Loan request response method, device, electronic equipment and program product

Abstract

The application discloses a loan request response method and device, electronic equipment and program product, and relates to the field of financial technology. The method comprises the following steps: extracting customer features from customer information of a loan request, and constructing a customer state vector sequence according to the customer features; constructing a target solving equation according to the customer state vector sequence and an external environment feature sequence based on a fractional order diffusion equation solving equation, solving the target solving equation at a preset time step, and obtaining a default probability prediction value of the loan request in a future time window; responding to the loan request according to the default probability prediction value of the future time window, obtaining a response result, and sending the response result to a target customer. Through the application, the problem that the prediction result is unstable and the response result of the loan request is unstable when the loan request is audited by predicting the loan default probability in the related art is solved.

Description

Technical Field

[0001] This application relates to the field of financial technology, and more specifically, to a method, apparatus, electronic device, and program product for responding to loan requests. Background Technology

[0002] In the current field of bank credit risk management, the prediction of loan default probability generally relies on traditional statistical models or machine learning methods, such as logistic regression, random forests, support vector machines, or shallow neural networks. Their core assumption is that credit risk is determined solely by current or recent customer behavior, i.e., it follows the Markov property. These methods typically use only static or short-term dynamic data such as the customer's repayment history, credit score, and income level over the past 6 to 12 months for modeling. However, traditional models lack a memory mechanism for historical disturbances and cannot effectively capture this non-instantaneously decaying dependency, resulting in predictions that are highly sensitive to short-term data noise and fluctuate wildly. Especially during periods of sudden events, the models are prone to misjudgment because they cannot quantify the cumulative effect of historical events—low-risk customers are mistakenly labeled as high-risk, or high-risk customers are wrongly approved due to good short-term repayment performance. This instability in prediction directly impacts the loan approval process, causing a series of chain reactions, including a decline in customer experience.

[0003] There is currently no effective solution to the problem that when reviewing loan requests by predicting the probability of loan default, the relevant technologies often rely on short-term historical data, which makes the prediction results unstable and thus leads to unstable response results for loan requests. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, electronic device, and program product for responding to loan requests, in order to solve the problem in the related art that when reviewing loan requests by predicting the probability of loan default, the reliance on short-term historical data usually leads to unstable prediction results, which in turn results in unstable response results to loan requests.

[0005] To achieve the above objectives, according to one aspect of this application, a method for responding to a loan request is provided. The method includes: receiving a loan request initiated by a target customer; extracting customer features from customer information in the loan request with the authorization of the target customer; constructing a customer state vector sequence based on the customer features; and extracting an external environment feature sequence from external environment information; constructing a target solution equation based on a fractional diffusion equation using the customer state vector sequence and the external environment feature sequence; solving the target solution equation at a preset time step to obtain a predicted default probability value for the loan request in a future time window; responding to the loan request based on the predicted default probability value in the future time window; obtaining a response result; and sending the response result to the target customer.

[0006] Furthermore, based on the fractional diffusion equation, a target solution equation is constructed using the customer state vector sequence and the external environment feature sequence. This target solution equation is solved at a preset time step to obtain a predicted default probability value for the loan request within a future time window. This process includes: processing the customer state vector sequence using a preset algorithm to obtain a fractional order, where the preset algorithm quantifies the duration of the influence of features in the customer state vector sequence on loan behavior; determining the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of those features; fitting the external environment feature sequence using a historical regression model to obtain an external environment feature mapping relationship; and constructing a target solution equation based on the grid discretization results of the customer state vector sequence, the fractional order, the diffusion coefficient, and the external environment feature mapping relationship, using a finite difference scheme and central difference approximation based on the fractional derivative, and solving the target solution equation at a preset time step to obtain a predicted default probability value for the future time window.

[0007] Furthermore, based on the grid discretization results of the customer state vector sequence, fractional order, diffusion coefficient, and the mapping relationship of external environment features, a target solution equation is constructed using a finite difference scheme with fractional derivatives and a central difference approximation. This target solution equation is then solved based on a preset time step to obtain the predicted default probability value for the future time window. This process includes: grid discretizing the customer state vector sequence to obtain the probability distribution sequence of each grid point; determining the duration of the influence of features in the customer state vector sequence on each time step using a finite difference scheme with fractional derivatives based on the probability distribution sequence of each grid point, the fractional order, and the preset time step; determining the diffusion characteristics of each spatial step based on the probability distribution sequence of each grid point, the preset spatial step, the diffusion coefficient, and the mapping relationship of external environment features; constructing a target solution equation based on the duration of influence at each time step and the diffusion characteristics at each spatial step; and iteratively solving the target solution equation step by step from the start of the future time window, obtaining the predicted default probability value corresponding to that time step, until the end of the future time window, to obtain the predicted default probability value for the future time window.

[0008] Furthermore, the diffusion coefficient is determined based on the statistical characteristics of the customer state vector sequence and the duration of the influence of the features in the customer state vector sequence. This includes: determining the historical variance of the customer state vector sequence within a historical time window, where the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; determining the duration of the influence of the features in the customer state vector sequence based on the features in the customer state vector sequence within the historical time window and a preset threshold; and determining the diffusion coefficient based on the fractional-order gamma function, the fractional-order, the historical variance, and the duration of the influence of the features in the customer state vector sequence.

[0009] Furthermore, the customer state vector sequence is processed based on a preset algorithm to obtain the fractional order, including: obtaining the historical default rate of the target customer in multiple sub-time windows within the historical time window, obtaining multiple historical default rate sequences, and determining the historical mean of each historical default rate sequence; determining the cumulative deviation sequence of each sub-time window based on the historical mean and each historical default rate sequence, and determining the standard deviation of each sub-time window; determining the range of each sub-time window based on the cumulative deviation sequence of each sub-time window; determining the rescaled range of each sub-time window based on the range and the standard deviation of each sub-time window; performing linear fitting on the logarithm of the rescaled range of each sub-time window and the logarithm of each sub-time window to obtain slope information, and determining the fractional order based on the slope information.

[0010] Furthermore, customer features are extracted from the customer information of loan requests, and a customer state vector sequence is constructed based on these features. An external environment feature sequence is also extracted from external environment information, including: obtaining the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and extracting asset load characteristics from the asset information; constructing a customer state vector sequence based on the loan information, historical default information, asset load characteristics, and credit risk score corresponding to different times within the preset time period; obtaining external environment information within a preset time period from preset channels, and processing the external environment information through a pre-trained semantic processing model to obtain an external environment feature sequence. The external environment information includes: preset indicator growth values ​​and natural language text contained in preset websites.

[0011] Furthermore, the loan request is responded to based on the predicted default probability value for the future time window, resulting in a response. This includes: numerically integrating the predicted default probability value for the future time window to obtain the target default probability corresponding to the loan request; if the target default probability is greater than a preset probability threshold, a risk warning is triggered, and the loan request is sent to the target recipient, who reviews the loan request and processes it based on the review results; if the target default probability is less than or equal to the preset probability threshold, the loan request is adjusted based on the target customer's repayment behavior information and credit risk score, and the response result is determined based on the adjusted loan request.

[0012] To achieve the above objectives, according to another aspect of this application, a loan request response apparatus is provided, comprising: a construction unit, configured to receive a loan request initiated by a target customer, and, with the authorization of the target customer, extract customer features from the customer information of the loan request, construct a customer state vector sequence based on the customer features, and extract an external environment feature sequence from external environment information; a prediction unit, configured to solve an equation based on a fractional diffusion equation, construct a target solution equation based on the customer state vector sequence and the external environment feature sequence, solve the target solution equation at a preset time step, and obtain a predicted default probability value for the loan request in a future time window; and a response unit, configured to respond to the loan request based on the predicted default probability value in the future time window, obtain a response result, and send the response result to the target customer.

[0013] Furthermore, the prediction unit includes: a first processing subunit, used to process the customer state vector sequence based on a preset algorithm to obtain a fractional order, wherein the preset algorithm is used to quantify the duration of the influence of features in the customer state vector sequence on loan behavior; a determination subunit, used to determine the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of features in the customer state vector sequence; a fitting subunit, used to fit the external environment feature sequence through a historical regression model to obtain the external environment feature mapping relationship; and a first construction subunit, used to construct the target solution equation based on the grid discretization result of the customer state vector sequence, the fractional order, the diffusion coefficient, and the external environment feature mapping relationship, using the finite difference scheme of the fractional derivative and the central difference approximation, and solve the target solution equation based on a preset time step to obtain the predicted default probability value for the future time window.

[0014] Further, the first construction subunit includes: a processing module for performing grid discretization on the customer state vector sequence to obtain a probability distribution sequence for each grid point; a first determination module for determining the duration of the influence of features in the customer state vector sequence on each time step using a finite difference scheme with fractional derivatives based on the probability distribution sequence of each grid point, the fractional order, and a preset time step; a second determination module for determining the diffusion characteristics of each spatial step based on the probability distribution sequence of each grid point, the preset spatial step, the diffusion coefficient, and the mapping relationship of external environmental features; a construction module for constructing a target solution equation based on the duration of influence of each time step and the diffusion characteristics of each spatial step; and a prediction module for iteratively solving the target solution equation step by step from the start time of the future time window with a preset time step to obtain the predicted default probability value corresponding to that time step, until the end time of the future time window, to obtain the predicted default probability value of the future time window.

[0015] Furthermore, the determining subunit includes: a third determining module, used to determine the historical variance of the customer state vector sequence within a historical time window, wherein the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; a fourth determining module, used to determine the duration of the influence of the features in the customer state vector sequence based on the features in the customer state vector sequence within the historical time window and a preset threshold; and a fifth determining module, used to determine the diffusion coefficient based on the fractional-order gamma function, the fractional-order, the historical variance, and the duration of the influence of the features in the customer state vector sequence.

[0016] Further, the first processing subunit includes: an acquisition module, used to acquire the historical default rates of the target customer in multiple sub-time windows within a historical time window, obtain multiple historical default rate sequences, and determine the historical mean of each historical default rate sequence; a sixth determination module, used to determine the cumulative deviation sequence of each sub-time window based on the historical mean and each historical default rate sequence, and to determine the standard deviation of each sub-time window; a seventh determination module, used to determine the range of each sub-time window based on the cumulative deviation sequence of each sub-time window; an eighth determination module, used to determine the rescaled range of each sub-time window based on the range and the standard deviation of each sub-time window; and a ninth determination module, used to perform linear fitting on the logarithm of the rescaled range of each sub-time window and the logarithm of each sub-time window to obtain slope information, and to determine the fractional order based on the slope information.

[0017] Furthermore, the construction unit includes: an extraction subunit, used to obtain the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and to extract asset load features from the asset information; a second construction subunit, used to construct a customer state vector sequence based on the loan information, historical default information, asset load features, and credit risk score corresponding to different times within the preset time period; and a second processing subunit, used to obtain external environment information within a preset time period from preset channels, and to process the external environment information through a pre-trained semantic processing model to obtain an external environment feature sequence, wherein the external environment information includes: preset indicator growth values ​​and natural language text contained in preset websites.

[0018] Furthermore, the response unit includes: a third processing subunit, used to numerically integrate the predicted default probability value for a future time window to obtain the target default probability corresponding to the loan request; an early warning subunit, used to trigger a risk warning when the target default probability is greater than a preset probability threshold, and send the loan request to the target object, wherein the target object reviews the loan request and processes the loan request based on the review result; and an adjustment subunit, used to adjust the loan request based on the target customer's repayment behavior information and the target customer's credit risk score when the target default probability is less than or equal to the preset probability threshold, and determine the response result based on the adjusted loan request.

[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 a response method for any of the above-described loan requests, and when executed by a processor, implements the steps of the response method for loan requests 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, comprising stored computer instructions, wherein, when the computer instructions are executed by a processor, a response method for any of the above-described loan requests is implemented.

[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 the one or more processors, the one or more processors cause the one or more processors to implement any of the above-described loan request response methods.

[0022] In this embodiment, by receiving a loan request initiated by a target customer, and with the target customer's authorization, customer features are extracted from the customer information of the loan request. A customer state vector sequence is constructed based on the customer features, and an external environment feature sequence is extracted from the external environment information. Based on the fractional diffusion equation, a target solution equation is constructed according to the customer state vector sequence and the external environment feature sequence. The target solution equation is solved with a preset time step to obtain a predicted default probability value for the loan request in a future time window. The loan request is responded to based on the predicted default probability value in the future time window, and the response result is sent to the target customer. This solves the technical problem that when reviewing loan requests by predicting the loan default probability, the reliance on short-term historical data usually leads to unstable prediction results, which in turn leads to unstable response results for the loan request.

[0023] Through the above steps, with customer authorization, a customer state vector sequence and an external environment feature sequence are constructed. An objective solution equation is built based on the fractional-order credit risk diffusion equation. Fractional derivatives are used to model the long-range dependence and memory kernel effect of credit risk, enabling the model to capture the non-local, decaying lag effect of historical events on the current risk state. Even if short-term historical data of the customer is noisy, the long-term evolution path can be integrated through the memory mechanism of fractional derivatives, thereby improving the stability and robustness of default probability prediction. This avoids the problem of prediction instability caused by traditional Markov models that rely solely on the current state and ignore historical events. Therefore, it can solve the technical problem in related technologies where loan requests are reviewed based on predicted loan default probabilities, which often rely on short-term historical data, leading to unstable prediction results and consequently unstable loan request response results. This achieves the technical effect of reducing misjudgments caused by external environmental noise and improving the stability of credit risk prediction results. 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 response method for a loan request, according to Embodiment 1 of this application;

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

[0027] Figure 3 This is a schematic diagram of an optional loan request response flow provided according to Embodiment 1 of this application;

[0028] Figure 4 This is a schematic diagram of a loan request response device according to Embodiment 2 of this application;

[0029] Figure 5 This is a schematic diagram of a loan request response 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 stability of credit processing results of financial institutions during the credit business process. 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 and standards of the relevant regions, 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 process.

[0033] Example 1

[0034] According to an embodiment of this application, a method embodiment for responding to a loan request 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 loan request response method according to Embodiment 1 of this application. Figure 1As 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 the program instructions / data storage device corresponding to the loan request response 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 loan request response 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 method for responding to a loan request is shown. Figure 2 This is a flowchart of an optional loan request response method provided according to Embodiment 1 of this application.

[0041] Step S201: Receive a loan request initiated by the target customer. With the authorization of the target customer, extract customer features from the customer information of the loan request, construct a customer state vector sequence based on the customer features, and extract an external environment feature sequence from the external environment information.

[0042] The executing entity in this embodiment 1 can be a loan request response system. This system is used for a credit default risk prediction method based on fractional partial differential equations. The system acquires customer static characteristics and external environment characteristics to construct a multidimensional customer state vector sequence, which is then used as initial conditions and driving inputs to solve the time-fractional Fokker-Planck equation, thereby outputting the probability density evolution of the default state. The system then integrates within a preset default region to calculate the predicted default probability value. This technology enables the characterization of long-range memory effects of credit risk in the time dimension, the modeling of nonlocal propagation characteristics in the state space dimension, and the adaptive calibration of model parameters through a data-driven approach, thereby improving the robustness and stability of default probability prediction.

[0043] In this embodiment 1, the loan request initiated by the target customer refers to the loan request data submitted by the target customer through mobile banking or a corporate credit platform, which includes the application intention and basic customer information. With the customer's explicit authorization (e.g., by checking the "Personal Information Use Consent Form"), static information provided by the customer is read from the loan application form. Multiple extracted customer features are organized into an ordered vector sequence according to the time dimension, representing a multidimensional numerical sequence representing the changes in the customer's credit status at continuous points in time. External environmental information related to the customer's industry or region is obtained from a preset data source. The external environmental feature sequence is a numerical sequence that changes over time and reflects external information of the system.

[0044] By receiving loan requests and extracting customer and external environment features, a customer state vector sequence and an external environment feature sequence can be constructed, providing a dynamic, temporal, and multidimensional input foundation for fractional-order models.

[0045] Step S202: Solve the equation based on the fractional diffusion equation. Construct the target solution equation based on the customer state vector sequence and the external environment feature sequence. Solve the target solution equation with a preset time step to obtain the predicted default probability value for the loan request in the future time window.

[0046] In this embodiment 1, both the customer state vector sequence and the external environment feature sequence are used as input variables and substituted into the mathematical framework of fractional partial differential equations (FPDEs) to establish a governing equation characterizing the evolution of credit risk density over time and space. The core of this equation is the Caputo fractional derivative (CFD), a form that allows the model to retain the decaying memory effect of historical risk states on current risk, rather than relying solely on the current state (as in traditional Markov models). For example, if a customer had a high credit score in 2022 but their income declined after 2023, traditional models might ignore credit accumulation before 2022. However, this method, through the memory kernel of the fractional derivative, can quantify the long-term decaying impact of historical events in 2022, making risk evolution more realistic. Continuous time is discretized into fixed-interval time points (e.g., daily, weekly, or monthly), and the distribution of risk density at each time point is calculated iteratively at a preset step size.

[0047] For example, if the time step is set to 1 day and the prediction window is 30 days, then starting from t=0, the process progresses daily to t=30. At each step, the space and time are discretized using the Finite Difference Method (FDM) to construct the objective solution equation. Through the above numerical solution, the cumulative probability that the customer's credit risk density crosses the default threshold region within the preset time window is obtained, which is the aforementioned default probability prediction value.

[0048] By introducing a memory kernel mechanism with fractional derivatives through the above steps, the model is equipped with the ability to model long-range dependencies and dynamically quantify the decaying impact of historical events (such as the fact that an interest rate increase three months ago still affects current repayment ability). This significantly improves the stability and robustness of predictions over time, especially for sudden events. It avoids prediction jumps caused by data gaps or noise, and achieves stable predictions of loan requests. This solves the problem in related technologies where traditional models (such as logistic regression and random forest) only use current or recent data and ignore the continuous impact of historical events on current default risk, resulting in unstable prediction results.

[0049] Step S203: Respond to the loan request based on the predicted default probability value of the future time window, obtain the response result, and send the response result to the target customer.

[0050] In this embodiment 1, in order to respond to a customer's loan application, the loan request can be responded to based on the predicted default probability value of a future time window, an approval conclusion can be generated and fed back to the customer, thereby improving loan approval efficiency and customer experience.

[0051] Specifically, after calculating the predicted probability of default for a customer within a specific future time window (e.g., 3 months, 6 months, or 1 year) using a fractional-order credit risk model, the system automatically generates a processing decision for the loan application based on a comparison of this value with a preset risk threshold (e.g., 5%), resulting in the aforementioned response. For example, the response result includes, but is not limited to: loan approval, rejection, or transfer to manual review, along with recommended loan amount, interest rate, and risk mitigation suggestions. The response result is displayed to the customer through the mobile banking app interface, where the customer can choose to confirm, adjust, or resubmit. After customer confirmation, the system automatically generates an electronic contract and pushes it to the customer's terminal, completing the response loop. The entire process requires no manual intervention, achieving automated and personalized responses. For instance, if the integral value of the risk density function in the default zone in the solution for the next 30 days (future time window) is 0.07, then the predicted probability of default for the customer within the next 30 days is 7%, which can be used to assist in loan approval decisions.

[0052] Optionally, in the loan request response method provided in Embodiment 1 of this application, a target solution equation is constructed based on the customer state vector sequence and the external environment feature sequence, and the target solution equation is solved at a preset time step to obtain a predicted default probability value for the loan request in a future time window. This includes: processing the customer state vector sequence based on a preset algorithm to obtain a fractional order, wherein the preset algorithm is used to quantify the duration of the influence of features in the customer state vector sequence on loan behavior; determining the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of features in the customer state vector sequence; fitting the external environment feature sequence through a historical regression model to obtain an external environment feature mapping relationship; constructing a target solution equation based on the grid discretization result of the customer state vector sequence, the fractional order, the diffusion coefficient, and the external environment feature mapping relationship, using a finite difference scheme and central difference approximation based on the fractional derivative, and solving the target solution equation at a preset time step to obtain a predicted default probability value for the future time window.

[0053] In this embodiment 1, in order to achieve quantitative memory and dynamic propagation of historical impacts during the evolution of credit risk, the fractional order and diffusion coefficient can be calibrated based on the statistical characteristics of customer state vector sequences, and a fractional diffusion equation can be constructed by integrating regression mapping of external environmental characteristics. This can improve the attention of default probability prediction to the lag effect of historical events.

[0054] Specifically, rescaled range analysis (R / S) is used to detect long-range dependencies of time series features such as credit scores and debt-to-equity ratios in the customer state vector sequence, calculates the Hurst exponent (H), and uses this exponent as the order of the fractional derivative. ,Right now =H, thus transforming the degree of influence of customers' historical behavior on subsequent events into a memory strength parameter in the mathematical model. Subsequently, based on the historical volatility of the customer state vector sequence... Calculate the diffusion coefficient with the time scale t of risk propagation characteristics. This coefficient reflects the nonlocal diffusion rate of risk in the state space caused by historical events. Then, by fitting the external environment feature sequence to a historical regression model, the nonlinear mapping relationship between it and the risk state drift term is obtained, i.e., the external environment feature mapping relationship. Finally, based on the two-dimensional grid discretization results of the customer state vector sequence (such as grid nodes divided by the debt-to-equity ratio axis and the credit score axis), and the aforementioned fractional order... diffusion coefficient Based on the mapping relationship between the external environment characteristics, the time derivative is approximated by the finite difference scheme of Caputo Fractional Derivative (CFD), and the spatial diffusion term is handled by the two-dimensional central difference approximation. The objective solution equation is constructed, and then implicit iterative solution is performed with a preset time step to output the predicted default probability value of the future time window.

[0055] Fractional order It is a dimensionless parameter that measures the duration of credit risk memory, and its value ranges from 0 to 1. ≤1; For example, if a customer's historical credit behavior is highly consistent (e.g., maintaining high debt even after 6 consecutive months of delinquency), then the Hurst index H=0.75, corresponding to =0.75, indicating that risk evolution has a significant long-range dependence. Diffusion coefficient It is a physical quantity that quantifies the rate of risk diffusion, with units of (state units² / time^2). The mapping relationship between external environmental characteristics is a functional relationship established through historical regression models between external variables and the direction and intensity of risk drift. For example, the regression model output F(x,t) = + External environmental characteristics A1+ • External environmental feature A2 serves as the drift term in the equation. The grid discretization result divides the customer state vector (x1 = debt-to-equity ratio, x2 = credit score) into a regular grid (e.g., Δx1 = 0.05, Δx2 = 10), forming the probability density value of each node in the discrete state space, which is used for numerical calculation.

[0056] By employing the above method, compared to traditional models that only use a fixed order or static diffusion coefficient, this scheme achieves fractional-order diffusion. With diffusion coefficient The dynamic adaptive calibration, based on the statistical characteristics of the customer's own behavioral sequences rather than manual settings, enables the model to truly reflect the long-term memory characteristics of individual credit risk. Simultaneously, by quantifying historical events in the external environment as drift terms through the mapping relationship of external environmental features, it solves the problems of neglecting the persistence of the impact of historical events and the nonlinear transmission of sudden events in existing technologies, achieving a significant improvement in the stability and accuracy of default probability prediction.

[0057] Optionally, in the loan request response method provided in Embodiment 1 of this application, a target solution equation is constructed based on the grid discretization result of the customer state vector sequence, the fractional order, the diffusion coefficient, and the mapping relationship of external environment features, using a finite difference scheme based on the fractional derivative and the central difference approximation. The target solution equation is then solved based on a preset time step to obtain a predicted default probability value for the future time window. This includes: performing grid discretization on the customer state vector sequence to obtain a probability distribution sequence for each grid point; and using a finite difference scheme based on the probability distribution sequence of each grid point, the fractional order, and the preset time step. The format determines the duration of the influence of features in the customer state vector sequence on each time step; the diffusion characteristics of each spatial step are determined based on the probability distribution sequence of each grid point, the preset spatial step, the diffusion coefficient, and the mapping relationship with external environmental features; the target solution equation is constructed based on the duration of influence of each time step and the diffusion characteristics of each spatial step; starting from the beginning of the future time window, the target solution equation is iteratively solved step by step with the preset time step to obtain the default probability prediction value corresponding to that time step, until the end of the future time window, to obtain the default probability prediction value of the future time window.

[0058] In this embodiment 1, in order to realize the nonlocal evolution modeling of credit risk over time in a multidimensional state space, the customer state vector sequence can be discretized by grid and a dynamic solution equation can be constructed by combining fractional memory mechanism and spatial diffusion characteristics. This allows for the accurate capture of the continuous impact of historical events and the spatial propagation path of risk.

[0059] Specifically, the customer state vector sequence is first divided into regular grid nodes to obtain the probability distribution sequence of each grid point at each time step. The grid discretization result divides the customer state space into a finite set of nodes. For example, the debt-to-equity ratio x1 is divided into the interval [0,1] with a step size Δx1=0.05, generating 21 discrete points; the credit score x2 is divided into the interval [300,900] with a step size Δx2=10, generating 61 discrete points, forming 21×61=1281 grid nodes, each node corresponding to a probability density value.

[0060] Subsequently, based on the probability distribution sequence of each grid point, the preset fractional order α, and the time step Δt, the finite difference scheme of Caputo fractional derivatives is used to calculate the weighted sum of the probability changes of all historical time steps affecting the grid point at the current time step. This weighting is determined by the memory decay coefficient. This determines the duration of the influence of features in the customer state vector sequence on each time step. Then, based on the probability distribution sequence of each grid point, preset spatial step sizes Δx1 and Δx2, and the calibrated diffusion coefficient... And the mapping relationship of external environmental features, using the central difference to approximate the calculation of the two-dimensional Laplace operator. ²u and drift term ·(F·u) is used to determine the diffusion characteristics of each spatial step, i.e., the local migration intensity of risk in the state space driven by both diffusion and drift. Then, based on the duration of the impact at each time step and the diffusion characteristics at each spatial step, a target solution equation containing fractional time derivatives and second-order spatial derivatives is constructed. Finally, starting from the beginning of the future time window, the target solution equation is implicitly iterated over a preset time step Δt, progressively outputting the predicted default probability value for each time step until the termination time, thus obtaining the default probability distribution for the complete future time window.

[0061] In an alternative embodiment, the implicit iterative scheme, i.e., the objective solution equation described above, can be expressed as: ,in, This represents the risk probability density distribution vector at all grid nodes at the (m+1)th time step, which is the set of predicted default probabilities for future times. Let represent the current risk probability density distribution vector at all grid nodes at the m-th time step. Let A represent the sparse coefficient matrix constructed by the fractional-order time discretization scheme and the implicit discretization of the spatial diffusion term; its structure depends on the fractional-order order. diffusion coefficient Spatial step size , And the grid topology, reflecting the implicit dependency between the current state and the next state. B represents a sparse coefficient matrix composed of the explicit discrete weights of the historical memory term and the explicit part of the spatial diffusion term, whose elements are the probability distribution of the previous time step, fractional memory weights, and so on. The spatial difference coefficients determine the state at the previous time step. The mapping is the driving force at the current moment, and is a numerical expression of the historical cumulative effect after the fractional derivative is discretized in time. This represents the mapping relationship (i.e., the drift term) based on external environmental features at the m-th time step. The terms are composed of a grid point and each component corresponds to the risk migration intensity caused by historical events in the external environment. They are the inputs of the model's response to external events.

[0062] The objective function equation described above is derived from the time fractional-order Fokker-Planck equation, which can be expressed as follows: ,in, Indicates risk density The Caputo fractional derivative with respect to time t characterizes the long-range time memory effect in the evolution of credit risk; The diffusion coefficient is adaptively calibrated by the degree of feature variation and feature time scale in the customer state vector sequence, reflecting the intensity and speed of risk diffusion in the customer state space. For a two-dimensional Laplace operator, it represents the risk density in the customer state space and the trend of risk migration and diffusion between adjacent state points; The external environment feature mapping relationship is a function of customer state x and time t, which is obtained by fitting a historical regression model and represents the direction and intensity of risk migration caused by historical events in the external environment. For drift term, It indicates its divergence, and the negative sign indicates the net migration direction of risk to low probability areas (i.e., safe areas) or high probability areas (i.e., default areas); , represents the noise item of sudden events in the external environment. In this scheme, it is usually set to zero or sparse impulse and is only activated when extreme events occur.

[0063] Based on the time-fractional Fokker-Planck equations, the Caputo fractional derivative can be used to provide a mathematical expression for the nonlocal time derivative, enabling the model to capture the persistent impact of historical events on current risk. The Caputo fractional derivative can be expressed as: , Used to quantify the rate of change of credit risk status u at the current moment due to its historical evolution path, when When =1, it degenerates into a regular derivative. ,in, , representing the risk state vector (e.g. =Debt-to-equity ratio =Credit score). This represents the gamma function. Represents the memory kernel function, describing historical moments. Influence weight, The smaller the value, the slower the historical influence decays (the stronger the memory). The closer to 1, the weaker the memory. This represents the instantaneous rate of change of risk density u at a historical moment τ, reflecting the slope of credit risk at that moment, i.e., the dynamic trend of customer status over time. This means that the integral from the initial time 0 to the current time t is performed, which means that the current risk change is the weighted cumulative effect of all historical events, rather than depending solely on the current state.

[0064] Furthermore, the continuous Caputo fractional derivative is transformed into a programmable time-discrete recursive formula, and the fractional memory effect is transformed into a weighted sum of probability increments over all historical time steps, making... The calculation depends on The discrete-time recursive formula can be expressed as: ,in, A fractional scaling factor representing the time step; Preset time step; This represents the fractional-order memory weighting coefficient, used to quantify the degree of decay in the contribution of historical time steps to the current derivative. This coefficient is determined by the order... The only certainty is that earlier historical states (the larger k is) have lower weights, but their decay rate is slower than exponential decay, reflecting long-range memory. Indicates time arrive The local change in the probability distribution of risk density between adjacent time points; Indicates the time from the current moment Tracing back to the initial moment All historical change increments are weighted and summed, where each increment... Multiply by its corresponding memory weight This allows for a numerical approximation of the fractional derivative.

[0065] The two-dimensional Laplace operator (spatial discrete two-dimensional diffusion term) in the time fractional Fokker-Planck equation can be expressed as: , Indicates risk density The Laplace operator in a two-dimensional state space This indicates the i-th element in the state space grid. Discrete points (such as the i-th interval of the debt-to-equity ratio) and the j-th interval The risk probability density value at a discrete point (such as the j-th interval of a credit score) represents the likelihood of a customer defaulting under that particular combination of conditions. , Indicates in The density value of adjacent grid points in the direction, i.e., the density value of the current point. Density of adjacent state points in terms of debt-to-equity ratio , Indicates in The density value of adjacent grid points in the direction, i.e., the density value of the current point. Density of adjacent state points in the credit scoring dimension. This represents the central difference approximation of risk density along the debt-to-equity ratio dimension, used to quantify whether the risk tends to concentrate (negative curvature) or diffuse (positive curvature) along this dimension. This represents the central difference approximation of risk density along the credit scoring dimension, used to quantify the diffusion trend of risk along this dimension.

[0066] Using the above method, compared to traditional models that rely solely on Markov assumptions or single-point state transitions, this scheme incorporates the duration of memory over time (as described above). And CFD finite difference representation) and the diffusion characteristics of spatial dimension (by By jointly modeling credit risk at the grid level (using central difference representation), collaborative calculation of credit risk in both time memory and spatial propagation dimensions is achieved, resulting in a significant improvement in the stability of default probability prediction.

[0067] Optionally, in the loan request response method provided in Embodiment 1 of this application, the diffusion coefficient is determined based on the statistical characteristics of the customer state vector sequence and the duration of the influence of the features in the customer state vector sequence, including: determining the historical variance of the customer state vector sequence within a historical time window, wherein the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; determining the duration of the influence of the features in the customer state vector sequence based on the features in the customer state vector sequence within the historical time window and a preset threshold; and determining the diffusion coefficient based on the fractional-order gamma function, the fractional-order, the historical variance, and the duration of the influence of the features in the customer state vector sequence.

[0068] In this embodiment 1, in order to achieve dynamic adaptive calibration of the diffusion coefficient, the diffusion intensity can be calculated by jointly calculating the historical fluctuations of customer status and the memory duration, thereby accurately reflecting the response intensity of risk evolution to historical disturbances.

[0069] Specifically, the first step is to determine the historical variance of the customer state vector sequence within the historical time window. This variance is obtained by calculating the squared mean of the deviations of features such as credit scores and debt-to-equity ratios from their mean over a historical period, directly representing the volatility of a customer's risk status. Subsequently, based on the volatility behavior of features in the customer status vector sequence within this historical time window and a preset threshold (such as a Hurst exponent threshold of 0.5), it is determined whether they possess long-term dependence. If the volatility exhibits a persistent trend, its duration is determined to be a long-term, non-instantaneously decaying influence. Finally, based on the fractional-order gamma function (Γ) and the fractional-order... Historical variance and duration of impact Through formula Calculate the diffusion coefficient. The duration of impact is the length of time it takes for a risk state to be significantly affected by a historical event. Its value can be set based on historical events, and a default value of 6 months (or 12 months) can be set.

[0070] The diffusion coefficient is achieved through the above method. By dynamically calibrating the statistical characteristics (variance) and memory characteristics (duration of impact) of customers' historical behavior, the subsequent fractional diffusion equation can truly reflect the differentiated risk evolution path of different customer groups, significantly improving the individual adaptability and system robustness of default probability prediction.

[0071] Optionally, in the loan request response method provided in Embodiment 1 of this application, the customer state vector sequence is processed based on a preset algorithm to obtain the fractional order, including: obtaining the historical default rate of the target customer in multiple sub-time windows within a historical time window, obtaining multiple historical default rate sequences, and determining the historical mean of each historical default rate sequence; determining the cumulative deviation sequence of each sub-time window based on the historical mean and each historical default rate sequence, and determining the standard deviation of each sub-time window; determining the range of each sub-time window based on the cumulative deviation sequence of each sub-time window; determining the rescaled range of each sub-time window based on the range and the standard deviation of each sub-time window; performing linear fitting on the logarithm of the rescaled range of each sub-time window and the logarithm of each sub-time window to obtain slope information, and determining the fractional order based on the slope information.

[0072] In this embodiment 1, in order to dynamically determine the fractional order that characterizes the long-range memory properties of credit risk, the memory strength can be estimated by rescaled range analysis based on historical default rate sequences, thereby quantifying the nonlocal time dependence in risk evolution.

[0073] Specifically, the historical default rates of target customers across multiple sub-time windows within a historical time window are obtained, resulting in multiple historical default rate sequences. Each sub-time window represents the percentage of defaults by the target customer within a fixed duration (e.g., quarter or month). Then, the historical mean of each historical default rate sequence is calculated, representing the arithmetic mean of the default rates within that sub-window. Based on the historical mean and actual values ​​of each historical default rate sequence, the cumulative deviation sequence for each sub-time window is calculated, representing the cumulative deviation of the default rate from the mean from the first sub-window to the current sub-window. Simultaneously, the standard deviation for each sub-time window is determined, reflecting the degree of volatility in the default rate within that window. Based on the cumulative deviation sequence for each sub-time window, its range is calculated, representing the difference between the maximum and minimum cumulative deviations, indicating the cumulative magnitude of risk deviation within that period. Finally, the rescaled range (R / S), the ratio of the range to the standard deviation, is calculated based on the range and standard deviation for each sub-time window. Finally, the natural logarithm of the rescaled range for each sub-time window is taken and linearly fitted with the natural logarithm of the corresponding sub-window length. The resulting slope is the Hurst exponent. The fractional order is determined directly from the slope and a constant. The estimated value enables the automatic extraction of memory parameters from historical default behaviors.

[0074] For example, fractional order The calculation process for the estimated value can be expressed as follows: Where H is the Hurst exponent, i.e., the slope mentioned above, H > 0.5 indicates long-range dependence ( ). ,in, Indicates the length of the sub-time window. This indicates the rescaled range mentioned above.

[0075] Using the above method, the rescaled range is applied to the bank customer default series, and the fractional order is estimated using the Hurst exponent. Unlike traditional models that use a fixed α=1 (Markov assumption) or are set based on experience, this fractional-order model can truly adapt to the long-term risk dependence characteristics of different customer groups, significantly improving the generalization ability of default probability prediction.

[0076] Optionally, in the loan request response method provided in Embodiment 1 of this application, customer features are extracted from the customer information in the loan request, a customer state vector sequence is constructed based on the customer features, and an external environment feature sequence is obtained from a preset channel. This includes: obtaining the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and extracting asset load features from the asset information; constructing a customer state vector sequence based on the loan information, historical default information, asset load features, and credit risk score corresponding to different times within the preset time period; obtaining external environment information within the preset time period from a preset channel, and processing the external environment information through a pre-trained semantic processing model to obtain an external environment feature sequence, wherein the external environment information includes: preset indicator growth values ​​and natural language text contained in a preset website.

[0077] In this embodiment 1, in order to construct a multi-dimensional representation system that integrates customer dynamic status and external environmental events, a time-series state vector can be constructed by integrating customer profile data, credit scores and external text environment, thereby enabling multi-source dynamic evolution modeling of credit risk.

[0078] Specifically, the system retrieves credit risk scores, loan information, historical default information, and asset information of target customers within a preset time period from a pre-defined storage space. It then extracts asset-liability characteristics from the asset information, referring to quantitative indicators such as the customer's debt-to-asset ratio at different times. The credit risk score is a comprehensive risk level value generated by financial institutions based on the customer's credit history and repayment behavior. Subsequently, based on the changes in loan balance, default history, asset-liability characteristics, and credit risk scores at different times within the preset time period (e.g., weekly), these are combined in chronological order to form a customer state vector sequence. For example, the state vector for week 1 is [credit score 685, debt-to-income ratio 7.2, number of defaults in the past year 0], and for week 2 it is [678, 7.8, 0]. Simultaneously, it retrieves external environmental information for the same time period from preset channels, including preset indicator growth values ​​such as GDP growth rate, as well as natural language text such as reports and interpretations published by various financial websites. The text is semantically encoded using a pre-trained semantic processing model (such as the BERT model), and its semantic vectors are extracted to form an external environment feature sequence.

[0079] By using the above method, the customer profiles, behavioral information, and external environmental events of financial institutions are uniformly modeled into a time-series state vector, which significantly enhances the robustness of default probability prediction and provides real, multi-dimensional input for subsequent fractional-order models, thereby benefiting financial institutions in making decisions.

[0080] Optionally, in the loan request response method provided in Embodiment 1 of this application, the loan information in the loan request is adjusted based on the predicted default probability value of the future time window to obtain a loan plan, including: numerically integrating the predicted default probability value of the future time window to obtain the target default probability corresponding to the loan request; if the target default probability is greater than a preset probability threshold, a risk warning is triggered, and the loan request is sent to the target object, wherein the target object reviews the loan request and processes the loan request based on the review result; if the target default probability is less than or equal to the preset probability threshold, the loan request is adjusted based on the repayment behavior information and credit risk score of the target customer, and the response result is determined based on the adjusted loan request.

[0081] In this embodiment 1, in order to achieve dynamic adaptive decision-making in loan approval, integral decision-making can be carried out based on the default probability distribution predicted by the fractional-order model and linked with risk control strategies, thereby enabling accurate risk classification and personalized credit response.

[0082] Specifically, the system first integrates the predicted default probability for the future time window to obtain the target default probability for the loan request. This integral is the accumulation of the default probability density function output by the fractional-order model over the entire time window, reflecting the overall probability that the customer will eventually default during the loan term. For example, if the integral result of the predicted density function over 36 months is 6.2%, the target default probability is 6.2%. Then, if the target default probability exceeds a preset probability threshold (e.g., 5%), a risk warning is triggered, and the loan request is sent to a human reviewer (the aforementioned target group). This human reviewer, considering the customer's historical behavior and industry background, determines whether to approve, freeze, or require supplementary collateral. If the target default probability is less than or equal to the preset probability threshold, the loan amount in the request is automatically adjusted based on the customer's repayment behavior information and credit risk score. A floating interest rate is matched according to the score range, and a loan recommendation is generated based on the score and risk level. Finally, a complete loan plan including the amount, interest rate, term, and repayment method is generated.

[0083] In one optional embodiment, adjusting the loan request based on the target customer's repayment behavior information and credit risk score includes: adjusting the loan amount of the loan request based on the target customer's repayment behavior information and credit risk score when the target default probability is less than or equal to a preset probability threshold; adjusting the interest rate of the loan request based on the credit risk score; determining a loan recommendation for the loan request based on the credit risk score; and determining a loan plan based on the adjusted loan request. For example, the loan interest rate and loan amount (r, L) of the optimal loan plan can be determined by the following formula: Among them, the loan amount is constrained by the upper limit of the risk limit, that is , For bank funding costs, Let C be the predicted probability of customer default at time T, and let C be the default loss rate.

[0084] Using the above method, the continuous probability density function output by the fractional-order model is transformed into the target default probability through numerical integration. Based on this, two decision-making mechanisms are constructed: manual intervention and automatic loan scheme generation. Unlike traditional models that only output single-point default probabilities and use fixed rules for approval, this method solves the problem of large differences in approval results and poor customer experience caused by unstable predictions in existing technologies. It automates the approval process and significantly improves the efficiency and stability of credit approval.

[0085] Optionally, in this embodiment 1, Figure 3 This is a schematic diagram of an optional loan request response flow provided according to Embodiment 1 of this application. For example... Figure 3As shown, in step S301, customers can access the loan application function through the mobile banking app or corporate credit platform to enter the credit business processing interface. In step S302, the mobile banking app automatically pushes a credit risk assessment questionnaire. Customers fill it out and submit it based on their own circumstances, and the system matches the customer's credit risk level based on the questionnaire results. In step S303, customers fill out the loan application form, upload basic information, and submit it to the financial institution's credit system. In step S304, the financial institution's credit system calls a function based on fractional partial differential equations, inputting the following parameters: customer static data: credit score, historical default records, debt-to-asset ratio, industry attributes, etc.; external environmental indicators (such as GDP growth rate, CPI index); and the order of the fractional derivative. diffusion coefficient Step S305: Based on the fractional-order diffusion equation (such as the time-fractional-order Fokker-Planck equation), the equation is solved using numerical algorithms (such as the finite difference method or Monte Carlo simulation), and the predicted probability of customer default at a specific future time point is output. Step S306: The system compares the predicted target default probability with a preset risk threshold (such as a default probability ≤ 5% being considered low risk): if the target default probability... If the risk threshold is ≤, the loan application will be automatically approved, and the optimal loan amount and interest rate will be determined; if the target default probability is ≤, the loan application will be automatically approved. If the risk threshold is reached, a risk warning is triggered, and the process transitions to manual review. The calculation of the target default probability can be expressed as follows: Step S307: Based on the target default probability and customer information, the system adjusts the following: dynamically adjusts the loan amount according to the customer's repayment ability and risk level; matches floating interest rates to high-risk customers and preferential interest rates to low-risk customers; and generates risk mitigation suggestions for high-risk customers, such as requiring collateral or shortening the repayment period. Step S308: The customer views the recommended plan on the mobile banking page, including the loan amount, interest rate, repayment plan, and risk warnings. The customer can choose to: accept the recommended plan: directly confirm and submit a loan application; or manually adjust parameters: such as modifying the loan term, repayment method, etc. The system recalculates the default probability and updates the plan. The financial institution's credit system automatically generates an electronic loan contract, specifying the loan amount, interest rate, repayment plan, and risk clauses, and pushes it to the customer via mobile banking. After the customer signs the electronic contract online, the financial institution completes the loan approval process, and funds are disbursed to the customer's account immediately or the next day. Financial institutions' credit systems continuously monitor customers' credit risk indicators (such as changes in income and fluctuations in debt-to-equity ratios), and dynamically update default probability predictions using a fractional-order model: If the default probability increases significantly, a post-loan warning is triggered, initiating collection or early repayment procedures; if the default probability decreases, customers are allowed to apply for interest rate discounts or credit limit adjustments. After the loan transaction is completed, customers can view the loan status, repayment records, and risk assessment report through mobile banking.

[0086] 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.

[0087] In summary, the loan request response method provided in this application receives a loan request initiated by a target customer, extracts customer features from the customer information of the loan request with the authorization of the target customer, constructs a customer state vector sequence based on the customer features, and extracts an external environment feature sequence from the external environment information; constructs a target solution equation based on the fractional diffusion equation, constructs the target solution equation based on the customer state vector sequence and the external environment feature sequence, solves the target solution equation with a preset time step, and obtains a predicted value of the default probability of the loan request in a future time window; responds to the loan request based on the predicted value of the default probability in the future time window, obtains a response result, and sends the response result to the target customer. This solves the problem in related technologies where reviewing loan requests by predicting the loan default probability usually relies on short-term historical data, resulting in unstable prediction results and thus unstable response results for loan requests.

[0088] Through the above steps, with customer authorization, a customer state vector sequence and an external environment feature sequence are constructed. An objective solution equation is built based on the fractional-order credit risk diffusion equation. Fractional derivatives are used to model the long-range dependence and memory kernel effect of credit risk, enabling the model to capture the non-local, decaying lag effect of historical events on the current risk state. Even if short-term historical data of the customer is noisy, the long-term evolution path can be integrated through the memory mechanism of fractional derivatives, thereby improving the stability and robustness of default probability prediction. This avoids the problem of prediction instability caused by traditional Markov models that rely solely on the current state and ignore historical events. Therefore, it can solve the technical problem in related technologies where loan requests are reviewed based on predicted loan default probabilities, which often rely on short-term historical data, leading to unstable prediction results and consequently unstable loan request response results. This achieves the technical effect of reducing misjudgments caused by external environmental noise and improving the stability of credit risk prediction results.

[0089] Example 2

[0090] This application also provides a loan request response device. It should be noted that the loan request response device of this application can be used to execute the loan request response method provided in this application. The loan request response device provided in this application is described below.

[0091] According to an embodiment of this application, an apparatus for implementing the above-described loan request response method is also provided. Figure 4 This is a schematic diagram of a loan request response device according to Embodiment 2 of this application. Figure 4 As shown, the device includes: a construction unit 401, a prediction unit 402, and a response unit 403.

[0092] Specifically, the construction unit 401 is used to receive loan requests initiated by target customers, extract customer features from the customer information of the loan request with the authorization of the target customer, construct a customer state vector sequence based on the customer features, and extract an external environment feature sequence from the external environment information.

[0093] The prediction unit 402 is used to solve the equation based on the fractional diffusion equation. It constructs the target solution equation based on the customer state vector sequence and the external environment feature sequence, and solves the target solution equation with a preset time step to obtain the predicted value of the default probability of the loan request in the future time window.

[0094] The response unit 403 is used to respond to the loan request based on the predicted default probability value of the future time window, obtain the response result, and send the response result to the target customer.

[0095] The loan request response device provided in this application embodiment receives a loan request initiated by a target customer through a construction unit 401. With the authorization of the target customer, it extracts customer features from the customer information of the loan request, constructs a customer state vector sequence based on the customer features, and extracts an external environment feature sequence from the external environment information. A prediction unit 402 solves the equation based on the fractional diffusion equation, constructs a target solution equation based on the customer state vector sequence and the external environment feature sequence, and solves the target solution equation with a preset time step to obtain a predicted value of the default probability of the loan request in a future time window. A response unit 403 responds to the loan request based on the predicted value of the default probability in the future time window, obtains a response result, and sends the response result to the target customer. This solves the problem in related technologies where reviewing loan requests by predicting the loan default probability usually relies on short-term historical data, which makes the prediction result unstable and thus leads to an unstable response result for the loan request.

[0096] Through the above steps, with customer authorization, a customer state vector sequence and an external environment feature sequence are constructed. An objective solution equation is built based on the fractional-order credit risk diffusion equation. Fractional derivatives are used to model the long-range dependence and memory kernel effect of credit risk, enabling the model to capture the non-local, decaying lag effect of historical events on the current risk state. Even if short-term historical data of the customer is noisy, the long-term evolution path can be integrated through the memory mechanism of fractional derivatives, thereby improving the stability and robustness of default probability prediction. This avoids the problem of prediction instability caused by traditional Markov models that rely solely on the current state and ignore historical events. Therefore, it can solve the technical problem in related technologies where loan requests are reviewed based on predicted loan default probabilities, which often rely on short-term historical data, leading to unstable prediction results and consequently unstable loan request response results. This achieves the technical effect of reducing misjudgments caused by external environmental noise and improving the stability of credit risk prediction results.

[0097] Optionally, in the loan request response device provided in Embodiment 2 of this application, the prediction unit 402 includes: a first processing subunit, used to process the customer state vector sequence based on a preset algorithm to obtain a fractional order, wherein the preset algorithm is used to quantify the duration of the influence of features in the customer state vector sequence on loan behavior; a determination subunit, used to determine the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of features in the customer state vector sequence; a fitting subunit, used to fit the external environment feature sequence through a historical regression model to obtain the external environment feature mapping relationship; and a first construction subunit, used to construct the target solution equation based on the grid discretization result of the customer state vector sequence, the fractional order, the diffusion coefficient, and the external environment feature mapping relationship, using the finite difference scheme of the fractional derivative and the central difference approximation, and to solve the target solution equation based on a preset time step to obtain the predicted value of the default probability in the future time window.

[0098] Optionally, in the loan request response device provided in Embodiment 2 of this application, the first construction subunit includes: a processing module, used to perform grid discretization on the customer state vector sequence to obtain a probability distribution sequence for each grid point; a first determination module, used to determine the duration of the influence of features in the customer state vector sequence on each time step using a finite difference scheme with fractional derivatives based on the probability distribution sequence of each grid point, the fractional order, and a preset time step; a second determination module, used to determine the diffusion characteristics of each spatial step based on the probability distribution sequence of each grid point, the preset spatial step, the diffusion coefficient, and the mapping relationship of external environmental features; a construction module, used to construct a target solution equation based on the duration of influence of each time step and the diffusion characteristics of each spatial step; and a prediction module, used to iteratively solve the target solution equation step by step from the start time of the future time window with a preset time step to obtain the default probability prediction value corresponding to the time step, until the end time of the future time window, to obtain the default probability prediction value of the future time window.

[0099] Optionally, in the loan request response device provided in Embodiment 2 of this application, the aforementioned determining subunit includes: a third determining module, used to determine the historical variance of the customer state vector sequence within a historical time window, wherein the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; a fourth determining module, used to determine the duration of the influence of the features in the customer state vector sequence based on the features in the customer state vector sequence within the historical time window and a preset threshold; and a fifth determining module, used to determine the diffusion coefficient based on the fractional-order gamma function, the fractional-order, the historical variance, and the duration of the influence of the features in the customer state vector sequence.

[0100] Optionally, in the loan request response device provided in Embodiment 2 of this application, the first processing subunit includes: an acquisition module, configured to acquire the historical default rates of the target customer in multiple sub-time windows within a historical time window, obtain multiple historical default rate sequences, and determine the historical mean of each historical default rate sequence; a sixth determination module, configured to determine the cumulative deviation sequence of each sub-time window based on the historical mean of each historical default rate sequence and each historical default rate sequence, and to determine the standard deviation of each sub-time window; a seventh determination module, configured to determine the range of each sub-time window based on the cumulative deviation sequence of each sub-time window; an eighth determination module, configured to determine the rescaled range of each sub-time window based on the range of each sub-time window and the standard deviation of each sub-time window; and a ninth determination module, configured to perform linear fitting on the logarithm of the rescaled range of each sub-time window and the logarithm of each sub-time window to obtain slope information, and determine the fractional order based on the slope information.

[0101] Optionally, in the loan request response device provided in Embodiment 2 of this application, the aforementioned construction unit 401 includes: an extraction subunit, used to obtain the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and extract asset load features from the asset information; a second construction subunit, used to construct a customer state vector sequence based on the loan information, historical default information, asset load features, and credit risk score corresponding to different times within the preset time period; and a second processing subunit, used to obtain external environment information within a preset time period from a preset channel, and process the external environment information through a pre-trained semantic processing model to obtain an external environment feature sequence, wherein the external environment information includes: preset indicator growth values ​​and natural language text contained in a preset website.

[0102] Optionally, in the loan request response device provided in Embodiment 2 of this application, the response unit 403 includes: a third processing subunit, used to perform numerical integration on the predicted default probability value of a future time window to obtain the target default probability corresponding to the loan request; an early warning subunit, used to trigger a risk warning and send the loan request to the target object when the target default probability is greater than a preset probability threshold, wherein the target object reviews the loan request and processes the loan request based on the review result; and an adjustment subunit, used to adjust the loan request based on the repayment behavior information and credit risk score of the target customer when the target default probability is less than or equal to the preset probability threshold, and determine the response result based on the adjusted loan request.

[0103] It should be noted that the aforementioned construction unit 401, prediction unit 402, and response unit 403 correspond to steps S201 to S203 in Embodiment 1. The two modules and their 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 aforementioned modules or units may 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 aforementioned modules may also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0104] Example 3

[0105] Embodiments of this application may provide an electronic device. Figure 5 This is a schematic diagram of a loan request response 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.

[0106] 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.

[0107] The processor can access information and applications stored in memory via a transmission device to perform the following steps: receiving a loan request initiated by a target customer; with the target customer's authorization, extracting customer features from the customer information in the loan request; constructing a customer state vector sequence based on the customer features; and extracting an external environment feature sequence from the external environment information; solving the equation based on the fractional diffusion equation; constructing a target solution equation based on the customer state vector sequence and the external environment feature sequence; solving the target solution equation at a preset time step to obtain a predicted default probability value for the loan request in a future time window; responding to the loan request based on the predicted default probability value in the future time window; obtaining a response result; and sending the response result to the target customer.

[0108] The processor can access information and applications stored in memory via a transmission device to execute the following steps: Solving the equation based on the fractional-order diffusion equation; constructing a target equation based on the customer state vector sequence and the external environment feature sequence; solving the target equation at a preset time step to obtain a predicted default probability value for the loan request in a future time window; including: processing the customer state vector sequence based on a preset algorithm to obtain the fractional-order order, where the preset algorithm is used to quantify the duration of the influence of features in the customer state vector sequence on loan behavior; determining the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of features in the customer state vector sequence; fitting the external environment feature sequence through a historical regression model to obtain the external environment feature mapping relationship; constructing the target equation based on the grid discretization result of the customer state vector sequence, the fractional-order order, the diffusion coefficient, and the external environment feature mapping relationship; and solving the target equation at a preset time step to obtain a predicted default probability value for the future time window.

[0109] The processor can access information and applications stored in memory via a transmission device to execute the following steps: Based on the grid discretization results of the customer state vector sequence, fractional order, diffusion coefficient, and external environment feature mapping relationship, construct the target solution equation using a finite difference scheme and central difference approximation based on the fractional derivative, and solve the target solution equation based on a preset time step to obtain the predicted default probability value for the future time window. This includes: performing grid discretization on the customer state vector sequence to obtain the probability distribution sequence of each grid point; and using the fractional derivative based on the probability distribution sequence of each grid point, the fractional order, and the preset time step. The finite difference scheme is used to determine the duration of the influence of features in the customer state vector sequence on each time step; the diffusion characteristics of each spatial step are determined based on the probability distribution sequence of each grid point, the preset spatial step, the diffusion coefficient, and the mapping relationship of external environmental features; the target solution equation is constructed based on the duration of influence of each time step and the diffusion characteristics of each spatial step; starting from the beginning of the future time window, the target solution equation is iteratively solved step by step with the preset time step to obtain the default probability prediction value corresponding to that time step, until the end of the future time window, to obtain the default probability prediction value of the future time window.

[0110] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: determining the diffusion coefficient based on the statistical characteristics of the customer state vector sequence and the duration of the influence of the features in the customer state vector sequence, including: determining the historical variance of the customer state vector sequence within a historical time window, wherein the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; determining the duration of the influence of the features in the customer state vector sequence based on the features in the customer state vector sequence within the historical time window and a preset threshold; and determining the diffusion coefficient based on the fractional-order gamma function, the fractional-order, the historical variance, and the duration of the influence of the features in the customer state vector sequence.

[0111] The processor can access information and applications stored in memory via a transmission device to execute the following steps: processing the customer state vector sequence based on a preset algorithm to obtain the fractional order, including: obtaining the historical default rates of the target customer in multiple sub-time windows within a historical time window, obtaining multiple historical default rate sequences, and determining the historical mean of each historical default rate sequence; determining the cumulative deviation sequence of each sub-time window based on the historical mean and each historical default rate sequence, and determining the standard deviation of each sub-time window; determining the range of each sub-time window based on the cumulative deviation sequence of each sub-time window; determining the rescaled range of each sub-time window based on the range and the standard deviation of each sub-time window; performing linear fitting on the logarithm of the rescaled range of each sub-time window and the logarithm of each sub-time window to obtain slope information, and determining the fractional order based on the slope information.

[0112] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: extracting customer features from customer information in loan requests, constructing a customer state vector sequence based on the customer features, and extracting an external environment feature sequence from external environment information, including: obtaining the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and extracting asset load features from the asset information; constructing a customer state vector sequence based on loan information, historical default information, asset load features, and credit risk score corresponding to different times within the preset time period; obtaining external environment information within a preset time period from preset channels, and processing the external environment information through a pre-trained semantic processing model to obtain an external environment feature sequence, wherein the external environment information includes: preset indicator growth values ​​and natural language text contained in preset websites.

[0113] The processor can access information and applications stored in memory via a transmission device to execute the following steps: responding to a loan request based on a predicted default probability value for a future time window, and obtaining a response result, including: numerically integrating the predicted default probability value for the future time window to obtain a target default probability corresponding to the loan request; triggering a risk warning if the target default probability is greater than a preset probability threshold, and sending the loan request to the target object, wherein the target object reviews the loan request and processes it based on the review result; if the target default probability is less than or equal to the preset probability threshold, adjusting the loan request based on the target customer's repayment behavior information and credit risk score, and determining the response result based on the adjusted loan request.

[0114] This application provides a method for responding to loan requests. By receiving a loan request initiated by a target customer, and with the target customer's authorization, customer features are extracted from the customer information in the loan request. A customer state vector sequence is constructed based on these features, and an external environment feature sequence is extracted from external environment information. A target solution equation is constructed based on the customer state vector sequence and the external environment feature sequence using a fractional diffusion equation. This target solution equation is solved at a preset time step to obtain a predicted default probability value for the loan request in a future time window. The loan request is responded to based on the predicted default probability value in the future time window, and the response result is sent to the target customer. This solves the technical problem that when reviewing loan requests by predicting default probability, the reliance on short-term historical data often leads to unstable prediction results, resulting in unstable response results for loan requests.

[0115] Through the above steps, with customer authorization, a customer state vector sequence and an external environment feature sequence are constructed. An objective solution equation is built based on the fractional-order credit risk diffusion equation. Fractional derivatives are used to model the long-range dependence and memory kernel effect of credit risk, enabling the model to capture the non-local, decaying lag effect of historical events on the current risk state. Even if short-term historical data of the customer is noisy, the long-term evolution path can be integrated through the memory mechanism of fractional derivatives, thereby improving the stability and robustness of default probability prediction. This avoids the problem of prediction instability caused by traditional Markov models that rely solely on the current state and ignore historical events. Therefore, it can solve the technical problem in related technologies where loan requests are reviewed based on predicted loan default probabilities, which often rely on short-term historical data, leading to unstable prediction results and consequently unstable loan request response results. This achieves the technical effect of reducing misjudgments caused by external environmental noise and improving the stability of credit risk prediction results.

[0116] 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.

[0117] 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.

[0118] Example 4

[0119] 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 loan request response method provided in Embodiment 1.

[0120] 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.

[0121] This application also provides a computer program product, which, when executed on a data processing device, is adapted to perform response method steps for a loan request.

[0122] 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.

[0123] 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.

[0124] 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.

[0125] 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.

[0126] 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.

[0127] 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.

[0128] 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 method for responding to a loan request, characterized in that, include: Upon receiving a loan request initiated by a target customer, and with the authorization of the target customer, extracting customer features from the customer information of the loan request, constructing a customer state vector sequence based on the customer features, and extracting an external environment feature sequence from external environment information; Based on the fractional diffusion equation, the target solution equation is constructed according to the customer state vector sequence and the external environment feature sequence. The target solution equation is solved with a preset time step to obtain the predicted default probability value for the loan request in the future time window. The loan request is responded to based on the predicted default probability value of the future time window, and the response result is sent to the target customer.

2. The method according to claim 1, characterized in that, Based on the fractional diffusion equation, a target solution equation is constructed using the customer state vector sequence and the external environment feature sequence. This target solution equation is solved at a preset time step to obtain a predicted default probability value for the loan request within a future time window, including: The customer state vector sequence is processed based on a preset algorithm to obtain a fractional order, wherein the preset algorithm is used to quantify the duration of the influence of features in the customer state vector sequence on loan behavior. The diffusion coefficient is determined based on the statistical characteristics of the customer state vector sequence and the duration of the influence of the characteristics in the customer state vector sequence. The external environment feature sequence is fitted using a historical regression model to obtain the external environment feature mapping relationship; Based on the grid discretization results of the customer state vector sequence, the fractional order, the diffusion coefficient, and the mapping relationship of the external environment features, the target solution equation is constructed using the finite difference scheme based on the fractional derivative and the central difference approximation. The target solution equation is then solved based on the preset time step to obtain the predicted default probability value for the future time window.

3. The method according to claim 2, characterized in that, Based on the grid discretization results of the customer state vector sequence, the fractional order, the diffusion coefficient, and the mapping relationship of the external environment features, the target solution equation is constructed using a finite difference scheme based on the fractional derivative and the central difference approximation. The target solution equation is then solved based on the preset time step to obtain the predicted default probability value for the future time window, including: The customer state vector sequence is discretized into a grid to obtain the probability distribution sequence of each grid point; The duration of the influence of features in the customer state vector sequence on each time step is determined using a finite difference scheme with fractional derivatives based on the probability distribution sequence of each grid point, the fractional order, and the preset time step. The diffusion characteristics of each spatial step are determined based on the probability distribution sequence of each grid point, the preset spatial step size, the diffusion coefficient, and the mapping relationship of the external environment characteristics. The objective solution equation is constructed based on the duration of the influence at each time step and the diffusion characteristics at each spatial step. Starting from the beginning of the future time window, the target equation is iteratively solved step by step with the preset time step to obtain the default probability prediction value corresponding to the time step, until the end of the future time window, to obtain the default probability prediction value of the future time window.

4. The method according to claim 2, characterized in that, The diffusion coefficient is determined based on the statistical characteristics of the customer state vector sequence and the duration of the influence of the characteristics in the customer state vector sequence, including: Determine the historical variance of the customer state vector sequence within a historical time window, wherein the historical variance characterizes the degree of fluctuation of the features in the customer state vector sequence; The duration of the influence of the features in the customer state vector sequence is determined based on the features in the customer state vector sequence within the historical time window and a preset threshold. The diffusion coefficient is determined based on the gamma function of the fractional order, the fractional order, the historical variance, and the duration of the influence of features in the customer state vector sequence.

5. The method according to claim 2, characterized in that, The customer state vector sequence is processed based on a preset algorithm to obtain a fractional order, including: Obtain the historical default rate of the target customer in multiple sub-time windows within the historical time window to obtain multiple historical default rate sequences, and determine the historical mean of each historical default rate sequence; Based on the historical mean of each historical default rate series and each historical default rate series, the cumulative deviation series for each sub-time window is determined, as well as the standard deviation for each sub-time window is determined; The range of each sub-time window is determined based on the cumulative deviation sequence of each sub-time window; The rescaled range of each sub-time window is determined based on the range and standard deviation of each sub-time window; Linear fitting is performed on the logarithm of the rescaled range for each sub-time window and the logarithm of each sub-time window to obtain slope information, and the fractional order is determined based on the slope information.

6. The method according to claim 1, characterized in that, Extracting customer features from the customer information of the loan request, constructing a customer state vector sequence based on the customer features, and extracting an external environment feature sequence from the external environment information, including: The system retrieves the target customer's credit risk score, loan information, historical default information, and asset information within a preset time period from a preset storage space, and extracts asset load characteristics from the asset information. The customer status vector sequence is constructed based on loan information, historical default information, asset and liability characteristics, and credit risk scores at different times within the preset time period. External environment information within a preset time period is obtained from a preset channel, and the external environment information is processed by a pre-trained semantic processing model to obtain the external environment feature sequence. The external environment information includes: preset indicator growth values ​​and natural language text contained in preset websites.

7. The method according to claim 1, characterized in that, The loan request is responded to based on the predicted default probability value for the future time window, and the response result includes: The predicted default probability for the future time window is numerically integrated to obtain the target default probability corresponding to the loan request. If the target default probability is greater than a preset probability threshold, a risk warning is triggered, and the loan request is sent to the target object. The target object reviews the loan request and processes it based on the review results. If the target default probability is less than or equal to the preset probability threshold, the loan request is adjusted based on the target customer's repayment behavior information and the target customer's credit risk score, and the response result is determined based on the adjusted loan request.

8. A loan request response device, characterized in that, include: The construction unit is used to receive a loan request initiated by a target customer, and, with the authorization of the target customer, extract customer features from the customer information of the loan request, construct a customer state vector sequence based on the customer features, and extract an external environment feature sequence from the external environment information. The prediction unit is used to solve the equation based on the fractional diffusion equation, construct the target solution equation according to the customer state vector sequence and the external environment feature sequence, solve the target solution equation with a preset time step, and obtain the predicted value of the default probability for the loan request in the future time window. The response unit is used to respond to the loan request based on the predicted default probability value of the future time window, obtain a response result, and send the response result to the target customer.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, executes the response method for a loan request as described in 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 response method for a loan request as described in any one of claims 1 to 7.

Images