steps: collecting target customer information from multiple channels, including: obtaining target customer credit service information from the credit business system of financial institutions; obtaining target customer credit record information from third-party credit reporting agencies and obtaining financial environment information via the Internet; obtaining target customer profile information, transaction behavior information, and asset information from the customer information system of financial institutions; and determining target customer information based on credit service information, credit record information, financial environment information, customer profile information, transaction behavior information, and asset information.

[0100] This application provides a data processing method. By detecting that a target customer is using credit services and with the customer's authorization, target customer information is collected from multiple channels at preset intervals, and target credit features are extracted from this information. A target state space is determined based on the target credit information and features, describing the set of all credit features of the target customer during their use of credit services. This target state space is then input into a target decision model to obtain the output decision action and its evaluation result. The target decision model is trained on a neural network using historical customer information and historical credit features based on a deep deterministic policy gradient algorithm. Based on the evaluation result and the decision action, a target adjustment action for the credit service to the target customer is determined. This method addresses the technical problem that financial institutions, which typically rely on static data to provide services, cannot promptly capture and respond to dynamic customer information, resulting in a delayed response mechanism.

[0101] By regularly updating target customer information and extracting credit characteristics from multiple channels, a target state space is determined, enabling timely capture of dynamic customer information. This addresses, to some extent, the issue of financial institutions relying on static data and enhances their responsiveness. By inputting the target state space into a target decision-making model trained using a deep deterministic strategy gradient algorithm, the model can output decision actions and their evaluation results, improving the accuracy and speed of decision-making and reducing reliance on traditional static data evaluation. Furthermore, by using the evaluation results and decision actions to determine specific adjustments to credit services, the model dynamically optimizes credit service strategies, further achieving the technical effects of reducing credit risk and improving service efficiency for financial institutions.

[0102] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0103] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0104] Example 4

[0105] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the data processing method provided in Embodiment 1.

[0106] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0107] This application also provides a computer program product, which, when executed on a data processing device, is a program adapted to perform data processing method steps.

[0108] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0109] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0111] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0112] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0113] If the integrated unit is implemented as 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 this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0114] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A data processing method, characterized in that, include: Upon detecting that a target customer is applying for credit services, and with the target customer's authorization, target customer information of the target customer is collected from multiple channels every preset time interval, and target credit features are extracted from the target customer information; A target state space is determined based on the target credit information and the target credit features in the target customer information, wherein the target state space is used to describe the set of all credit features of the target customer during the use of the credit service; The target state space is input into the target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action. The target decision model is obtained by training a neural network with historical customer information and historical credit features based on a deep deterministic policy gradient algorithm. Based on the evaluation results of the decision-making actions and the decision-making actions themselves, target adjustment actions for credit services to the target customer are determined.

2. The method according to claim 1, characterized in that, Before inputting the target state space into the target decision model, the method further includes: Obtain the historical customer information of historical customers, and extract the historical credit characteristics from the historical customer information, wherein the historical customer refers to a customer who has applied for credit services within a historical time period; The first neural network generates predictive adjustment actions based on the historical state space and the action space. The historical state space is used to describe the set of all credit characteristics of the historical customer during the use of the credit service, and the action space is used to describe the set of various adjustment actions taken by the financial institution for the credit service based on the historical state space. The prediction adjustment action is input into multiple agents to obtain a first observation result determined by the multiple agents according to the reward function, and the historical state space is updated according to the first observation result. The multiple agents are determined based on the various credit types of the credit service, and the reward function is determined based on the repayment behavior of the historical customers. Sample data is randomly extracted multiple times from the updated historical state space. Each extracted sample data is input into the second neural network. The first neural network is iteratively trained based on the value evaluation results output by the second neural network to obtain the target decision model.

3. The method according to claim 2, characterized in that, The first neural network generates predictive and adjusted actions based on the historical state space and action space, including: Historical credit information related to the credit services of the historical customers is extracted from the historical customer information, and the historical state space is determined based on the historical credit information and the historical credit characteristics. The historical credit information includes at least one of the following: loan interest rate information, loan amount information, and minimum collateral amount information. The system obtains multiple credit types of the credit service from the credit business system of the financial institution, and determines the action space based on the adjustment actions corresponding to the multiple credit types. The action space includes at least one of the following: adjusting the loan interest rate, adjusting the loan amount, early loan recall, and increasing the minimum amount of collateral. The historical state space is input into the first neural network, and the probability distribution information of each adjustment action in the action space is calculated by the first neural network. The predicted adjustment action is determined based on the probability distribution information.

4. The method according to claim 2, characterized in that, The first neural network is iteratively trained based on the value assessment results output by the second neural network to obtain the target decision model, including: The randomly selected sample data is input into the second neural network to obtain the value assessment result output by the second neural network; The model parameters of the second neural network are updated based on the loss function according to the preset value target and the value assessment result, resulting in the updated second neural network. The policy gradient method determines gradient information based on the updated second neural network. The model parameters of the first neural network are updated based on the gradient information, and sample data is repeatedly extracted. The model parameters of the first neural network are iteratively updated based on the sample data extracted each time until the number of iterations reaches a preset number, thereby obtaining the target decision model.

5. The method according to claim 2, characterized in that, After determining the target adjustment actions for credit services to the target customer based on the evaluation results of the decision-making actions and the decision-making actions, the method further includes: The target adjustment action is input into the plurality of agents to obtain the second observation result determined by the plurality of agents based on the reward function; The repayment behavior of the target customer is collected from the multiple channels after a preset time period; The target state space is updated based on the second observation result, the target adjustment action, and the repayment behavior; The model parameters of the second neural network and the model parameters of the target decision model are updated based on the updated target state space to obtain the updated second neural network and the updated target decision model.

6. The method according to claim 1, characterized in that, Extracting target credit features from the target customer information includes: The target customer information is preprocessed to obtain first customer information, wherein the preprocessing operation includes at least one of the following operations: data cleaning, data integration, data transformation, and noise reduction; Missing values ​​in the first customer information are filled in, and outliers in the first customer information are modified to obtain the second customer information; The original credit characteristics of the target customer are extracted from the second customer information through feature engineering, wherein the original credit characteristics include at least: asset characteristics, credit score characteristics, repayment stability characteristics and financial environment characteristics; The original credit features are normalized to obtain the target credit features.

7. The method according to claim 1, characterized in that, Target customer information for the target customers is collected from multiple channels, including: Obtain the credit service information of the target customer from the credit business system of the financial institution; Obtain credit record information of the target customers from third-party credit reporting agencies and obtain financial environment information through the Internet; Obtain customer profile information, transaction behavior information, and asset information of the target customer from the financial institution's customer information system; The target customer information is determined based on the credit service information, the credit record information, the financial environment information, the customer profile information, the transaction behavior information, and the asset information.

8. A data processing apparatus, characterized in that, include: The first collection unit is used to collect target customer information of the target customer from multiple channels every preset time interval when it is detected that the target customer is applying for credit services and with the authorization of the target customer, and to extract target credit features from the target customer information. The first determining unit is used to determine a target state space based on the target credit information and the target credit features in the target customer information, wherein the target state space is used to describe the set of all credit features of the target customer during the use of the credit service; A decision unit is used to input the target state space into a target decision model to obtain the decision action output by the target decision model and the evaluation result of the decision action. The target decision model is obtained by training a neural network based on a deep deterministic policy gradient algorithm using historical customer information and historical credit features. The second determining unit is used to determine the target adjustment action for the credit service for the target customer based on the evaluation result of the decision action and the decision action.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the data processing method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the data processing method according to any one of claims 1 to 7.

Images