Credit risk control method and device based on multi-source data fusion and reinforcement learning

Abstract

This invention discloses a credit risk control method and apparatus based on multi-source data fusion and reinforcement learning, relating to the fintech field or other related fields. The method includes: responding to a borrower's loan application; acquiring a credit data set from multiple data sources; preprocessing the credit data set; using an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data point in the credit data set; fusing all credit data based on the weight vector to obtain a target feature vector; inputting the target feature vector into a pre-constructed risk assessment model to obtain the model output, which includes a risk assessment score; approving the borrower's loan application when the rate of return on the target financial capital requested by the borrower exceeds a preset rate of return threshold; and determining the loan amount, interest rate, and term based on the risk assessment score. This invention solves the technical problem in related technologies where credit risk assessment cannot comprehensively and accurately reflect the borrower's creditworthiness.

Description

Technical Field

[0001] This invention relates to the field of financial technology or other related fields, and more specifically, to a credit risk control method and apparatus based on multi-source data fusion and reinforcement learning. Background Technology

[0002] With the rapid development of internet finance, the demand for credit is increasing, placing higher demands on the convenience and immediacy of credit services. Financial institutions are migrating their credit approval processes online, pursuing more efficient and faster approval mechanisms. However, while online and internet-based approaches have improved efficiency, they have also brought new challenges, especially in the control of credit risk. Accurately identifying and assessing risk amidst a massive volume of online credit applications is a crucial issue for financial institutions.

[0003] In related technologies, financial institutions' credit risk assessment strategies typically rely on a single data source and a fixed assessment model. A single data source cannot fully capture the borrower's credit status, and a fixed assessment model lacks flexibility and self-updating capabilities. In the context of the rapid digitalization of financial business and the increasingly complex market environment, these limitations are even more prominent, making it impossible to fully and accurately reflect the borrower's credit status.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] This invention provides a credit risk control method and apparatus based on multi-source data fusion and reinforcement learning, which at least solves the technical problem in related technologies that credit risk assessment cannot fully and accurately reflect the borrower's credit status.

[0006] To achieve the above objectives, according to one aspect of this application, a credit risk control method based on multi-source data fusion and reinforcement learning is provided. The method includes: responding to a borrower's loan application; obtaining a set of the borrower's credit data from multiple data sources and preprocessing the credit data set, wherein the credit data set includes at least: basic customer information, financial statements, credit records, and transaction records; using an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data point in the preprocessed credit data set, and fusing all credit data based on the weight vector to obtain a target feature vector; inputting the target feature vector into a pre-constructed risk assessment model to obtain the model output, wherein the model output includes at least: a risk assessment score; approving the borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and determining the loan amount, interest rate, and term based on the risk assessment score.

[0007] Optionally, the step of obtaining a set of borrowers' credit data from multiple data sources includes: obtaining basic customer information from the financial institution's internal information system or a third-party institution, wherein the basic customer information includes: the borrower's age, occupation, and income; obtaining financial statements from data files provided by the borrower or from a third-party institution, wherein the financial statements include: the borrower's assets and liabilities; obtaining credit records from the financial institution's internal information system or a third-party institution, wherein the credit records include: historical borrowing information, number of overdue payments, amount owed, and violation records; and obtaining transaction records from the financial institution's internal information system, wherein the transaction records include: the borrower's historical transactions, transaction frequency, transaction amount, and fund flow information.

[0008] Optionally, the data types of various credit data in the credit dataset include: numerical data and categorical data. The preprocessing steps for the credit dataset include: removing duplicate data from the various credit data in the credit dataset; the steps for removing duplicate data include: calculating the similarity value between any two credit data points; removing either data point if the similarity value is greater than a preset similarity threshold; filling missing values ​​in the credit data using the mean, median, or mode; encoding the categorical data if it is categorical data and converting it into numerical data; standardizing the numerical data if it is numerical data and mapping it to a target interval; and normalizing the various credit data in the credit dataset.

[0009] Optionally, the steps of determining the weight vectors of various credit data in the credit dataset using the hierarchical analysis strategy include: obtaining the importance ratios between credit data, constructing a judgment matrix based on the importance ratios, and normalizing the judgment matrix; calculating the eigenvectors of the judgment matrix, normalizing the eigenvectors to obtain the weight vectors of the credit data, wherein the weight vectors corresponding to each credit data are different; calculating the consistency ratio of the weight vectors, and adjusting the judgment matrix and recalculating the weight vectors if the consistency ratio is greater than a preset index threshold, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix.

[0010] Optionally, the risk assessment model is trained as follows: The risk assessment model is initialized using a reinforcement learning algorithm, defining a state space, action space, reward function, and Q-value. The state space includes a target feature vector; the action space includes actions such as approving a loan, rejecting a loan, and adjusting the loan amount; the reward function includes expected return and expected risk; and the Q-value is the expected return of selecting the target action in the current state. The risk assessment model is then trained based on historical target feature vectors. The training steps include: inputting historical target feature vectors into the risk assessment model; updating the risk assessment model based on the output Q-value; and terminating the training of the risk assessment model when the Q-value reaches the convergence condition.

[0011] Optionally, the step of inputting the target feature vector into the risk assessment model and calculating the risk assessment score includes: inputting the target feature vector into the risk assessment model to obtain the default risk probability and default risk loss; obtaining the borrower's historical credit score; and calculating the risk assessment score based on the default risk probability, default risk loss, and historical credit score.

[0012] Optionally, the method further includes: obtaining the target financial capital applied for by the borrower, and extracting the capital loan interest rate and capital loan term from the model output; calculating the expected return based on the target financial capital, the capital loan interest rate, and the capital loan term; obtaining the loan amount cost based on the target financial capital, obtaining the loan base cost from the financial institution's internal information system, and calculating the operating cost based on the loan amount cost and the loan base cost; calculating the risk exposure value based on the target financial capital and the probability of default, wherein the risk exposure value is used to measure the risk of lending by the financial institution; obtaining the risk-free interest rate value from the financial institution's internal information system, wherein the risk-free interest rate value is used to measure the return obtained by the financial institution under risk-free conditions; and calculating the rate of return on capital of the target financial capital applied for by the borrower based on the expected return, operating cost, risk exposure value, and risk-free interest rate value.

[0013] Optionally, the steps of determining the loan amount, interest rate, and term based on the risk assessment score include: obtaining the benchmark loan amount from the model output; obtaining the benchmark interest rate and benchmark term from the financial institution's internal information system; and determining the loan amount, interest rate, and term based on the benchmark loan amount, benchmark interest rate, and benchmark term, according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

[0014] According to another aspect of the present invention, a credit risk control device based on multi-source data fusion and reinforcement learning is also provided. The device includes: an acquisition unit, configured to respond to a borrower's loan application, acquire a set of the borrower's credit data from multiple data sources, and preprocess the credit data set, wherein the credit data set includes at least: basic customer information, financial statements, credit records, and transaction records; a fusion unit, configured to use an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data in the preprocessed credit data set, and fuse all credit data based on the weight vector to obtain a target feature vector; an input unit, configured to input the target feature vector into a pre-constructed risk assessment model to obtain the model output result, wherein the model output result includes at least: a risk assessment score; and a decision unit, configured to approve the borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and determine the loan amount, interest rate, and term based on the risk assessment score.

[0015] Optionally, the acquisition unit includes: a first acquisition module for acquiring basic customer information from the financial institution's internal information system or a third-party institution, wherein the basic customer information includes: the borrower's age, occupation, and income; a second acquisition module for acquiring financial statements from data files provided by the borrower or from a third-party institution, wherein the financial statements include: the borrower's assets and liabilities; a third acquisition module for acquiring credit records from the financial institution's internal information system or a third-party institution, wherein the credit records include: historical lending information, number of overdue payments, amount owed, and violation records; and a fourth acquisition module for acquiring transaction records from the financial institution's internal information system, wherein the transaction records include: the borrower's historical transaction history, transaction frequency, transaction amount, and fund flow information.

[0016] Optionally, the acquisition unit further includes: a first preprocessing module, used to process various types of credit data in the credit data set, including numerical data and categorical data, to remove duplicate data from the various types of credit data in the credit data set. The steps for removing duplicate data include: calculating the similarity value between any two pieces of credit data; removing any one of the two pieces of data if the similarity value is greater than a preset similarity threshold; and filling missing values ​​in the credit data with the mean, median, or mode; a second preprocessing module, used to encode the categorical data if the credit data is categorical data, and convert the categorical data into numerical data; a third preprocessing module, used to standardize the numerical data if the credit data is numerical data, and map the numerical data to a target interval; and a fourth preprocessing module, used to normalize the various types of credit data in the credit data set.

[0017] Optionally, the fusion unit includes: a construction module for obtaining the importance ratio between credit data, constructing a judgment matrix based on the importance ratio, and normalizing the judgment matrix; a first calculation module for calculating the eigenvector of the judgment matrix, normalizing the eigenvector to obtain the weight vector of the credit data, wherein the weight vector corresponding to each credit data is different; and a second calculation module for calculating the consistency ratio of the weight vector, adjusting the judgment matrix and recalculating the weight vector if the consistency ratio is greater than a preset index threshold, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix.

[0018] Optionally, the input unit includes: a definition module, used to initialize the risk assessment model using a reinforcement learning algorithm, defining the state space, action space, reward function, and Q-value, wherein the state space includes: target feature vector, the action space includes: loan approval action, loan rejection action, and loan amount adjustment action, the reward function includes: expected return and expected risk, and the Q-value is the expected return of selecting the target action in the current state; and a training module, used to train the risk assessment model based on historical target feature vectors, the training steps including: inputting historical target feature vectors into the risk assessment model, updating the risk assessment model based on the output Q-value, and ending the training of the risk assessment model when the Q-value reaches the convergence condition.

[0019] Optionally, the input unit further includes: a fifth acquisition module for inputting the target feature vector into the risk assessment model to obtain the default risk probability and default risk loss; a sixth acquisition module for acquiring the borrower's historical credit score; and a third calculation module for calculating the risk assessment score based on the default risk probability, default risk loss, and historical credit score.

[0020] Optionally, the decision-making unit includes: a fourth calculation module, used to obtain the target financial capital applied for by the borrower, and extract the capital loan interest rate and capital loan term from the model output results, and calculate the expected return based on the target financial capital, capital loan interest rate, and capital loan term; a fifth calculation module, used to obtain the loan amount cost based on the target financial capital, obtain the loan base cost from the financial institution's internal information system, and calculate the operating cost based on the loan amount cost and the loan base cost; a sixth calculation module, used to calculate the risk exposure value based on the target financial capital and the probability of default risk, wherein the risk exposure value is used to measure the risk of lending by the financial institution; a seventh acquisition module, used to obtain the risk-free interest rate value from the financial institution's internal information system, wherein the risk-free interest rate value is used to measure the return obtained by the financial institution under risk-free conditions; and a seventh calculation module, used to calculate the rate of return on capital of the target financial capital applied for by the borrower based on the expected return, operating cost, risk exposure value, and risk-free interest rate value.

[0021] Optionally, the decision-making unit further includes: an eighth acquisition module for acquiring the benchmark loan amount from the model output; a ninth acquisition module for acquiring the benchmark interest rate and benchmark term from the financial institution's internal information system; and a determination module for determining the loan amount, interest rate, and term based on the benchmark loan amount, benchmark interest rate, and benchmark term, according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

[0022] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the credit risk control method based on multi-source data fusion and reinforcement learning as described above.

[0023] According to another aspect of the present invention, an electronic device is also 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 the credit risk control method based on multi-source data fusion and reinforcement learning as described above.

[0024] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the credit risk control method based on multi-source data fusion and reinforcement learning as described above.

[0025] In this disclosure, in response to a borrower's loan application, a set of the borrower's credit data is obtained from multiple data sources. This set of credit data is preprocessed, and a hierarchical analysis strategy is used to determine the weight vector of each credit data point in the preprocessed set. Based on these weight vectors, all credit data are fused to obtain a target feature vector. This target feature vector is then input into a pre-constructed risk assessment model to obtain the model output, which includes at least a risk assessment score. Finally, if the return on the borrower's target financial capital exceeds a preset return threshold, the loan application is approved, and the loan amount, interest rate, and term are determined based on the risk assessment score. This disclosure, by fusing multi-source credit data, can form a comprehensive and multi-dimensional credit profile of the borrower. By using a risk assessment model to predict lending risk, it can accurately assess the borrower's creditworthiness, thus solving the technical problem in related technologies where credit risk assessment cannot comprehensively and accurately reflect the borrower's creditworthiness. Attached Figure Description

[0026] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0027] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a credit risk control method based on multi-source data fusion and reinforcement learning is shown.

[0028] Figure 2 This is a flowchart of an optional credit risk control method based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of another optional credit risk control method based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of an optional credit risk control device based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention;

[0031] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] It should be noted that the credit risk control method and apparatus based on multi-source data fusion and reinforcement learning disclosed herein can be used in the field of fintech for credit risk control based on multi-source data fusion and reinforcement learning, and can also be used in any field other than fintech for credit risk control based on multi-source data fusion and reinforcement learning. This disclosure does not limit the application field of the credit risk control method and apparatus based on multi-source data fusion and reinforcement learning.

[0035] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected in this public disclosure are 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 related data all comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or organization.

[0036] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0037] The following embodiments of the present invention can be applied to various systems / applications / devices for credit risk control based on multi-source data fusion and reinforcement learning. The present invention obtains a set of borrower credit data from multiple data sources, preprocesses the credit data set, and then uses an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data point in the preprocessed credit data set. Based on the weight vector, all credit data are fused to obtain a target feature vector, thereby forming a comprehensive and multi-dimensional borrower credit profile. The target feature vector is then input into a pre-built risk assessment model, which can accurately assess the borrower's creditworthiness. If the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, the borrower's loan application is approved, and the loan amount, interest rate, and term are determined based on the risk assessment score, completing a comprehensive and accurate lending process.

[0038] The present invention will now be described in detail with reference to various embodiments.

[0039] Example 1

[0040] According to an embodiment of the present invention, an embodiment of a credit risk control method based on multi-source data fusion and reinforcement learning is 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.

[0041] The credit risk control method based on multi-source data fusion and reinforcement learning provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a credit risk control method based on multi-source data fusion and reinforcement learning is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 The processor 102 (which may include, but is not limited to, a microprocessor MCU (Microcontroller Unit) or a programmable logic device FPGA (Field Programmable Gate Array)) is illustrated using 102a, 102b, ..., 102n. It also includes a memory 104 for storing data and a transmission device 106 for communication functions. In addition, it may 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.

[0042] 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).

[0043] 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 credit risk control method based on multi-source data fusion and reinforcement learning in this embodiment of the application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned credit risk control method based on multi-source data fusion and reinforcement learning. 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.

[0044] 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.

[0045] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0046] Under the aforementioned operating environment, this application provides the following: Figure 2 The credit risk control method shown is based on multi-source data fusion and reinforcement learning. Figure 2 This is a flowchart of an optional credit risk control method based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0047] Step S201: In response to the loan application initiated by the borrower, obtain the borrower's credit data set from multiple data sources and preprocess the credit data set. The credit data set includes at least: customer basic information, financial statements, credit records and transaction records.

[0048] In this embodiment of the invention, after a borrower initiates a loan application, the invention responds to the application by acquiring a set of credit data related to the borrower from multiple data sources. The aim is to construct a comprehensive and detailed credit profile to support more accurate risk assessment. After collecting the credit data set, it undergoes data preprocessing to ensure data quality and usability.

[0049] Multi-source data fusion refers to the collection and processing of data from different channels, formats, and qualities to form a comprehensive, multi-dimensional, and accurate dataset. In credit risk control, multi-source data includes internal data from financial institutions, credit reporting data, social network data, and e-commerce platform transaction data. This data can provide information about borrowers' financial status, credit history, social behavior, consumption habits, and other aspects, thus reflecting their creditworthiness more comprehensively. Multi-source data fusion strategies can overcome the information limitations of single data sources, improving the accuracy and comprehensiveness of credit risk assessment. Furthermore, preprocessing the collected multi-source data can effectively remove noise and redundant information, providing high-quality data support for subsequent risk assessment models.

[0050] In this embodiment of the invention, the multi-source data obtained from multiple data sources includes: basic customer information, financial statements, credit records, and transaction records. These data constitute a credit data set, providing a solid data foundation for subsequent risk assessment and risk control. It should be noted that the multi-source data in this embodiment is not limited to basic customer information, financial statements, credit records, and transaction records; it can also include any other data that helps assess borrower lending risk. This comprehensive use of multi-source data not only improves the comprehensiveness and accuracy of credit risk assessment but also helps identify potential credit risk points, thereby assisting financial institutions in generating more reasonable credit policies, effectively controlling credit risk, and improving the quality of credit business.

[0051] Step S202: The weight vector of each credit data in the preprocessed credit data set is determined by the hierarchical analysis strategy, and all credit data are fused based on the weight vector to obtain the target feature vector.

[0052] In the collected multi-source data, each type of data (including basic customer information, financial statements, credit records, and transaction records) comes from an independent data source. In order to effectively integrate these credit data from different channels and with different properties, this embodiment of the invention uses hierarchical analysis to determine the weight vector of various credit data in the credit data set, and then gives each type of credit data an appropriate degree of importance during the fusion process, so as to ensure that the final target feature vector is both comprehensive and accurate.

[0053] After determining the weight vector of each credit data point in the credit dataset, a weighted average strategy is used to fuse all credit data to obtain the target feature vector. The fusion formula is as follows:

[0054]

[0055] Where F is the target feature vector, n is the number of credit data, and D i Let w be the feature vector of the i-th credit data. i Let be the weight vector for the i-th credit data. It should be noted that the feature vector of the credit data is obtained after feature extraction. The feature extraction strategy depends on the type of credit data and can be implemented using deep learning models or mathematical transformations.

[0056] Step S203: Input the target feature vector into the pre-built risk assessment model to obtain the model output results, wherein the model output results include at least the risk assessment score.

[0057] After collecting, preprocessing, and integrating multi-source credit data to form a comprehensive and accurate target feature vector, this vector is input into a pre-built risk assessment model to generate a quantitative evaluation of the borrower's credit risk. Once the target feature vector is input into the model, the risk assessment model begins operation. Combining past learning experience with the input target feature vector, the model calculates a risk assessment score for each loan application. The risk assessment score is derived from an analysis of the borrower's comprehensive characteristics. Based on multi-source data such as the borrower's personal information, financial situation, credit history, and transaction history, the risk assessment model predicts the borrower's probability of default and potential losses, calculating the risk assessment score based on these predictions. A higher risk assessment score indicates a lower risk level and greater suitability for a loan; a lower risk assessment score indicates a higher risk level, requiring careful consideration of lending strategies.

[0058] Step S204: If the rate of return on the target financial capital applied for by the borrower is greater than the preset rate of return threshold, the borrower's loan application is approved, and the loan amount, interest rate and term are determined based on the risk assessment score.

[0059] In this embodiment of the invention, the indicator for determining whether to approve a loan to a borrower is the rate of return on capital. The rate of return on capital is an indicator that measures credit risk and return. This embodiment of the invention obtains the rate of return on capital of the target financial capital applied for by the borrower and compares it with a preset rate of return threshold. If the rate of return on capital is greater than the preset rate of return threshold, the borrower's loan application is approved; if the rate of return on capital is not greater than the preset rate of return threshold, the borrower's loan application is rejected.

[0060] After approving a loan application, this embodiment of the invention dynamically adjusts key conditions such as loan amount, interest rate, and term based on a risk assessment score—a quantitative assessment of the borrower's credit risk. This dynamic adjustment mechanism for loan conditions based on risk assessment scores enables this embodiment of the invention to effectively control credit risk and achieve optimal allocation of credit resources while ensuring profit targets. Through this dynamic adjustment mechanism, this embodiment of the invention can provide personalized and differentiated loan services according to each borrower's specific credit situation, meeting the needs of borrowers while ensuring the safe and efficient operation of financial institutions' credit business.

[0061] Optionally, the step of obtaining a set of borrowers' credit data from multiple data sources includes: obtaining basic customer information from the financial institution's internal information system or a third-party institution, wherein the basic customer information includes: the borrower's age, occupation, and income; obtaining financial statements from data files provided by the borrower or from a third-party institution, wherein the financial statements include: the borrower's assets and liabilities; obtaining credit records from the financial institution's internal information system or a third-party institution, wherein the credit records include: historical borrowing information, number of overdue payments, amount owed, and violation records; and obtaining transaction records from the financial institution's internal information system, wherein the transaction records include: the borrower's historical transactions, transaction frequency, transaction amount, and fund flow information.

[0062] In step S201, to comprehensively and accurately assess the risk of borrowers, this embodiment of the invention obtains a set of credit data of borrowers from multiple data sources. First, basic customer information is obtained from the internal information system of financial institutions or third-party institutions, which includes the borrower's age, occupation, and income. Second, financial statements are obtained from data files provided by the borrower or from third-party institutions, which include the borrower's assets and liabilities. Third, credit records are obtained from the internal information system of financial institutions or third-party institutions, which include the borrower's historical borrowing information, number of overdue payments, amount owed, and violation records. Finally, transaction records are obtained from the internal information system of financial institutions, which include the borrower's historical transactions, transaction frequency, transaction amount, and fund flow information.

[0063] This invention, through comprehensive and systematic collection and organization of borrowers' basic customer information, financial statement data, credit records, and transaction details, constructs a multi-dimensional credit data set, providing a solid data foundation for subsequent credit risk assessment and control. This process involves not only in-depth data mining within financial institutions' internal systems but also extensive cooperation with external institutions (third-party organizations), ensuring the richness and accuracy of the credit data set. This enables financial institutions to provide more scientific and refined risk assessment services for their credit operations.

[0064] It should be noted that, according to the characteristics of different data sources, the frequency and scope of data collection can also be set. For example, basic customer information can be updated regularly at long intervals, while transaction records and credit records need to be updated in real time or at high frequency to ensure the timeliness, integrity and accuracy of the data.

[0065] Optionally, the data types of various credit data in the credit dataset include: numerical data and categorical data. The preprocessing steps for the credit dataset include: removing duplicate data from the various credit data in the credit dataset; the steps for removing duplicate data include: calculating the similarity value between any two credit data points; removing either data point if the similarity value is greater than a preset similarity threshold; filling missing values ​​in the credit data using the mean, median, or mode; encoding the categorical data if it is categorical data and converting it into numerical data; standardizing the numerical data if it is numerical data and mapping it to a target interval; and normalizing the various credit data in the credit dataset.

[0066] In step S201, the collected credit data includes numerical data and categorical data. For example, a borrower's income and assets are numerical data, while their occupation and gender are categorical data. During preprocessing of the credit data, duplicate data removal is performed first. This embodiment of the invention determines whether duplicate data exists by calculating the similarity value between two data sets. The similarity calculation is performed separately for numerical data and categorical data. For numerical data, Euclidean distance or cosine similarity can be used to measure the closeness between two data sets; for categorical data, Hamming distance or Jaccard distance can be used. When determining whether duplicate data exists, a preset similarity threshold is typically set, such as 0.85. If the calculated similarity value is greater than the preset similarity threshold, they are considered duplicate data, and either data set needs to be removed, retaining only one data set to maintain the purity of the credit data.

[0067] After removing duplicate data from credit data, missing values ​​need to be imputed to ensure the integrity of the credit data. For numerical data, the mean or median can be used to impute missing values; for categorical data, the mode can be used.

[0068] After populating the credit data, categorized data needs to be encoded, which can be done using hot coding to convert categorized data into numerical data. For numerical data, standardization is required to map the numerical data to a target range. The standardization formula is as follows:

[0069]

[0070] Where x′ represents the standardized credit data, x represents the credit data, and x' represents the credit data. min x is the minimum value in each credit dataset. max This represents the maximum value in each set of credit data. This calculation formula maps numerical data to the interval [0,1] to facilitate subsequent processing of credit data of different magnitudes.

[0071] Finally, the credit data is normalized. The calculation formula for normalization in this embodiment of the invention is as follows:

[0072]

[0073] Where x″ represents the normalized credit data, x′ represents the credit data after deduplication, padding, and standardization (or encoding), μ represents the mean of each credit data set, and σ represents the standard deviation of each credit data set. Normalization ensures that the credit data are on the same scale, which facilitates the processing of subsequent risk assessment models.

[0074] Through a series of preprocessing operations, the accuracy and consistency of credit data were ensured, providing a solid data foundation for subsequent risk assessment.

[0075] It should be noted that credit datasets include not only numerical and categorical data, but also other types of data, such as time-series or text data. After preprocessing the time-series or text data, these data can be merged with the numerical and categorical data.

[0076] Optionally, the steps of determining the weight vectors of various credit data in the credit dataset using the hierarchical analysis strategy include: obtaining the importance ratios between credit data, constructing a judgment matrix based on the importance ratios, and normalizing the judgment matrix; calculating the eigenvectors of the judgment matrix, normalizing the eigenvectors to obtain the weight vectors of the credit data, wherein the weight vectors corresponding to each credit data are different; calculating the consistency ratio of the weight vectors, and adjusting the judgment matrix and recalculating the weight vectors if the consistency ratio is greater than a preset index threshold, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix.

[0077] In step S202, when determining the weight vectors of various credit data in the credit dataset using the analytic hierarchy process (AHP) strategy, it is first necessary to obtain the importance ratios between the credit data. Importance ratios are typically set empirically, determining the importance ratio of one type of credit data relative to another by comparing their respective importance. A scaling strategy is usually used when determining importance ratios, where "1" represents that the two types of credit data are equally important, "9" represents that one type of credit data is far more important than the other, and values ​​between "0" and "9" reflect different degrees of importance difference.

[0078] Based on the determined importance ratio, a judgment matrix A is constructed, and the judgment matrix is ​​as follows:

[0079]

[0080] Where n is the number of types of credit data, A ij Let i be the importance ratio of the i-th type of credit data to the j-th type of credit data, for example, "A ij =3" indicates that the importance of the i-th type of credit data is 3 times that of the j-th type of credit data, "A ij =1 / 3” means that the importance of the i-th type of credit data is 1 / 3 times that of the j-th type of credit data.

[0081] After constructing the judgment matrix, the judgment matrix is ​​normalized, and then the eigenvectors of the judgment matrix, i.e., the relative weights of various credit data, are calculated. The calculation formula is as follows:

[0082]

[0083] Among them, W i Let B be the relative weight of the i-th type of credit data. ij To determine the j-th element in the i-th row of the matrix, where n is the number of elements in the i-th row (i.e., the number of types of credit data).

[0084] Then, based on the relative weights of all credit data, the relative weights of various credit data are normalized to obtain weight vectors for each type of credit data. Simultaneously, since the judgment matrix is ​​based on empirical settings, a consistency check is needed to ensure the logical consistency of the judgment matrix. During the consistency check, the largest eigenvalue of the judgment matrix is ​​first calculated, and then the consistency index is calculated. The formula for calculating the consistency index is:

[0085]

[0086] Where CI is the consistency index, and λ maxTo determine the largest eigenvalue of the matrix, n represents the number of credit data categories. After calculating the consistency index, the consistency ratio is calculated using the following formula:

[0087]

[0088] Wherein, CR is the consistency ratio, CI is the consistency index, and RI is the random consistency index. The random consistency index is obtained from the analytic hierarchy process (AHP) strategy standard table based on the number of categories, n. After calculating the consistency ratio, it is compared with a preset index threshold, which can be set to 0.1. When the consistency ratio is less than or equal to the preset index threshold, the consistency check is considered passed. When the consistency ratio is greater than the preset index threshold, the judgment matrix needs to be adjusted, and the weight vector of each credit data needs to be recalculated.

[0089] Different credit data have varying degrees of impact on credit risk, and their weight vectors differ. By employing an analytic hierarchy process (AHP) strategy to calculate these different weight vectors, the impact of multi-source data on the risk assessment model is comprehensively considered, resulting in a more reasonable credit risk assessment.

[0090] Optionally, the risk assessment model is trained as follows: The risk assessment model is initialized using a reinforcement learning algorithm, defining a state space, action space, reward function, and Q-value. The state space includes a target feature vector; the action space includes actions such as approving a loan, rejecting a loan, and adjusting the loan amount; the reward function includes expected return and expected risk; and the Q-value is the expected return of selecting the target action in the current state. The risk assessment model is then trained based on historical target feature vectors. The training steps include: inputting historical target feature vectors into the risk assessment model; updating the risk assessment model based on the output Q-value; and terminating the training of the risk assessment model when the Q-value reaches the convergence condition.

[0091] In step S203, the risk assessment model is pre-built and trained. First, a reinforcement learning algorithm is used to initialize the risk assessment model, defining the state space, action space, reward function, and Q-value. The state space includes the target feature vector, the action space includes loan actions such as approving a loan, rejecting a loan, and adjusting the loan amount, the Q-value is the expected return of choosing the target action in the current state, and the reward function includes expected return and expected risk, both of which are outputs of the risk assessment model.

[0092] When training the risk assessment model, the historical target feature vector is input into the risk assessment model, and the risk assessment model is updated based on the output Q value. This update process is continuously iterated, and the risk assessment model gradually maximizes the lending return. Finally, when the Q value reaches the convergence condition, the training of the risk assessment model ends, and the trained risk assessment model is then used in real-world scenarios.

[0093] Optionally, the step of inputting the target feature vector into the risk assessment model and calculating the risk assessment score includes: inputting the target feature vector into the risk assessment model to obtain the default risk probability and default risk loss; obtaining the borrower's historical credit score; and calculating the risk assessment score based on the default risk probability, default risk loss, and historical credit score.

[0094] In step S203, the risk assessment score is calculated based on the output of the risk assessment model. In this embodiment of the invention, after inputting the target feature vector into the risk assessment model, the borrower's default risk probability and default risk loss are predicted. Simultaneously, this embodiment of the invention obtains the borrower's historical credit score, and then calculates the risk assessment score based on the default risk probability, default risk loss, and historical credit score. The calculation formula is as follows:

[0095]

[0096] In this formula, Score represents the risk assessment score, P represents the predicted probability of default, L represents the predicted loss due to default, R represents the borrower's historical credit score, and α, β, and γ are the weighting coefficients for the probability of default, the loss due to default, and the historical credit score, respectively. These weighting coefficients are set by the user. It should be noted that R in this formula does not have to be the historical credit score; other credit risk factors that can measure the borrower's historical credit can be used. Through this formula, the risk assessment score is controlled within the range of 0 to 1, used to measure the level of risk of the borrower.

[0097] Optionally, the method further includes: obtaining the target financial capital applied for by the borrower, and extracting the capital loan interest rate and capital loan term from the model output; calculating the expected return based on the target financial capital, the capital loan interest rate, and the capital loan term; obtaining the loan amount cost based on the target financial capital, obtaining the loan base cost from the financial institution's internal information system, and calculating the operating cost based on the loan amount cost and the loan base cost; calculating the risk exposure value based on the target financial capital and the probability of default, wherein the risk exposure value is used to measure the risk of lending by the financial institution; obtaining the risk-free interest rate value from the financial institution's internal information system, wherein the risk-free interest rate value is used to measure the return obtained by the financial institution under risk-free conditions; and calculating the rate of return on capital of the target financial capital applied for by the borrower based on the expected return, operating cost, risk exposure value, and risk-free interest rate value.

[0098] In step S204, the rate of return on capital is calculated based on expected returns, operating costs, risk exposure, and the risk-free interest rate, using the following formula:

[0099]

[0100] In the above formula, RAROC is the return on capital, R is the expected return, C is the operating cost, E is the risk exposure value, and r is the risk-free rate.

[0101] In the calculation parameters of return on capital, the expected return is calculated based on the target financial capital, the capital loan interest rate, and the capital loan term. First, the target financial capital applied for by the borrower is obtained. Then, the capital loan interest rate and capital loan term are extracted from the output of the risk assessment model. The formula for calculating the expected return is as follows:

[0102] R = H × r loan ×T

[0103] In the above formula, H represents the target financial capital, and r loan t is the capital loan interest rate, and T is the capital loan term.

[0104] Operating costs are calculated based on the cost of loan amount and the basic cost of loan. First, the cost of loan amount is obtained based on the target financial capital. Then, the basic cost of loan is obtained from the financial institution's internal information system. Finally, the operating cost is calculated based on the cost of loan amount and the basic cost of loan. The calculation formula is as follows:

[0105] C = C fixed +C var ×H

[0106] In the above formula, C represents the operating cost, and H represents the target financial capital. fixed For the base cost of the loan, C var This refers to the variable costs of loans. These variable costs are also obtained from the financial institution's internal information system.

[0107] The risk exposure value is calculated based on the target financial capital and the probability of default risk output by the risk assessment model. The calculation formula is as follows:

[0108] E = H × P

[0109] In the above formula, E represents risk exposure, H represents target financial capital, and P represents the probability of default. Risk exposure is one of the indicators used to measure the lending risk of financial institutions.

[0110] The risk-free rate is a metric used to measure the return a financial institution can obtain in a risk-free environment, and it is usually obtained from the financial institution's internal information system.

[0111] Optionally, the steps of determining the loan amount, interest rate, and term based on the risk assessment score include: obtaining the benchmark loan amount from the model output; obtaining the benchmark interest rate and benchmark term from the financial institution's internal information system; and determining the loan amount, interest rate, and term based on the benchmark loan amount, benchmark interest rate, and benchmark term, according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

[0112] In step S204, if the rate of return on capital is greater than a preset rate of return threshold, the borrower's loan application is approved, and the loan amount, interest rate, and term are determined based on the risk assessment score. Specifically, first, a benchmark loan amount is obtained from the output of the risk assessment model, and then a benchmark interest rate and benchmark term are obtained from the financial institution's internal information system. When determining the borrower's loan amount, the benchmark loan amount is used as a basis; the higher the risk assessment score, the larger the approved loan amount, and the lower the risk assessment score, the smaller the approved loan amount. When determining the borrower's loan term, the benchmark term is used as a basis; the higher the risk assessment score, the longer the approved loan term, and the lower the risk assessment score, the shorter the approved loan term. When determining the borrower's loan interest rate, the benchmark interest rate is used as a basis; the higher the risk assessment score, the lower the approved loan interest rate, and the lower the risk assessment score, the higher the approved loan interest rate.

[0113] Figure 3 This is a schematic diagram of another optional credit risk control method based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention, such as... Figure 3 As shown, this method first obtains multi-source data from the internal systems of financial institutions or third-party institutions. Then, it performs preprocessing on the acquired multi-source data, including removing duplicate data, filling missing values, standardization, and normalization. Next, it uses an analytic hierarchy process (AHP) strategy to determine the weight vectors of the preprocessed multi-source data. Based on these weight vectors, the multi-source data is fused, and the resulting target feature vector is input into a pre-constructed risk assessment model to obtain model output results, including a risk assessment score. Finally, it obtains the borrower's return on capital and determines whether to approve the loan based on the return on capital. Then, based on the risk assessment score, it determines the loan amount, interest rate, and term. Finally, this embodiment of the invention stores the assessment results using a distributed system and database and displays the results in the form of charts.

[0114] This invention, by fusing multi-source data, can form a comprehensive and multi-dimensional credit profile of borrowers. Employing reinforcement learning algorithms to construct a risk assessment model, it can more accurately predict borrowers' default risk. Based on this strategy, this invention can automatically screen high-quality borrowers, identify potential risky borrowers, and provide financial institutions with detailed assessment reports and recommendation strategies, reducing the need for manual intervention and improving the efficiency of credit processing. By identifying potential risky borrowers and taking timely risk control measures, this invention can reduce the overall risk level of financial institutions' credit operations. Furthermore, this invention supports customizable risk thresholds and control rules, which can be adjusted and optimized according to the different business needs and risk preferences of financial institutions.

[0115] The aforementioned credit risk control method based on multi-source data fusion and reinforcement learning involves the following steps: responding to a borrower's loan application, acquiring a set of the borrower's credit data from multiple data sources, preprocessing the credit data set, using an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data point in the preprocessed set, and fusing all credit data based on the weight vector to obtain a target feature vector. This target feature vector is then input into a pre-constructed risk assessment model to obtain the model output, which includes at least a risk assessment score. Finally, if the return on capital for the borrower's target financial capital exceeds a preset return threshold, the loan application is approved, and the loan amount, interest rate, and term are determined based on the risk assessment score. This invention, by fusing multi-source credit data, can form a comprehensive and multi-dimensional borrower credit profile. By using a risk assessment model to predict lending risk, it can accurately assess the borrower's creditworthiness, thus solving the technical problem in related technologies where credit risk assessment cannot comprehensively and accurately reflect the borrower's creditworthiness.

[0116] The following is a detailed description with reference to another embodiment.

[0117] Example 2

[0118] The credit risk control device based on multi-source data fusion and reinforcement learning provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in Embodiment 1 above.

[0119] Figure 4 This is a schematic diagram of an optional credit risk control device based on multi-source data fusion and reinforcement learning according to an embodiment of the present invention, such as... Figure 4 As shown, the credit risk control device based on multi-source data fusion and reinforcement learning may include: an acquisition unit 41, a fusion unit 42, an input unit 43, and a decision unit 44.

[0120] The acquisition unit 41 is used to respond to a loan application initiated by a borrower, acquire a set of credit data of the borrower from multiple data sources, and preprocess the credit data set. The credit data set includes at least: basic customer information, financial statements, credit records and transaction records.

[0121] The fusion unit 42 is used to determine the weight vector of each credit data in the preprocessed credit data set using the hierarchical analysis strategy, and to fuse all credit data based on the weight vector to obtain the target feature vector.

[0122] Input unit 43 is used to input the target feature vector into a pre-built risk assessment model to obtain the model output results, wherein the model output results include at least the risk assessment score.

[0123] Decision unit 44 is used to approve a borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and to determine the loan amount, interest rate and term based on the risk assessment score.

[0124] The aforementioned credit risk control device based on multi-source data fusion and reinforcement learning can respond to a borrower's loan application through the acquisition unit 41, acquire a set of the borrower's credit data from multiple data sources, preprocess the credit data set, and then use the fusion unit 42 to determine the weight vector of each credit data in the preprocessed credit data set using a hierarchical analysis strategy. Based on the weight vector, all credit data are fused to obtain a target feature vector. Then, the input unit 43 inputs the target feature vector into a pre-constructed risk assessment model to obtain the model output result, which includes at least a risk assessment score. Finally, the decision unit 44 approves the borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and determines the loan amount, interest rate, and term based on the risk assessment score. This embodiment of the invention, by fusing multi-source credit data, can form a comprehensive and multi-dimensional credit profile of the borrower, and uses a risk assessment model to predict lending risk, accurately assessing the borrower's credit status, thereby solving the technical problem in related technologies where credit risk assessment cannot comprehensively and accurately reflect the borrower's credit status.

[0125] Optionally, the acquisition unit 41 includes: a first acquisition module for acquiring basic customer information from the financial institution's internal information system or a third-party institution, wherein the basic customer information includes: the borrower's age, occupation, and income; a second acquisition module for acquiring financial statements from data files provided by the borrower or from a third-party institution, wherein the financial statements include: the borrower's assets and liabilities; a third acquisition module for acquiring credit records from the financial institution's internal information system or a third-party institution, wherein the credit records include: historical lending information, number of overdue payments, amount owed, and violation records; and a fourth acquisition module for acquiring transaction records from the financial institution's internal information system, wherein the transaction records include: the borrower's historical transaction business, transaction frequency, transaction amount, and fund flow information.

[0126] Optionally, the acquisition unit 41 further includes: a first preprocessing module, used to process various types of credit data in the credit data set, including numerical data and categorical data, to remove duplicate data from the various types of credit data in the credit data set. The steps for removing duplicate data include: calculating the similarity value between any two pieces of credit data; removing any one of the two pieces of data if the similarity value is greater than a preset similarity threshold; and filling missing values ​​in the credit data with the mean, median, or mode; a second preprocessing module, used to encode the categorical data if the credit data is categorical data, and convert the categorical data into numerical data; a third preprocessing module, used to standardize the numerical data if the credit data is numerical data, and map the numerical data to a target interval; and a fourth preprocessing module, used to normalize the various types of credit data in the credit data set.

[0127] Optionally, the fusion unit 42 includes: a construction module for obtaining the importance ratio between credit data, constructing a judgment matrix based on the importance ratio, and normalizing the judgment matrix; a first calculation module for calculating the feature vector of the judgment matrix, normalizing the feature vector to obtain the weight vector of the credit data, wherein the weight vector corresponding to each credit data is different; and a second calculation module for calculating the consistency ratio of the weight vector, adjusting the judgment matrix and recalculating the weight vector if the consistency ratio is greater than a preset index threshold, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix.

[0128] Optionally, the input unit 43 includes: a definition module, used to initialize the risk assessment model using a reinforcement learning algorithm, defining the state space, action space, reward function, and Q-value, wherein the state space includes: target feature vector, the action space includes: loan approval action, loan rejection action, and loan amount adjustment action, the reward function includes: expected return and expected risk, and the Q-value is the expected return of selecting the target action in the current state; and a training module, used to train the risk assessment model based on historical target feature vectors, the training steps including: inputting historical target feature vectors into the risk assessment model, updating the risk assessment model based on the output Q-value, and ending the training of the risk assessment model when the Q-value reaches the convergence condition.

[0129] Optionally, the input unit 43 further includes: a fifth acquisition module for inputting the target feature vector into the risk assessment model to obtain the default risk probability and default risk loss; a sixth acquisition module for acquiring the borrower's historical credit score; and a third calculation module for calculating the risk assessment score based on the default risk probability, default risk loss, and historical credit score.

[0130] Optionally, the decision-making unit 44 includes: a fourth calculation module, used to obtain the target financial capital applied for by the borrower, and extract the capital loan interest rate and capital loan term from the model output results, and calculate the expected return based on the target financial capital, capital loan interest rate, and capital loan term; a fifth calculation module, used to obtain the loan amount cost based on the target financial capital, obtain the loan base cost from the financial institution's internal information system, and calculate the operating cost based on the loan amount cost and the loan base cost; a sixth calculation module, used to calculate the risk exposure value based on the target financial capital and the probability of default risk, wherein the risk exposure value is used to measure the risk of lending by the financial institution; a seventh acquisition module, used to obtain the risk-free interest rate value from the financial institution's internal information system, wherein the risk-free interest rate value is used to measure the return obtained by the financial institution under risk-free conditions; and a seventh calculation module, used to calculate the rate of return on capital of the target financial capital applied for by the borrower based on the expected return, operating cost, risk exposure value, and risk-free interest rate value.

[0131] Optionally, the decision-making unit 44 further includes: an eighth acquisition module for acquiring the benchmark loan amount from the model output; a ninth acquisition module for acquiring the benchmark interest rate and benchmark term from the financial institution's internal information system; and a determination module for determining the loan amount, interest rate, and term based on the benchmark loan amount, benchmark interest rate, and benchmark term, according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

[0132] The credit risk control device based on multi-source data fusion and reinforcement learning may also include a processor and a memory. The acquisition unit 41, fusion unit 42, input unit 43, decision unit 44, etc. are all stored in the memory as program units, and the processor executes the program units stored in the memory to realize the corresponding functions.

[0133] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and credit risk control based on multi-source data fusion and reinforcement learning can be achieved by adjusting kernel parameters.

[0134] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0135] Example 3

[0136] Embodiments of this application may provide an electronic device. Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 Only one of the components 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.

[0137] 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.

[0138] The processor can access information and applications stored in memory via a transmission device to perform the following steps: responding to a borrower's loan application, obtaining a set of the borrower's credit data from multiple data sources, and preprocessing the credit data set, wherein the credit data set includes at least: basic customer information, financial statements, credit records, and transaction records; using an analytic hierarchy process (AHP) strategy to determine the weight vector of each credit data point in the preprocessed credit data set, and fusing all credit data based on the weight vector to obtain a target feature vector; inputting the target feature vector into a pre-built risk assessment model to obtain the model output, wherein the model output includes at least: a risk assessment score; approving the borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and determining the loan amount, interest rate, and term based on the risk assessment score.

[0139] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: obtaining basic customer information from the financial institution's internal information system or a third-party institution, including the borrower's age, occupation, and income; obtaining financial statements from data files provided by the borrower or from a third-party institution, including the borrower's assets and liabilities; obtaining credit records from the financial institution's internal information system or a third-party institution, including historical lending information, number of overdue payments, amount owed, and violation records; and obtaining transaction records from the financial institution's internal information system, including the borrower's historical transactions, transaction frequency, transaction amount, and fund flow information.

[0140] The processor can also access information and applications stored in the memory via a transmission device to execute the following steps: The credit data set contains various data types, including numerical data and categorical data. The processor removes duplicate data from the various credit data in the credit data set. The steps for removing duplicate data include: calculating the similarity value between any two credit data entries; removing any one of the two data entries if the similarity value exceeds a preset similarity threshold; filling missing values ​​in the credit data using the mean, median, or mode; encoding the categorical data if it is categorical data and converting it to numerical data; standardizing the numerical data if it is numerical data and mapping it to a target interval; and normalizing the various credit data in the credit data set.

[0141] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: obtain the importance ratio between credit data, construct a judgment matrix based on the importance ratio, and normalize the judgment matrix; calculate the eigenvector of the judgment matrix, normalize the eigenvector to obtain the weight vector of the credit data, wherein the weight vector corresponding to each credit data is different; calculate the consistency ratio of the weight vector, and if the consistency ratio is greater than a preset index threshold, adjust the judgment matrix and recalculate the weight vector, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix.

[0142] The processor can also access information and applications stored in the memory via a transmission device to execute the following steps: Initialize the risk assessment model using a reinforcement learning algorithm, defining the state space, action space, reward function, and Q-value. The state space includes a target feature vector; the action space includes actions such as approving a loan, rejecting a loan, and adjusting the loan amount; the reward function includes expected return and expected risk; and the Q-value is the expected return of selecting the target action in the current state. Train the risk assessment model based on historical target feature vectors. The training steps include: inputting historical target feature vectors into the risk assessment model and updating the risk assessment model based on the output Q-value; and ending the training of the risk assessment model when the Q-value reaches the convergence condition.

[0143] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: inputting the target feature vector into the risk assessment model to obtain the default risk probability and default risk loss; obtaining the borrower's historical credit score; and calculating the risk assessment score based on the default risk probability, default risk loss, and historical credit score.

[0144] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: obtaining the target financial capital applied for by the borrower, and extracting the capital loan interest rate and capital loan term from the model output; calculating the expected return based on the target financial capital, capital loan interest rate, and capital loan term; obtaining the loan amount cost based on the target financial capital, obtaining the loan base cost from the financial institution's internal information system, and calculating the operating cost based on the loan amount cost and loan base cost; calculating the risk exposure value based on the target financial capital and the probability of default, where the risk exposure value is used to measure the risk of lending by the financial institution; obtaining the risk-free interest rate value from the financial institution's internal information system, where the risk-free interest rate value is used to measure the return obtained by the financial institution under risk-free conditions; and calculating the rate of return on capital of the target financial capital applied for by the borrower based on the expected return, operating cost, risk exposure value, and risk-free interest rate value.

[0145] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: obtain the benchmark loan amount from the model output results; obtain the benchmark interest rate and benchmark term from the financial institution's internal information system; and determine the loan amount, interest rate, and term based on the benchmark loan amount, benchmark interest rate, and benchmark term, according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

[0146] This invention provides a credit risk control method based on multi-source data fusion and reinforcement learning. In response to a borrower's loan application, it acquires a set of the borrower's credit data from multiple data sources, preprocesses the data, and then uses an analytic hierarchy process (AHP) to determine the weight vector of each credit data point. Based on these weight vectors, all credit data are fused to obtain a target feature vector. This target feature vector is then input into a pre-constructed risk assessment model to obtain the model's output, which includes at least a risk assessment score. Finally, if the return on the borrower's target financial capital exceeds a preset threshold, the loan application is approved, and the loan amount, interest rate, and term are determined based on the risk assessment score. This invention, by fusing multi-source credit data, can form a comprehensive and multi-dimensional borrower credit profile. By using a risk assessment model to predict lending risk, it can accurately assess the borrower's creditworthiness, thus solving the technical problem in related technologies where credit risk assessment cannot comprehensively and accurately reflect the borrower's creditworthiness.

[0147] 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.

[0148] Those skilled in the art will understand that all or part of the steps in the various credit risk control methods based on multi-source data fusion and reinforcement learning in the above embodiments can be implemented by a program instructing the hardware of 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.

[0149] Example 4

[0150] 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 credit risk control method based on multi-source data fusion and reinforcement learning provided in Embodiment 1.

[0151] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the credit risk control method based on multi-source data fusion and reinforcement learning as described in any of the embodiments above.

[0152] 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.

[0153] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the credit risk control method based on multi-source data fusion and reinforcement learning described in various embodiments of this application.

[0154] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the credit risk control method based on multi-source data fusion and reinforcement learning described in various embodiments of this application.

[0155] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0156] In the above embodiments of the present invention, 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.

[0157] 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 can be a logical functional division, and in actual implementation, there may be other division methods. For instance, 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.

[0158] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] Furthermore, the functional units in the various embodiments of the present invention 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.

[0160] 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 the present invention, 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 the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0161] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A credit risk control method based on multi-source data fusion and reinforcement learning, characterized in that, include: In response to a borrower's loan application, the system obtains a set of credit data for the borrower from multiple data sources and preprocesses the set of credit data. The set of credit data includes at least: basic customer information, financial statements, credit records, and transaction history. The weight vector of each credit data in the preprocessed credit data set is determined by the hierarchical analysis strategy, and all the credit data are fused based on the weight vector to obtain the target feature vector; The target feature vector is input into a pre-built risk assessment model to obtain the model output, wherein the model output includes at least a risk assessment score, calculated using the following formula: ; in, Score As a risk assessment score, P To predict the probability of default risk, L To predict the resulting default risk loss, R Assess the borrower's historical credit score. α , β , γ These are the weighting coefficients for the probability of default risk, the loss due to default risk, and the historical credit score, respectively. If the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, the borrower's loan application is approved, and the loan amount, interest rate and term are determined based on the risk assessment score; The steps for determining the weight vectors of various credit data in the credit dataset using a hierarchical analysis strategy include: obtaining the importance ratios between the credit data; constructing a judgment matrix based on the importance ratios; and normalizing the judgment matrix; calculating the eigenvectors of the judgment matrix; normalizing the eigenvectors to obtain the weight vectors of the credit data, wherein the weight vectors corresponding to each credit data are different; calculating the consistency ratio of the weight vectors; and, if the consistency ratio is greater than a preset index threshold, adjusting the judgment matrix and recalculating the weight vectors, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix. The risk assessment model is trained as follows: The risk assessment model is initialized using a reinforcement learning algorithm, defining a state space, action space, reward function, and Q-value. The state space includes the target feature vector; the action space includes actions such as loan approval, loan rejection, and loan amount adjustment; the reward function includes expected return and expected risk; and the Q-value is the expected return of selecting the target action in the current state. The risk assessment model is then trained based on historical target feature vectors. The training steps include: inputting the historical target feature vectors into the risk assessment model; updating the risk assessment model based on the output Q-value; and terminating the training of the risk assessment model when the Q-value reaches the convergence condition.

2. The credit risk control method according to claim 1, characterized in that, The steps of obtaining the borrower's credit data set from multiple data sources include: The basic customer information is obtained from the internal information system of the financial institution or a third-party institution, wherein the basic customer information includes: the borrower's age, occupation and income; The financial statements are obtained from data files provided by the borrower or from a third-party institution, wherein the financial statements include: the borrower's assets and liabilities; The credit records are obtained from the internal information systems of financial institutions or third-party institutions, and the credit records include: historical loan information, number of overdue payments, amount owed, and violation records; The transaction records are obtained from the financial institution's internal information system. The transaction records include: the borrower's historical transaction data, transaction frequency, transaction amount, and fund flow information.

3. The credit risk control method according to claim 1, characterized in that, The credit data set includes various data types, including numerical data and categorical data. The preprocessing steps for the credit data set include: The credit data set is subjected to deduplication processing, which includes: calculating the similarity value between any two credit data sets; if the similarity value is greater than a preset similarity threshold, removing any one of the two data sets; and filling missing values ​​in the credit data with the mean, median, or mode. In the case where the credit data is the categorized data, the categorized data is encoded and converted into numerical data; When the credit data is numerical data, the numerical data is standardized and mapped to a target range. The various credit data in the credit data set are normalized.

4. The credit risk control method according to claim 1, characterized in that, The steps of inputting the target feature vector into the risk assessment model and calculating the risk assessment score include: The target feature vector is input into the risk assessment model to obtain the default risk probability and default risk loss; Obtain the borrower's historical credit score; The risk assessment score is calculated based on the default risk probability, the default risk loss, and the historical credit score.

5. The credit risk control method according to claim 1, characterized in that, Also includes: Obtain the target financial capital applied for by the borrower, and extract the capital loan interest rate and capital loan term from the model output. Calculate the expected return based on the target financial capital, the capital loan interest rate, and the capital loan term. Based on the target financial capital, the loan amount cost is obtained; the basic loan cost is obtained from the financial institution's internal information system; and based on the loan amount cost and the basic loan cost, the operating cost is calculated. Based on the target financial capital and the probability of default risk, a risk exposure value is calculated, wherein the risk exposure value is used to measure the risk of lending by financial institutions; The risk-free interest rate is obtained from the internal information system of the financial institution, wherein the risk-free interest rate is used to measure the return obtained by the financial institution under risk-free conditions; The rate of return on capital for the target financial capital applied for by the borrower is calculated based on the expected return, the operating costs, the risk exposure, and the risk-free interest rate.

6. The credit risk control method according to claim 1, characterized in that, The steps for determining the loan amount, interest rate, and term based on the risk assessment score include: Obtain the benchmark loan amount from the model output; Obtain benchmark interest rates and benchmark maturities from the internal information systems of financial institutions; Based on the benchmark loan amount, the benchmark interest rate, and the benchmark term, the loan amount, interest rate, and term are determined according to the risk assessment score, wherein the loan amount and term are positively proportional to the risk assessment score, and the interest rate is inversely proportional to the risk assessment score.

7. A credit risk control device based on multi-source data fusion and reinforcement learning, characterized in that, include: The acquisition unit is used to respond to a loan application initiated by a borrower, acquire a set of credit data of the borrower from multiple data sources, and preprocess the set of credit data, wherein the set of credit data includes at least: basic customer information, financial statements, credit records and transaction records; The fusion unit is used to determine the weight vector of each credit data in the preprocessed credit data set using a hierarchical analysis strategy, and to fuse all the credit data based on the weight vector to obtain the target feature vector. The input unit is used to input the target feature vector into a pre-built risk assessment model to obtain the model output result, wherein the model output result includes at least: a risk assessment score, calculated using the following formula: ; in, Score As a risk assessment score, P To predict the probability of default risk, L To predict the resulting default risk loss, R Assess the borrower's historical credit score. α , β , γ These are the weighting coefficients for the probability of default risk, the loss due to default risk, and the historical credit score, respectively. The decision-making unit is used to approve the borrower's loan application if the rate of return on the target financial capital applied for by the borrower is greater than a preset rate of return threshold, and to determine the loan amount, interest rate and term based on the risk assessment score; The fusion unit includes: a construction module for obtaining the importance ratio between credit data, constructing a judgment matrix based on the importance ratio, and normalizing the judgment matrix; a first calculation module for calculating the eigenvector of the judgment matrix, normalizing the eigenvector to obtain the weight vector of the credit data, wherein the weight vector corresponding to each credit data is different; and a second calculation module for calculating the consistency ratio of the weight vector, adjusting the judgment matrix and recalculating the weight vector if the consistency ratio is greater than a preset index threshold, wherein the consistency ratio is used to measure the logical consistency of the judgment matrix. The input unit includes: a definition module, used to initialize the risk assessment model using a reinforcement learning algorithm, defining the state space, action space, reward function, and Q-value. The state space includes the target feature vector; the action space includes actions such as approving a loan, rejecting a loan, and adjusting the loan amount; the reward function includes expected return and expected risk; and the Q-value is the expected return of choosing the target action in the current state. A training module is used to train the risk assessment model based on historical target feature vectors. The training steps include: inputting the historical target feature vectors into the risk assessment model; updating the risk assessment model based on the output Q-value; and ending the training of the risk assessment model when the Q-value reaches the convergence condition.

8. An electronic device, characterized in that, It includes 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 the credit risk control method based on multi-source data fusion and reinforcement learning as described in any one of claims 1 to 6.

Images