Credit business automatic processing method and system based on large language model

Abstract

The invention provides an automatic credit business processing method and system based on a large language model, and belongs to the technical field of financial information. When determining that the change condition of the user credit granting risk in the target period meets the requirement, determining position distribution data of the credit granting application user on the basis of an analysis result of the credit granting application information of the credit granting application user, determining suspected fraud users in the credit granting application users by combining analysis results of the credit granting application information of different credit granting application users, and vectorizing all credit granting application data of the suspected fraud users in the credit granting application process; and in combination with the vectorization processing result of the credit granting application information of the suspected fraudulent user, the credit report is automatically generated by utilizing the large language model, so that the efficiency of generating and processing the credit report is improved.

Description

Technical Field

[0001] The present invention belongs to the field of financial information technology, and in particular relates to a method and system for automatically processing credit business based on a large language model. Background Art

[0002] When processing credit business using existing technical solutions, it is often necessary for credit employees to process credit applications based on the personal information of credit applicants. Not only is the processing efficiency slow, but the accuracy of the processing results is also difficult to meet requirements.

[0003] To address the above technical issues, more and more consumer finance companies have set up their own credit application models to automatically process credit applications. Specifically, invention patent CN202410880254.0, "A Method and System for Generating and Processing Optimized Risk Control Strategies," and invention patent application CN202311738093.3, "A Method and System for Managing Credit Risk Control Based on a Multi-Task Boosting Tree," both implement automated approval and processing of credit applications. However, these technical solutions have the following technical issues: When processing credit business, credit reports are often generated on a regular basis. Existing credit reports are often generated manually, which inevitably makes the generation and processing of credit reports very difficult. Therefore, how to reduce the difficulty of generating and processing credit reports has become a technical problem that needs to be solved urgently.

[0004] In response to the above technical problems, the present invention proposes a method and system for automated processing of credit business based on a large language model. Summary of the Invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: According to one aspect of the present invention, a method for automated processing of credit business based on a large language model is provided.

[0006] A method for automated credit business processing based on a large language model, specifically comprising: S1 obtains overdue user data of a target credit business within a target period, and when the target credit business is determined to have an approval risk using the overdue user data, proceeds to the next step; S2 determines the identification result of the credit risk of the credit application information of the credit application users in the target period in the credit approval model, and through the changes in the distribution data of the credit application users in different credit risk intervals in the target period, determines that the changes in the user credit risk in the target period meet the requirements, and then proceeds to the next step; S3 determines the location distribution data of the credit applicant based on the analysis results of the credit application information of the credit applicant, and determines suspected fraudulent users among the credit applicants by combining the analysis results of the credit application information of different credit applicants; S4 vectorizes all the credit application data of the suspected fraudulent user during the credit application process, and automatically generates a credit report using a large language model based on the vectorized processing results of the credit application information excluding the suspected fraudulent user.

[0007] The beneficial effects of the present invention are: 1. In the present invention, overdue user data is used to determine whether a target credit business has an approval risk, thereby achieving an accurate assessment of the approval risk of the target credit business from multiple angles such as the number of overdue users and the overdue amount. This dynamically achieves the assessment of the approval risk and lays the foundation for further generating differentiated credit reports based on the size and changes of the approval risk. This not only reduces the difficulty of generating and processing the credit report, but also allows for faster attribution and dynamic adjustment of credit businesses with higher approval risks.

[0008] 2. In the present invention, all credit application data of suspected fraudulent users in the credit application process, excluding the vectorized processing results of the credit application information of suspected fraudulent users, are used to automatically generate credit reports using a large language model. This not only ensures the comprehensiveness of the data of suspected fraudulent users and accurately identifies the fraud risks of the target credit business, but also reduces the information dimension of the input data. On the basis of reducing the difficulty of large language model processing, the efficiency and accuracy of credit report generation and processing are guaranteed.

[0009] A further technical solution is that the target period ranges from 1 month to 3 months, and is specifically determined according to the number of credit application users of the target credit business.

[0010] A further technical solution is that the overdue user data includes the number of overdue users and the overdue amount of the overdue users.

[0011] A further technical solution is that all the credit application data of the suspected fraudulent user includes credit application information, communication information during the credit application process, and identity identification information.

[0012] A further technical solution is to automatically generate credit reports, specifically including: Using the vectorized processing results of all credit application data of the suspected fraudulent user and the vectorized processing results of the credit application information excluding the suspected fraudulent user as inputs to the large language model, and using the large language model to determine the analysis results of indicators in multiple dimensions based on the inputs; Based on the analysis results of indicators in multiple dimensions, credit reports are automatically generated according to preset templates.

[0013] In a second aspect, the present invention provides a computer system comprising: a memory and a processor in communication connection, and a computer program stored in the memory and capable of running on the processor, characterized in that when the processor runs the computer program, it executes the above-mentioned method for automated processing of credit business based on a large language model.

[0014] Other features and advantages will be described in the following description. The objectives and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the accompanying drawings.

[0017] Figure 1 It is a flow chart of a method for automating credit business processing based on a large language model; Figure 2 It is a flow chart of a method for determining whether a target credit business has approval risk; Figure 3 It is a flow chart to determine whether the changes in user credit risk within the target period meet the requirements; Figure 4 A flowchart of a method for identifying suspected fraudulent users among credit application users; Figure 5 It is a framework diagram of a computer system. DETAILED DESCRIPTION

[0018] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this specification without creative work should fall within the scope of protection of this specification.

[0019] When processing credit business, credit reports are often generated on a regular basis. Existing credit reports are often generated manually, resulting in problems such as high report production costs, high review costs, high costs for re-submissions, and lack of automatic pre-examination and verification capabilities.

[0020] In response to the above technical problems, the present invention proposes a method and system for automated processing of credit business based on a large language model.

[0021] Determine whether the target credit business has approval risks: Use the proportion of overdue users of the target credit business within the target period to determine whether the target credit business has approval risks, and define the target credit business with an overdue user proportion greater than 0.05 as a target credit business with approval risks.

[0022] Determine whether the changes in user credit risk within the target period meet the requirements: determine the average deviation coefficient of the comparison target period based on the deviation between the distribution ratio of credit application users in different credit risk intervals and the distribution ratio of the comparison target period, and determine the deviation target period in the comparison target period based on the average deviation coefficient of different comparison target periods. When the number of deviation target periods accounts for more than 0.2, it is determined that the changes in user credit risk within the target period do not meet the requirements.

[0023] Suspected fraudulent users: other credit applicants within a preset range of distance from the credit applicant are divided into the same applicant group, the number of credit applicants in the applicant group is determined, and similar information items with other credit applicants are used to determine other credit applicants with similar information. The fraud risk coefficient of the credit applicant is determined based on the number of credit applicants and the number of similar information users, and credit applicants with a fraud risk coefficient greater than 0.6 are regarded as suspected fraudulent users.

[0024] The vectorized processing results of all credit application data of suspected fraudulent users and the vectorized processing results of credit application information excluding the suspected fraudulent users are used as the input of the large language model. Based on the input, the large language model is used to determine the analysis results of indicators in multiple dimensions. Based on the analysis results of indicators in multiple dimensions, a credit report is automatically generated according to a preset template.

[0025] Example 1 To solve the above problem, according to one aspect of the present invention, Figure 1 As shown, a method for automatic processing of credit business based on a large language model is provided, which specifically includes: S1 obtains overdue user data of a target credit business within a target period, and when the target credit business is determined to have an approval risk using the overdue user data, proceeds to the next step; Furthermore, the target period ranges from 1 month to 3 months, and is specifically determined according to the number of credit application users of the target credit business.

[0026] Optionally, the overdue user data includes the number of overdue users and the overdue amount of the overdue users.

[0027] It should be noted that if Figure 2 As shown, it is determined that the target credit business has approval risks, specifically including: Determining the overdue amounts of different overdue users based on the overdue user data of the target credit business within the target period; Dividing the overdue users into different overdue amount intervals based on the overdue amount, and determining the overdue proportion factors for the different overdue amount intervals according to the proportion of overdue users in the credit application users in the different overdue amount intervals; Determine the preset weight coefficients for different overdue amount intervals, and determine the approval risk factor of the target credit business in combination with the overdue proportion factors of different overdue amount intervals, and determine whether the target credit business has approval risk through the approval risk factor.

[0028] Furthermore, determining whether the target credit business has an approval risk based on the approval risk factor specifically includes: When the approval risk factor is not within a preset range, it is determined that the target credit business has an approval risk.

[0029] It is understandable that when there is no approval risk in the target credit business, the credit report is automatically generated using a large language model based on the vectorized processing results of the credit application information of the credit applicant.

[0030] Optionally, determining whether the target credit business has approval risk specifically includes: Determining the overdue amounts of different overdue users based on the overdue user data of the target credit business within the target period, and determining the overdue risk factor of the target credit business within the target period based on the average of the overdue amounts of different overdue users; Determining the proportion of overdue users of the target credit business within a target period based on the analysis results of the overdue user data; An approval risk factor is determined based on the product of the overdue risk factor and the proportion of overdue users, and whether the target credit business has an approval risk is determined by the approval risk factor.

[0031] It is understandable that the overdue risk factor of the target credit business within the target period is determined based on the preset risk factor corresponding to the average value of the overdue amount.

[0032] Optionally, determining whether the target credit business has approval risk specifically includes: S11 determines the overdue amounts of different overdue users based on the overdue user data of the target credit business within the target period; S12 classifies the overdue users into different overdue amount intervals based on the overdue amounts, and determines interval risk factors for the different overdue amount intervals according to the number of overdue users in the different overdue amount intervals and the overdue amounts of the different overdue users; S13 determines preset weight coefficients for different overdue amount intervals, and determines the approval risk factor of the target credit business in combination with the interval risk factors of different overdue amount intervals, and determines whether the target credit business has approval risk through the approval risk factor.

[0033] Optionally, the above step S11 includes steps S111-S113, specifically: S111 determines the number of overdue users of the target credit business within the target period based on the overdue user data of the target credit business within the target period. If the number of overdue users of the target credit business within the target period does not meet the requirement, it is determined that there is an approval risk for the target credit business. If the number of overdue users of the target credit business within the target period meets the requirement, the process proceeds to step S112. S112 determines the overdue amounts of different overdue users. When the number of overdue users of the target credit business within the target period is greater than the preset number of overdue users, the process proceeds to step S113. When the number of overdue users of the target credit business within the target period is not greater than the preset number of overdue users, the process directly proceeds to step S12. S113 When the average value of the overdue amounts of different overdue users is not within the preset overdue amount range, it is determined that the target credit business has an approval risk. When the average value of the overdue amounts of different overdue users is within the preset overdue amount range, proceed to step S12.

[0034] Optionally, the above step S12 includes steps S121-S123, specifically: S121 classifies the overdue users into different overdue amount intervals based on the overdue amount. If the number of overdue users within the preset overdue amount interval does not meet the requirement, it is determined that the target credit business has an approval risk. If the number of overdue users within the preset overdue amount interval meets the requirement, the process proceeds to step S122. S122 determines interval risk factors for different overdue amount intervals based on the number of overdue users within different overdue amount intervals and the overdue amounts of different overdue users. If the interval risk factors within the preset overdue risk intervals do not meet the requirements, it is determined that the target credit business has an approval risk. If the interval risk factors within the preset overdue risk intervals meet the requirements, the process proceeds to step S123. S123 uses the preset risk factor threshold to determine the overdue risk intervals whose interval risk factors do not meet the requirements. When the number of overdue risk intervals whose interval risk factors do not meet the requirements does not meet the requirements, it is determined that the target credit business has approval risks. When the number of overdue risk intervals whose interval risk factors do not meet the requirements meets the requirements, proceed to step S13.

[0035] Furthermore, the overdue risk factor of the target credit business within the target period is determined based on a preset risk factor corresponding to the average value of the overdue amount.

[0036] S2 determines the identification result of the credit risk of the credit application information of the credit application users in the target period in the credit approval model, and through the changes in the distribution data of the credit application users in different credit risk intervals in the target period, determines that the changes in the user credit risk in the target period meet the requirements, and then proceeds to the next step; Specifically, the change in the distribution data of the credit application user is determined based on the deviation from the distribution data of the target credit business within a preset number of target cycles.

[0037] Specifically, such as Figure 3 As shown, it is determined that the changes in the user's credit risk within the target period meet the requirements, specifically including: Using a preset number of target cycles of the target credit business as comparison target cycles, and determining the distribution ratio of credit applicants in different credit risk intervals based on the distribution data of credit applicants in different credit risk intervals within the target cycles; Determining, based on the distribution ratios of credit applicants within different credit risk intervals, deviations from the distribution ratios of different comparison target periods; determining, based on the deviations of the distribution ratios, deviation intervals within the credit risk intervals; and determining, based on the number of the deviation intervals, a distribution deviation factor for the comparison target period; The variation coefficient of the user credit risk within the target period is determined by taking the average value of the distribution deviation factors of different comparison target periods, and whether the variation of the user credit risk within the target period meets the requirements is determined by the variation coefficient.

[0038] Furthermore, when the change in the user's credit risk within the target period does not meet the requirements, the credit report is automatically generated using a large language model based on the vectorized processing results of the credit application information of the credit applicant.

[0039] It can be understood that the deviation range is the credit risk range in which the deviation of the distribution ratio does not meet the requirements.

[0040] It should also be noted that the determination of whether changes in user credit risk within the target period meet the requirements specifically includes: Using a preset number of target cycles of the target credit business as comparison target cycles, and determining the distribution ratio of credit applicants in different credit risk intervals based on the distribution data of credit applicants in different credit risk intervals within the target cycles; Determining deviations from distribution ratios of credit applicants in different credit risk intervals based on the distribution ratios of credit applicants in different comparison target periods, and determining an average deviation coefficient for the comparison target period based on the deviations of the distribution ratios; The deviation target period in the comparison target period is determined by using the average deviation coefficient of different comparison target periods, and the number of the deviation target periods is used to determine whether the change in the user credit risk within the target period meets the requirements.

[0041] Furthermore, the average deviation coefficient of the comparison target period is determined according to an average value of deviations of distribution ratios of the comparison target period in different credit risk intervals.

[0042] Optionally, determining whether the change in the user's credit risk within the target period meets the requirements specifically includes steps S21-S23: S21 uses a preset number of target cycles of the target credit business in the vicinity as comparison target cycles, and determines the distribution ratio of credit applicants in different credit risk intervals based on the distribution data of credit applicants in different credit risk intervals in the target cycles; S22: determining, based on the distribution ratios of credit applicants within different credit risk intervals, deviations from the distribution ratios of different comparison target periods; and determining, based on the deviations of the distribution ratios within the different credit risk intervals, deviation intervals within the credit risk intervals; and determining, based on the number of deviation intervals and the deviations of the different deviation intervals, a distribution deviation factor for the comparison target period; S23 determines the weight coefficients of different comparison target periods based on the time intervals between different comparison target periods and the target period, and determines the coefficient of variation of the user credit risk within the target period in combination with the distribution deviation factors of the different comparison target periods, and determines whether the change of the user credit risk within the target period meets the requirements through the coefficient of variation.

[0043] S3 determines the location distribution data of the credit applicant based on the analysis results of the credit application information of the credit applicant, and determines suspected fraudulent users among the credit applicants by combining the analysis results of the credit application information of different credit applicants; Furthermore, the location distribution data of the credit application user is determined based on the analysis result of the positioning data corresponding to the credit application information of the credit application user.

[0044] It should be noted that if Figure 4 As shown, the method for determining suspected fraud users among the credit application users is: Based on the location distribution data of the credit applicant, other credit applicants within a preset distance from the credit applicant are divided into the same applicant group, and the number of credit applicants in the applicant group is determined; Obtaining similarities between the credit application information of the credit applicant and other credit application users, and using the similarities to determine the ratio of similar information items with other credit application users, and determining similar information users among other credit application users based on the ratio of similar information items; The fraud risk coefficient of the credit application user is determined based on the number of the credit application users and the number of users with similar information, and the suspected fraud users among the credit application users are determined based on the fraud risk coefficient.

[0045] Furthermore, the fraud risk coefficient of the credit application user is determined based on the product of the number of credit application users, the number of users with similar information, and a preset proportional factor.

[0046] S4 vectorizes all the credit application data of the suspected fraudulent user during the credit application process, and uses a large language model to automatically generate a credit report based on the vectorized processing results of the credit application information excluding the suspected fraudulent user.

[0047] It is understandable that all the credit application data of the suspected fraudulent user include credit application information, communication information during the credit application process, and identity identification information.

[0048] It should be noted that the automatic generation of credit reports specifically includes: Using the vectorized processing results of all credit application data of the suspected fraudulent user and the vectorized processing results of the credit application information excluding the suspected fraudulent user as inputs to the large language model, and using the large language model to determine the analysis results of indicators in multiple dimensions based on the inputs; Based on the analysis results of indicators in multiple dimensions, credit reports are automatically generated according to preset templates.

[0049] Optionally, before entering step S21, the following steps S201-S202 are further included, specifically: S201 obtains the user credit risk of credit applicants within a target period. If the number of credit applicants whose user credit risk is greater than a preset risk coefficient does not meet the requirement, it is determined that the change in user credit risk within the target period does not meet the requirement. If the number of credit applicants whose user credit risk is greater than the preset risk coefficient meets the requirement, the process proceeds to step S202. S202 determines the average value of the user credit risk of the credit application users within the target period. When the average value of the user credit risk is not within the preset credit risk range, it is determined that the change in the user credit risk within the target period does not meet the requirements. When the average value of the user credit risk is within the preset credit risk range, proceed to step S21.

[0050] Optionally, the above step S21 includes steps S211-S213, specifically: S211 uses a preset number of target cycles of the target credit business as comparison target cycles. When the deviation between the average value of the user credit risk of the credit applicant in the target cycle and that of other comparison target cycles is not within the preset risk deviation range, it is determined that the change in the user credit risk in the target cycle does not meet the requirement. When the deviation between the average value of the user credit risk of the credit applicant in the target cycle and that of other comparison target cycles is within the preset risk deviation range, the process proceeds to step S212. S212: When the number of credit application users whose user credit risk in the target period is greater than the preset risk coefficient and the deviation from that of other target periods for comparison is not within the preset deviation range, it is determined that the change in user credit risk in the target period does not meet the requirement; when the number of credit application users whose user credit risk in the target period is greater than the preset risk coefficient and the deviation from that of other target periods for comparison is within the preset deviation range, the process proceeds to step S213; S123 determines the distribution ratio of credit application users in different credit risk intervals based on the distribution data of credit application users in different credit risk intervals in the target period, and proceeds to step S13.

[0051] Optionally, the above step S22 includes steps S221-S223, specifically: S221 determines the deviations from the distribution ratios of the credit application users in different credit risk intervals from the distribution ratios of different comparison target periods, and determines the deviation intervals in the credit risk intervals based on the deviations of the distribution ratios in the different credit risk intervals. If the number of comparison target periods with the deviation intervals does not meet the requirement, it is determined that the change in the user credit risk in the target period does not meet the requirement. If the number of comparison target periods with the deviation intervals meets the requirement, the process proceeds to step S222. S222: The comparison target period in which the number of deviation intervals is greater than the preset number of intervals is regarded as the deviation target period. If there is a deviation from the target period, it is determined that the change in the user's credit risk within the target period does not meet the requirements. If there is no deviation from the target period, the process proceeds to step S223. S223 determines the distribution deviation factor of the comparison target period based on the number of the deviation intervals and the deviation conditions of different deviation intervals, and when there is no comparison target period whose distribution deviation factor does not meet the requirements, enters step S23. When there is a comparison target period whose distribution deviation factor does not meet the requirements, it is determined that the change in the user credit risk within the target period does not meet the requirements.

[0052] Example 2 In the second aspect, as Figure 5 As shown, the present invention provides a computer system, comprising: a memory and a processor in communication connection, and a computer program stored in the memory and capable of running on the processor, characterized in that: when the processor runs the computer program, it executes the above-mentioned method for automated processing of credit business based on a large language model.

[0053] Optionally, determining a suspected fraudulent user among the credit application users includes steps S31-S33, specifically: S31, based on the location distribution data of the credit applicant, classifies other credit applicants within a preset range from the credit applicant into the same applicant group, and determines the applicant clustering coefficient of the credit applicant based on the number of credit applicants in the applicant group and the distances between different credit applicants; S32: obtaining similarities between the credit application information of the credit applicant and other credit applicants, and using the similarities to determine the ratio of similar information items with other credit applicants, determining information-similar users among other credit applicants based on the ratio of similar information items, and determining the information similarity coefficient of the credit applicant based on the number of similar information users, the number of similar information items of different similar information users, and the distance between the similar information users and the credit applicant; S33 determines the fraud risk coefficient of the credit application user based on the information similarity coefficient and the application user clustering coefficient, and determines the suspected fraud users among the credit application users based on the fraud risk coefficient.

[0054] Optionally, before entering step S31, it is also necessary to judge the similarity of the credit application information between the credit application user and other credit application users. When it is determined based on the similarity that there is no user with similar information to the credit application user, it is directly determined that the credit application user is not a suspected fraudulent user. When it is determined based on the similarity that there is a user with similar information to the credit application user, step S31 is entered again.

[0055] Optionally, the above step S31 includes steps S311 to S313, which are specifically: S311 determines, based on the location distribution data of the credit applicant, whether there are other credit applicants within a preset range from the credit applicant. If so, it is determined that the credit applicant is not a suspected fraud user. If there are other credit applicants within the preset range from the credit applicant, the process proceeds to step S312. In step S312, other credit applicants within a preset range from the credit applicant are grouped into the same applicant group. When the number of credit applicants in the applicant group is greater than a preset user number threshold, the credit applicant is determined to be a suspected fraudulent user. When the number of credit applicants in the applicant group is not greater than the preset user number threshold, the process proceeds to step S313. S313 determines the applicant user clustering coefficient of the credit applicant user by the number of credit applicant users in the applicant user group and the distance between different credit applicant users. When the applicant user clustering coefficient of the credit applicant user does not meet the requirements, it is determined that the credit applicant user is a suspected fraudulent user. When the applicant user clustering coefficient of the credit applicant user does not meet the requirements, step S33 is entered.

[0056] Optionally, the above step S32 includes steps S321-S323, specifically: S321: obtaining similarities between the credit application information of the credit applicant and other credit applicants, and using the similarities to determine the ratio of similar information items with other credit applicants, and determining similar information users among other credit applicants based on the ratio of similar information items. When the number of similar information users is greater than a preset similar user number threshold, the credit applicant is determined to be a suspected fraudulent user. When the number of similar information users is not greater than the preset similar user number threshold, proceeding to step S322; S322 calculates the number of similar users in the applicant user group based on the distance between different similar users and the credit applicant. If the number of similar users in the applicant user group is greater than the set value, the credit applicant is determined to be a suspected fraudulent user. If the number of similar users in the applicant user group is not greater than the set value, the process proceeds to step S323. S323 determines the information similarity coefficient of the credit application user based on the number of information-similar users, the number of similar information items of different information-similar users, and the distance between the user and the credit application user. When the information similarity coefficient does not meet the requirements, it is determined that the credit application user is a suspected fraudulent user. When the information similarity coefficient meets the requirements, step S33 is entered.

[0057] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.

[0058] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The foregoing description is merely one or more embodiments of this specification and is not intended to limit this specification. It will be apparent to those skilled in the art that various modifications and variations may be made to one or more embodiments of this specification. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of one or more embodiments of this specification are intended to be within the scope of the claims of this specification.

Claims

1. A method for automatic processing of credit business based on a large language model, characterized in that: Specifically include: Acquire overdue user data of the target credit business within the target period, and when determining that the target credit business has approval risk using the overdue user data, proceed to the next step; Determine the identification result of the user credit risk of the credit application information of the credit application user within the target period in the credit approval model, and through the change of the distribution data of the credit application users in different credit risk intervals within the target period, determine that the change of the user credit risk within the target period meets the requirements, and then proceed to the next step; Based on the analysis results of the credit application information of the credit application user, determine the location distribution data of the credit application user, and determine the suspected fraudulent users among the credit application users in combination with the analysis results of the credit application information of different credit application users; All credit application data of the suspected fraudulent user during the credit application process are vectorized, and combined with the vectorized processing results of the credit application information excluding the suspected fraudulent user, a large language model is used to automatically generate a credit report.

2. The method for automatically processing credit business based on a large language model according to claim 1, characterized in that: The target period ranges from 1 month to 3 months, and is specifically determined according to the number of credit application users of the target credit business.

3. The credit business automation processing method based on a large language model according to claim 1, characterized in that: The overdue user data includes the number of overdue users and the overdue amount of the overdue users.

4. The method for automatically processing credit business based on a large language model according to claim 1, characterized in that: Determine whether the target credit business has approval risks, including: Determining the overdue amounts of different overdue users based on the overdue user data of the target credit business within the target period; Based on the overdue amount, the overdue users are divided into different overdue amount intervals, and the overdue proportion factors of the different overdue amount intervals are determined according to the proportion of overdue users in the different overdue amount intervals among the credit application users; Determine the preset weight coefficients for different overdue amount intervals, and determine the approval risk factor of the target credit business in combination with the overdue proportion factors of different overdue amount intervals, and determine whether the target credit business has approval risk through the approval risk factor.

5. The method for automatically processing credit business based on a large language model according to claim 4, characterized in that: Determining whether the target credit business has an approval risk through the approval risk factor specifically includes: When the approval risk factor is not within a preset range, it is determined that the target credit business has an approval risk.

6. The credit business automation processing method based on a large language model according to claim 1, characterized in that: The change in the distribution data of the credit application user is determined based on the deviation from the distribution data of the target credit business within a preset number of target cycles.

7. The method for automatically processing credit business based on a large language model according to claim 1, characterized in that: Determine whether the changes in user credit risk during the target period meet the requirements, including: Taking the target credit business in the adjacent preset number of target cycles as the comparison target cycle, and determining the distribution ratio of the credit application users in different credit risk intervals based on the distribution data of the credit application users in different credit risk intervals in the target cycle; Determine the deviation from the distribution ratio of different comparison target periods based on the distribution ratio of credit application users in different credit risk intervals, and determine the deviation interval in the credit risk interval based on the deviation of the distribution ratio, and determine the distribution deviation factor of the comparison target period based on the number of the deviation intervals; The coefficient of variation of the user credit risk within the target period is determined by taking the average value of the distribution deviation factors of different comparison target periods, and whether the variation of the user credit risk within the target period meets the requirements is determined by the coefficient of variation.

8. The method for automatically processing credit business based on a large language model according to claim 7, characterized in that: When the change of the user's credit risk within the target period does not meet the requirements, the credit report is automatically generated using a large language model based on the vectorized processing result of the credit application information of the credit applicant.

9. The credit business automation processing method based on a large language model as claimed in claim 8, characterized in that: The deviation range is the credit risk range where the deviation of the distribution ratio does not meet the requirements.

10. A computer system comprising: A memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes a method for automated processing of credit business based on a large language model as described in any one of claims 1 to 9 when running the computer program.

Images