Risk credit approval method and device

Abstract

The invention provides a risk credit approval method and device. A processor obtains task data of a credit approval task; inputting the task data and the first-stage question set into the large model to obtain a first-stage answer result set; inputting the task data, the first-stage answer result set and the second-stage question set into a large model to obtain a second-stage answer result set; according to the first-stage answer result set and the second-stage answer result set, generating a plurality of prompt tags and a target credit review report; and outputting the plurality of prompt labels and the target credit review report. Therefore, in the application, the processor divides the whole examination and approval process into two stages, positive evidence or negative investigation is carried out on the question and answer result of the second stage through the question and answer result of the first stage, the accuracy of risk credit examination and approval is improved, the target credit examination report is directly generated through the large model, and the efficiency of risk credit examination and approval is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a risk credit approval method and device. Background Art

[0002] When customers apply for loans, risk approval usually inquires about the customer's credit report from the People's Bank of China, and sometimes asks customers to provide additional information such as bank transaction bills. This information contains a large amount of data and text information. Traditional machine learning methods have a weak understanding of text information and are unable to fully utilize the information contained therein, resulting in inaccurate conclusions on risk credit approval. How to improve the accuracy of risk credit approval is an urgent problem that needs to be solved. Summary of the invention

[0003] The present application provides a risk credit approval method and device, which divides the entire approval process into two stages, and uses the answer results of the first stage to positively corroborate or negatively check the answer results of the second stage, thereby improving the accuracy of risk credit approval, and directly generating a target credit review report through a large model, thereby improving the efficiency of risk credit approval.

[0004] In a first aspect, the present application provides a risk credit approval method, the method comprising: obtaining task data of a credit approval task; inputting the task data and a first-stage question set into a large model to obtain a first-stage answer result collection, wherein the first-stage question set includes at least one question regarding at least one aspect of work mode, behavioral preference, and comprehensive quality; inputting the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain a second-stage answer result collection; the second-stage question set includes at least one question regarding at least one aspect of income structure, assets and liabilities, and credit characteristics; generating multiple prompt labels and a target credit review report based on the first-stage answer result collection and the second-stage answer result collection, the target credit review report being associated with the risk assessment level of the credit approval task, and the multiple prompt labels being used to indicate the risk types that may exist in the credit approval task; and outputting the multiple prompt labels and the target credit review report.

[0005] In a second aspect, the present application provides a risk credit approval device, comprising: a receiving unit, used to obtain task data of a credit approval task; a processing unit, used to input the task data and a first-stage question set into a large model to obtain a first-stage answer result collection, wherein the first-stage question set includes at least one question for at least one aspect of work mode, behavioral preference, and comprehensive quality; and, input the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain a second-stage answer result collection; the second-stage question set includes at least one question for at least one aspect of income structure, assets and liabilities, and credit characteristics; and, based on the first-stage answer result collection and the second-stage answer result collection, generate multiple prompt labels and a target credit review report, the target credit review report is associated with the risk assessment level of the credit approval task, and the multiple prompt labels are used to indicate the risk types that may exist in the credit approval task; an output unit, used to output the multiple prompt labels and the target credit review report.

[0006] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the method described in any one of the first aspects is executed.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program / instruction stored thereon, wherein the computer program / instruction, when executed by a processor, implements the steps in the method as described in any one of the first aspects.

[0008] It can be seen that in the embodiment of the present application, the processor obtains the task data of the credit approval task; inputs the task data and the first-stage question set into the large model to obtain the first-stage answer result collection, and the first-stage question set includes at least one question for at least one aspect of work mode, behavior preference, and comprehensive quality; inputs the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain the second-stage answer result collection; the second-stage question set includes at least one question for at least one aspect of income composition, assets and liabilities, and credit characteristics; generates multiple prompt labels and target credit review reports based on the first-stage answer result collection and the second-stage answer result collection, and the target credit review report is associated with the risk assessment level of the credit approval task, and multiple prompt labels are used to indicate the risk type that may exist in the credit approval task; outputs multiple prompt labels and target credit review reports. Therefore, in the application, the processor divides the entire approval process into two stages, and positively corroborates or negatively checks the second-stage question and answer results through the first-stage question and answer results to improve the accuracy of risk credit approval, and directly generates the target credit review report through the large model to improve the efficiency of risk credit approval. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0010] Figure 1 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0011] Figure 2 A flowchart of a risk credit approval method provided in an embodiment of the present application;

[0012] Figure 3 A flowchart of another risk credit approval method provided in an embodiment of the present application;

[0013] Figure 4 A schematic diagram of the process flow of the first stage and the second stage provided in the embodiment of the present application;

[0014] Figure 5 A schematic diagram of preprocessing task data provided in an embodiment of the present application;

[0015] Figure 6 A schematic diagram of using a large model to conduct the first stage of question and answering provided in an embodiment of the present application;

[0016] Figure 7 A schematic diagram of using a large model to conduct second-stage question-answering provided in an embodiment of the present application;

[0017] Figure 8 This is a functional unit block diagram of the risk credit approval device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. 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 limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0020] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] In the embodiments of the present application, "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent the following three situations: A exists alone; A and B exist at the same time; B exists alone. Among them, A and B can be singular or plural.

[0022] In the embodiment of the present application, the symbol " / " can indicate that the objects associated with each other are in an "or" relationship. In addition, the symbol " / " can also indicate a division sign, that is, performing a division operation. For example, A / B can indicate A divided by B.

[0023] In the embodiments of the present application, "at least one item" or similar expressions refer to any combination of these items, including any combination of single items or plural items, and refer to one or more, and multiple refers to two or more. For example, at least one item of a, b, or c can represent the following seven situations: a, b, c, a and b, a and c, b and c, a, b, and c. Among them, each of a, b, and c can be an element or a set containing one or more elements.

[0024] In the embodiments of the present application, "equal to" can be used in conjunction with greater than, and is applicable to the technical solution adopted when greater than, and can also be used in conjunction with less than, and is applicable to the technical solution adopted when less than. When equal to is used in conjunction with greater than, it is not used in conjunction with less than; when equal to is used in conjunction with less than, it is not used in conjunction with greater than.

[0025] In order to solve the above problems, this application provides a risk credit approval method and device, which divides the entire approval process into two stages, and uses the question and answer results of the first stage to positively corroborate or negatively check the question and answer results of the second stage, thereby improving the accuracy of risk credit approval, and directly generating a target credit review report through a large model to improve the efficiency of risk credit approval.

[0026] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0027] See also Figure 1 , Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. For the sake of convenience, only the parts related to the embodiment of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present application. The electronic device 1 can be any terminal device including a mobile phone, a computer, a PDA, a POS, a car computer, etc. Figure 1 As shown, the electronic device 1 includes a memory 20, a processor 10, a communication bus 40, a communication interface 30 and one or more programs 21. The one or more programs 21 are stored in the memory 20 and are configured to be executed by the processor 10. The one or more programs 21 include instructions for executing any step in the following method embodiment. In a specific implementation, the processor 10 is used to execute the following Figure 2 and Figure 3 Any step instruction in the method embodiment shown, and when performing data transmission such as sending, the communication interface 30 can be selectively called to complete the corresponding operation.

[0028] In the embodiment of the present application, the processor 10 included in the electronic device 1 may have the functions corresponding to any method step in the embodiment.

[0029] Those skilled in the art will understand that Figure 1 The structure of the electronic device 1 shown in the figure does not constitute a limitation on the electronic device 1, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0030] See also Figure 2 , Figure 2 A flow chart of a risk credit approval method provided in an embodiment of the present application, wherein the method is applied to Figure 1 The processor 10 shown, the method includes:

[0031] Step S201: The processor obtains task data of a credit approval task.

[0032] In step S202, the processor inputs the task data and the first-stage question set into the large model to obtain a first-stage answer result set.

[0033] Among them, the first-stage question set includes at least one question regarding at least one aspect of work mode, behavioral preference, and comprehensive quality.

[0034] Step S203, the processor inputs the task data, the first stage answer result collection and the second stage question set into the large model to obtain the second stage answer result collection.

[0035] The second-stage question set includes at least one question regarding at least one aspect of income structure, assets and liabilities, and credit characteristics.

[0036] Step S204: The processor generates a plurality of prompt labels and a target credit review report according to the first-stage answer result collection and the second-stage answer result collection.

[0037] The target credit review report is associated with the risk assessment level of the credit approval task, and the multiple prompt tags are used to indicate the risk types that may exist in the credit approval task.

[0038] Step S205: the processor outputs the multiple prompt labels and the target credit review report.

[0039] according to Figure 2 The overall process of the first and second stages is as follows: Figure 4 As shown, in the first stage, the processor sends questions 1 to question n to the big model, and the big model retrieves according to the task data to obtain answers 1 to answer n respectively; the big model summarizes answers 1 to answer n to obtain the first stage answer result collection; in the second stage, the processor inputs questions 9 to question m into the big model, and the big model obtains answers 9 to answer m according to the first stage answer result collection and the task data; the big model summarizes the first stage answer result collection and answers 9 to answer m to obtain the target credit review report.

[0040] See also Figure 3 , Figure 3 A flow chart of another risk credit approval method provided in an embodiment of the present application is as follows: Figure 3 As shown, the method includes:

[0041] Step S301: The processor obtains task data of a credit approval task.

[0042] The processor creates a credit approval task for each loan application and obtains the task data of the loan applicant, which usually includes but is not limited to basic customer information, consumption behavior, finance, PBOC credit, bank transaction bills, etc. After the processor obtains the task data of a single credit approval task, it pre-processes the task data, and then sends the processed data and the preset question set to the big model, which generates the target credit review report from the task data.

[0043] In one or more embodiments of the present application, a large model refers to a deep learning model with large-scale model parameters. The large model is pre-trained using large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization capabilities, such as a large-scale language model (LLM, Large Language Model).

[0044] Step S302: the processor classifies the task data to obtain a plurality of classified data.

[0045] Among them, the task data specifically includes but is not limited to one or more of the following: all details of the People's Bank of China, all details of the bill, variables of the People's Bank of China, variables of the bill, and statistics of the bill; specifically, all details of the People's Bank of China can be divided into basic information of the People's Bank of China and all information of the People's Bank of China, and all details of the bill can be filtered and divided into bill income details and bill expenditure details, and the variables of the People's Bank of China can be summarized to obtain the summary of variables of the People's Bank of China, and the variables of the bill can be summarized to obtain the summary of bill variables, and the bill statistics remain unchanged. Therefore, multiple classified data include but are not limited to one or more of the following: basic information of the People's Bank of China, all information of the People's Bank of China, details of bill income, details of bill expenditure, summary of variables of the People's Bank of China, summary of variables of the bill, summary of variables of the bill, and statistics of the bill.

[0046] Among them, the PBOC variable and the bill variable are characteristic variables, which are used to calculate the accurate value corresponding to a single PBOC variable based on the information in all the PBOC details and / or the information in all the bill details, or to calculate the accurate value corresponding to a single bill variable based on the information in all the PBOC details and / or the information in all the bill details. For example, the PBOC variables include the total loan amount, the mortgage amount, and the car loan amount. The large model can calculate the specific total loan amount, the mortgage amount, and the car loan amount of the loan applicant based on the PBOC variables and the information in all the PBOC details and / or the information in all the bill details. Its role is to accurately and efficiently calculate the specific content corresponding to the variable. The large model does not need to perform a complex calculation process in the step of obtaining the variable value, which reduces interference with the large model.

[0047] For example, see Figure 5 The process of filtering and unpacking all the details of the bill includes:

[0048] Preprocessing 1: Filter out all amounts less than 100 yuan;

[0049] Preprocessing 2: Delete the account number;

[0050] Pre-processing 3: Split all bill details into income and expenses.

[0051] It can be understood by those skilled in the art that Figure 5 The preprocessing process shown is only an example provided in this embodiment. The data preprocessing process may of course include other suitable methods, such as filling missing values, deleting outliers, and data vectorization, etc. This application does not impose specific limitations on this.

[0052] Step S303: the processor obtains a first-stage question set and a second-stage question set.

[0053] Among them, the first-stage question set includes at least one question regarding at least one aspect of work mode, behavioral preference, and comprehensive quality.

[0054] Among them, the work model can reflect the stability of the client, the reliability of the income source, and the risk tolerance. For example, employees of government agencies or state-owned enterprises are generally more stable, while some industries (such as the Internet industry, freelancers, etc.) may face greater income fluctuations. Exemplary work models include but are not limited to unit types, etc.

[0055] Among them, behavioral preferences can reflect customers' consumption attitudes, repayment willingness, and potential risks. For example, some customers may have a tendency to overspend or have a bad borrowing history, which may indicate a higher risk of default. For example, behavioral preferences include but are not limited to consumption behavior.

[0056] Among them, comprehensive quality includes the customer's educational background, work experience, etc., as well as his awareness and behavior of abiding by the contract and repaying on time. Customers with strong comprehensive quality usually have more confidence and sense of responsibility to fulfill the loan contract.

[0057] Therefore, the first-stage question set focuses on the individual background, behavioral patterns, and psychological characteristics of the loan applicant, which affect the risk tolerance of the loan applicant and are subjective assessments.

[0058] The second-stage question set includes at least one question about at least one aspect of income structure, assets and liabilities, and credit characteristics. The second-stage question set focuses on the actual economic situation and financial capacity of the loan applicant, which is an objective assessment.

[0059] The first-stage question set mainly serves as a positive corroboration or reverse check for the second-stage question set. A customer may have a good income and asset status (second stage), but if he lacks a good sense of performance and behavioral norms (first stage), he may still be at risk of default; conversely, even if the customer has some deficiencies in income and assets, if he has a strong sense of performance and repayment ability, he may also be a low-risk customer. Therefore, the collection of the first-stage answer results is used to positively or negatively affect the large model on the second-stage summary results, so that the target credit review report comprehensively considers the customer's risk tolerance and actual economic situation.

[0060] In step S304, the processor establishes a first mapping relationship between each classified data in the multiple classified data and each first-stage question in the first-stage question set, and establishes a second mapping relationship between each classified data in the multiple classified data and each second-stage question in the second-stage question set.

[0061] The first mapping relationship is used to characterize the association between each first-stage question and the classification data required to answer the question; the second mapping relationship is used to characterize the association between each second-stage question and the classification data required to answer the question.

[0062] For details, see Figure 6 In some embodiments, the first mapping relationship includes: the first-stage questions about the working mode include but are not limited to the basic information of the People's Bank of China, the bill income details, and the bill expenditure details; the first-stage questions about the behavioral preferences include but are not limited to the bill expenditure details, the bill variable summary, and the bill statistics; the first-stage questions about the comprehensive quality include but are not limited to the basic information of the People's Bank of China.

[0063] The second mapping relationship includes: Figure 7 The second-stage issues regarding the income structure include but are not limited to the bill income details, the bill variable summary, and the bill statistics; the second-stage issues regarding assets and liabilities include but are not limited to all the information of the People's Bank of China, the bill income details, the bill expenditure details, and the People's Bank of China variable summary; the credit characteristics include but are not limited to all the information of the People's Bank of China and the People's Bank of China variable summary.

[0064] It can be seen that in this embodiment, by establishing a mapping relationship between questions and data, it is ensured that the large model only uses part of the data when answering specific questions, thereby avoiding calculation errors of the large model.

[0065] Step S305: The processor inputs the task data and the first-stage question set into the large model to obtain a first-stage answer result collection.

[0066] For specific implementation, see Figure 6 For any questions about comprehensive quality, work mode, behavioral preference, etc., the big model obtains the answer to each question based on the task data and the first mapping relationship, and summarizes the answer to each question to obtain the first-stage answer result collection.

[0067] In some embodiments, the specific process of the large model creating a single first-stage answer result specifically includes the following steps a to c:

[0068] Step a: The large model obtains at least one classification data corresponding to the single first-stage problem from the task data according to the first mapping relationship.

[0069] Step b: the large model performs vector similarity calculation based on the first-stage question vector corresponding to the single first-stage question and the set of classified data vectors corresponding to the at least one classified data to obtain a first target classified data vector having the highest similarity to the first-stage question vector.

[0070] Among them, the general steps of big model question and answer include data preparation, retrieval, and answer splicing. Specifically: first, the user uploads the data, and the big model divides the data into blocks and stores it using embedded technology; once a question is asked, the big model uses vector search technology to mine the stored data to find relevant information; the big model combines the retrieved relevant information with the question to generate the final prompt answer.

[0071] In the present application, the processor sends the task data and the first-stage question collection to the large model; after receiving the questions and data, the large model performs vector similarity calculation on the task data related to the question according to the first mapping relationship, and obtains the first target classification data vector with the highest similarity to the question. The classification data corresponding to the first target classification data vector is the answer to the question.

[0072] In some embodiments, the similarity between the first-stage question vector and the classification data vector set corresponding to the at least one classification data is measured by the vector distance between the vectors.

[0073] Step c: the large model determines the first target classification data corresponding to the first target classification data vector as the first stage answer result of the first stage question.

[0074] Step S306: The processor inputs the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain the second-stage answer result collection.

[0075] For specific implementation, see Figure 7For any questions about income structure, assets and liabilities, credit characteristics, etc., the big model obtains the answer to each question based on the task data and the second mapping relationship, and summarizes the answer to each question to obtain the second-stage answer result collection.

[0076] In some embodiments, the specific process of the large model creating the second-stage answer result of a single second-stage question includes the following steps d to g:

[0077] Step d: the large model obtains at least one classification data corresponding to the single second-stage problem from the task data according to the second mapping relationship.

[0078] Step e: the large model performs vector similarity calculation based on the second-stage question vector of the single second-stage question and the classified data set of its corresponding classified data to obtain a second target classified data vector having the highest similarity to the single second-stage question vector.

[0079] Among them, the specific method for the large model to calculate the second target classification data vector refers to the above step b and will not be repeated here.

[0080] In step f, the large model determines a first scoring index based on the answer results of the first stage and the preset scoring criteria of the first stage. The first scoring index is used to characterize the risk tolerance of the credit approval task. The risk tolerance is inversely proportional to the risk assessment level.

[0081] For each question, the lending institution sets a preset scoring standard. For example, for the unit type, the public servants are scored higher than the casual workers, and the employed are scored higher than the unemployed. Therefore, the first-stage question set is set with the first-stage preset scoring standard, and the second-stage question set is set with the second-stage preset scoring standard.

[0082] In some embodiments, the large model can determine the scoring index of the answer result by calculating the similarity between the answer result and the preset scoring criteria, scoring index = similarity × preset scoring criteria, where the similarity can be represented by the vector distance between the vector data of the answer result and the vector data of the preset scoring criteria.

[0083] Among them, the first scoring index is used to characterize the risk tolerance of the loan applicant. The greater the risk tolerance, the lower the risk. When conducting the second stage of question and answer, the first scoring index can reduce the risk level of the second stage answer result collection, so that the risk assessment level is lower, which means that the loan is safer; the lower the risk tolerance, the higher the risk. When conducting the second stage of question and answer, the first scoring index can increase the risk level of the second stage answer result collection, so that the risk assessment level is higher, which means that the loan has a higher risk.

[0084] Step g: the large model obtains the second stage answer result according to the first scoring index and the target classification data vector.

[0085] Among them, the large model verifies the target classification data found according to the first scoring index, and positively or negatively adds the target classification data vector through the first scoring index to obtain the final second-stage answer result. Exemplarily, through the first-stage question set, it is found that user A is a stable income group, then the income level calculated by the large model in the second stage can be regarded as the actual income level of user A; if user A is determined to be a group with declining income in the first stage, it is necessary to negatively add the income level calculated by the large model in the second stage to reduce the actual income of user A, so that the income level calculated by the large model is consistent with the income change trend of user A; if user A is determined to be a group with steadily rising income in the first stage, the income level calculated by the large model in the second stage can be positively added to moderately increase the actual income of user A.

[0086] Step S307: The processor generates a plurality of prompt labels and a target credit review report according to the first-stage answer result collection and the second-stage answer result collection.

[0087] In some embodiments, the processor generates a plurality of prompt labels and a target credit review report according to the first-stage answer result collection and the second-stage answer result collection, including the following steps h-j:

[0088] Step h, determining a first scoring index for at least one of the working mode, the behavioral preference, and the comprehensive quality based on the collection of answer results of the first stage and the preset scoring criteria of the first stage, and determining a prompt label for at least one of the working mode, the behavioral preference, and the comprehensive quality based on the first scoring index.

[0089] Step i: Determine a second scoring index for at least one of the income structure, the assets and liabilities, and the credit characteristics based on the collection of the answer results of the second stage and the preset scoring criteria of the second stage, and determine a prompt label for at least one of the income structure, the assets and liabilities, and the credit characteristics based on the second scoring index.

[0090] Among them, there is a mapping relationship between the scoring index and the prompt label. For example, for the work mode, if the sum of the first scoring index of at least one first-stage question included in the work mode exceeds the first preset threshold, the corresponding work is stable. If it is lower than the first preset threshold, the corresponding work is unstable. Therefore, according to the corresponding relationship between the scoring index and the prompt label, prompt labels can be generated for each aspect of work mode, behavioral preference, comprehensive quality, income structure, assets and liabilities, and credit characteristics to indicate possible risks, etc., to assist the reviewer in reviewing.

[0091] Step j, summarizing the collection of the first-stage answer results and the second-stage answer results to obtain the target credit review report.

[0092] Among them, the target credit review report includes the collection of the first-stage answer results and the second-stage answer results, which are used to summarize and refine the customer's basic information, liabilities, repayment ability, etc., so as to improve the approval efficiency of the approval personnel.

[0093] Step S308: The processor outputs the multiple prompt labels and the target credit review report.

[0094] In some embodiments, the method further includes: outputting a reference risk assessment level of the credit approval task according to the first scoring index and the second scoring index.

[0095] Among them, the processor can also calculate the reference risk assessment level of the current approval task based on the first scoring index and the second scoring index, which is used to assist the reviewer in determining whether the assessment conclusion made based on the prompt label and the target credit review report is accurate.

[0096] It can be seen that in the embodiment of the present application, the processor obtains the task data of the credit approval task; inputs the task data and the first-stage question set into the large model to obtain the first-stage answer result collection, and the first-stage question set includes at least one question for at least one aspect of work mode, behavioral preference, and comprehensive quality; inputs the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain the target credit review report, and the second-stage question set includes at least one question for at least one aspect of income structure, assets and liabilities, and credit characteristics. The target credit review report is used to characterize the risk assessment level of the credit approval task; wherein, the first-stage answer result collection is used to characterize the degree of influence on the risk assessment level. Therefore, in the application, the processor divides the entire approval process into two stages, and positively corroborates or negatively checks the second-stage question and answer results through the first-stage question and answer results, thereby improving the accuracy of risk credit approval, and directly generates the target credit review report through the large model, thereby improving the efficiency of risk credit approval.

[0097] In some embodiments, the processor also provides a professional knowledge base and inputs the professional knowledge base into the big model, so that the big model can add business knowledge during the question-answering process, making the big model's decisions more professional.

[0098] Among them, exemplary, common transaction types include consumption, repayment, investment, lending to others, temporary expenditure, real estate, etc., and each transaction is classified into the first and second levels:

[0099] Consumption, including basic consumption such as rent, food and beverage, living expenses, etc., and excessive consumption such as luxury goods;

[0100] Repayment, including but not limited to bank loans, credit card advances, etc.;

[0101] Investment, including but not limited to financial investment such as funds and stocks, industrial investment such as company shares, etc.;

[0102] lending money to others;

[0103] Temporary expenses, including but not limited to fines, unexpected events such as illness and other large expenses;

[0104] Property purchase, such as real estate purchase;

[0105] In addition, it can also include red envelopes, transfers, personal bank turnover, etc.

[0106] The above mainly introduces the scheme of the embodiment of the present application from the perspective of the execution process on the method side. It is understandable that the processor includes a hardware structure and / or software module corresponding to each function in order to realize the above functions. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiment provided herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present application.

[0107] The embodiment of the present application can divide the processor into functional units according to the above method example. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated unit can be implemented in the form of hardware or in the form of a software program module. It should be noted that the division of units in the embodiment of the present application is schematic, which is only a logical function division, and there may be other division methods in actual implementation.

[0108] In the case of integrated units, see Figure 8 , Figure 8 A risk credit approval device is provided in an embodiment of the present application. The risk credit approval device 8 includes:

[0109] Receiving unit 801, used to obtain task data of the credit approval task;

[0110] Processing unit 802 is used to input the task data and the first-stage question set into the big model to obtain a first-stage answer result collection, wherein the first-stage question set includes at least one question for at least one aspect of work mode, behavior preference, and comprehensive quality; and input the task data, the first-stage answer result collection, and the second-stage question set into the big model to obtain a second-stage answer result collection; the second-stage question set includes at least one question for at least one aspect of income structure, assets and liabilities, and credit characteristics; and generate multiple prompt labels and a target credit review report according to the first-stage answer result collection and the second-stage answer result collection, wherein the target credit review report is associated with the risk assessment level of the credit approval task, and the multiple prompt labels are used to indicate the risk type that may exist in the credit approval task;

[0111] The output unit 803 is used to output the multiple prompt labels and the target credit review report.

[0112] It can be seen that in the embodiment of the present application, the processor obtains the task data of the credit approval task; inputs the task data and the first-stage question set into the large model to obtain the first-stage answer result collection, and the first-stage question set includes at least one question for at least one aspect of work mode, behavior preference, and comprehensive quality; inputs the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain the second-stage answer result collection; the second-stage question set includes at least one question for at least one aspect of income composition, assets and liabilities, and credit characteristics; generates multiple prompt labels and target credit review reports based on the first-stage answer result collection and the second-stage answer result collection, and the target credit review report is associated with the risk assessment level of the credit approval task, and multiple prompt labels are used to indicate the risk type that may exist in the credit approval task; outputs multiple prompt labels and target credit review reports. Therefore, in the application, the processor divides the entire approval process into two stages, and positively corroborates or negatively checks the second-stage question and answer results through the first-stage question and answer results to improve the accuracy of risk credit approval, and directly generates the target credit review report through the large model to improve the efficiency of risk credit approval.

[0113] In some embodiments, after the processing unit 802 obtains the task data of the credit approval task, the method further includes:

[0114] The task data is classified and processed to obtain multiple classified data; the task data includes at least one of all details of the People's Bank of China, all details of the bill, People's Bank of China variables, bill variables, and bill statistics; the multiple classified data include at least one of basic information of the People's Bank of China, all information of the People's Bank of China, bill income details, bill expenditure details, People's Bank of China variable summary, bill variable summary, and bill statistics; a first mapping relationship is established between each classified data in the multiple classified data and each first-stage question in the first-stage question set, and a second mapping relationship is established between each classified data in the multiple classified data and each second-stage question in the second-stage question set; the first mapping relationship is used to characterize the association relationship between each first-stage question and the classified data required to answer the question.

[0115] In some embodiments, the first mapping relationship includes: the first-stage questions about the working mode include the basic information of the People's Bank of China, the bill income details, and the bill expenditure details; the first-stage questions about the behavioral preferences include the bill expenditure details, the bill variable summary, and the bill statistics; the first-stage questions about the comprehensive quality include the basic information of the People's Bank of China; the second mapping relationship includes: the second-stage questions about the income structure include the bill income details, the bill variable summary, and the bill statistics; the second-stage questions about assets and liabilities include all the information of the People's Bank of China, the bill income details, the bill expenditure details, and the People's Bank variable summary; the credit characteristics include all the information of the People's Bank of China and the People's Bank variable summary.

[0116] In some embodiments, the process of creating the first-stage answer result of a single first-stage question specifically includes: obtaining at least one classification data corresponding to the single first-stage question from the task data according to the first mapping relationship; performing vector similarity calculation based on the first-stage question vector corresponding to the single first-stage question and the classification data vector set corresponding to the at least one classification data, to obtain the first target classification data vector with the highest similarity to the first-stage question vector; and determining the first target classification data corresponding to the first target classification data vector as the first-stage answer result of the first-stage question.

[0117] In some embodiments, the specific process of creating the second-stage answer result of a single second-stage question includes: obtaining at least one classification data corresponding to the single second-stage question from the task data according to the second mapping relationship; performing vector similarity calculation based on the second-stage question vector of the single second-stage question and the classification data set of its corresponding classification data to obtain a second target classification data vector with the highest similarity to the single second-stage question vector; determining a first scoring index based on the first-stage answer result and the first-stage preset scoring criteria, the first scoring index being used to characterize the risk tolerance of the credit approval task, the risk tolerance being inversely proportional to the risk assessment level; and obtaining the second-stage answer result based on the first scoring index and the target classification data vector.

[0118] In some embodiments, the generating of multiple prompt labels and a target credit review report based on the first-stage answer result collection and the second-stage answer result collection includes: determining a first scoring index of at least one of the working mode, the behavioral preference, and the comprehensive quality according to the first-stage answer result collection and the first-stage preset scoring standard, and determining a prompt label of at least one of the working mode, the behavioral preference, and the comprehensive quality according to the first scoring index; determining a second scoring index of at least one of the income structure, the assets and liabilities, and the credit characteristics according to the second-stage answer result collection and the second-stage preset scoring standard, and determining a prompt label of at least one of the income structure, the assets and liabilities, and the credit characteristics according to the second scoring index; and summarizing the first-stage answer result collection and the second-stage answer result collection to obtain the target credit review report.

[0119] In some embodiments, the method further includes: outputting a reference risk assessment level of the credit approval task according to the first scoring index and the second scoring index.

[0120] An embodiment of the present application provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the method described in any possible embodiment are implemented.

[0121] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0122] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0123] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above-mentioned units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0124] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0125] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0126] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, 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. The computer software product is stored in a memory, including a number of instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the above-mentioned methods of each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0127] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, and the memory can include: a flash drive, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a magnetic disk or an optical disk, etc.

[0128] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A risk credit approval method, characterized in that: The method comprises: Get the task data of the credit approval task; Inputting the task data and the first-stage question set into the large model to obtain a first-stage answer result collection, wherein the first-stage question set includes at least one question for at least one aspect of work mode, behavior preference, and comprehensive quality; Inputting the task data, the first-stage answer result set and the second-stage question set into the large model to obtain the second-stage answer result set; the second-stage question set includes at least one question for at least one aspect of income composition, assets and liabilities, and credit characteristics; generating a plurality of prompt labels and a target credit review report according to the collection of the answer results of the first stage and the collection of the answer results of the second stage, wherein the target credit review report is associated with the risk assessment level of the credit approval task, and the plurality of prompt labels are used to indicate the risk types that may exist in the credit approval task; Output the multiple prompt labels and the target credit review report.

2. The method according to claim 1, characterized in that: After obtaining the task data of the credit approval task, the method further includes: The task data is classified and processed to obtain a plurality of classified data; the task data includes at least one of all details of the People's Bank of China, all details of the bill, variables of the People's Bank of China, variables of the bill, and bill statistics; the plurality of classified data includes at least one of basic information of the People's Bank of China, all information of the People's Bank of China, details of bill income, details of bill expenditure, summary of variables of the People's Bank of China, summary of variables of the bill, and bill statistics; Establishing a first mapping relationship between each classified data in the plurality of classified data and each first-stage question in the first-stage question set, and establishing a second mapping relationship between each classified data in the plurality of classified data and each second-stage question in the second-stage question set; The first mapping relationship is used to characterize the association relationship between each first-stage question and the classification data required to answer the question.

3. The method according to claim 2, characterized in that The first mapping relationship includes: the first-stage questions about the working mode include the basic information of the People's Bank of China, the bill income details, and the bill expenditure details; the first-stage questions about the behavior preference include the bill expenditure details, the bill variable summary, and the bill statistics; the first-stage questions about the comprehensive quality include the basic information of the People's Bank of China; The second mapping relationship includes: the second-stage problems in the income structure include the bill income details, the bill variable summary, and the bill statistics; the second-stage problems in the assets and liabilities include all the information of the People's Bank of China, the bill income details, the bill expenditure details, and the People's Bank of China variable summary; the credit characteristics include all the information of the People's Bank of China and the People's Bank of China variable summary.

4. The method according to claim 2 or 3, characterized in that: The process of creating the first-stage answer result for a single first-stage question specifically includes: Acquire at least one classification data corresponding to the single first-stage problem from the task data according to the first mapping relationship; Performing vector similarity calculation based on the first-stage question vector corresponding to the single first-stage question and the set of classified data vectors corresponding to the at least one classified data, to obtain a first target classified data vector having the highest similarity to the first-stage question vector; The first target classification data corresponding to the first target classification data vector is determined as the first stage answer result of the first stage question.

5. The method according to claim 4, characterized in that The specific process of creating the second-stage answer result for a single second-stage question includes: Acquire at least one classification data corresponding to the single second-stage question from the task data according to the second mapping relationship; Performing vector similarity calculation based on the second-stage question vector of the single second-stage question and the classified data set of the corresponding classified data to obtain a second target classified data vector having the highest similarity to the single second-stage question vector; Determine a first scoring index according to the answer result of the first stage and the preset scoring standard of the first stage, wherein the first scoring index is used to characterize the risk tolerance of the credit approval task, and the risk tolerance is in inverse proportion to the risk assessment level; The second-stage answer result is obtained according to the first scoring index and the target classification data vector.

6. The method according to claim 5, characterized in that The generating of a plurality of prompt labels and a target credit review report according to the collection of the answer results of the first stage and the collection of the answer results of the second stage includes: Determine a first scoring index of at least one of the working mode, the behavioral preference, and the comprehensive quality according to the collection of answer results of the first stage and the preset scoring standard of the first stage, and determine a prompt label of at least one of the working mode, the behavioral preference, and the comprehensive quality according to the first scoring index; Determine a second scoring index for at least one of the income composition, the assets and liabilities, and the credit characteristics based on the collection of the answer results of the second stage and the preset scoring criteria of the second stage, and determine a prompt label for at least one of the income composition, the assets and liabilities, and the credit characteristics based on the second scoring index; The target credit review report is obtained by summarizing the collection of the answer results of the first stage and the collection of the answer results of the second stage.

7. The method according to claim 6, characterized in that The method further comprises: A reference risk assessment level of the credit approval task is output according to the first scoring index and the second scoring index.

8. A risk credit approval device, characterized in that: include: A receiving unit, used to obtain task data of a credit approval task; a processing unit, configured to input the task data and the first-stage question set into the large model to obtain a first-stage answer result collection, wherein the first-stage question set includes at least one question for at least one aspect of work mode, behavior preference, and comprehensive quality; and input the task data, the first-stage answer result collection, and the second-stage question set into the large model to obtain a second-stage answer result collection; The second-stage question set includes at least one question for at least one aspect of income composition, assets and liabilities, and credit characteristics; and, based on the first-stage answer result set and the second-stage answer result set, a plurality of prompt labels and a target credit review report are generated, wherein the target credit review report is associated with the risk assessment level of the credit approval task, and the plurality of prompt labels are used to indicate the risk type that may exist in the credit approval task; An output unit is used to output the multiple prompt labels and the target credit review report.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 7 when calling the computer program in the memory.

10. A computer-readable storage medium having a computer program / instruction stored thereon, wherein the computer program / instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images