Credit management system, credit data mixed batch uploading method and server

Abstract

The invention discloses a credit management system, a credit data mixed batch uploading method and a server. The method comprises the following steps: receiving credit data submitted by a first user in batches through an uploading entrance; calling the target large model; providing the credit data to the target large model and inputting a first cue word; receiving a main category of each credit file returned by the target large model; calling the target large model again; providing each credit file and the main category thereof to the target large model, and inputting a second cue word; receiving a sub-category of each credit file returned by the target large model; and respectively storing the credit files into corresponding file directories according to the main categories and the sub-categories. According to the scheme, a traditional scattered uploading mode that single data corresponds to a single entrance is broken through, batch submission of multiple types of credit files is supported through the unified uploading entrance, a client manager can upload the credit files through one key without pre-classification, and the problem that traditional one-by-one entrance matching is tedious is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of credit risk management, and particularly relates to a credit management system and a credit data mixed batch uploading method and a server. BACKGROUND

[0002] In the field of credit business, in daily operations, a customer manager usually collects various credit data from a customer and then uploads the credit data one by one to each module entrance of a credit management system such as a credit system, a risk control platform and a monitoring platform for pre-loan risk assessment and in-loan risk monitoring. In this process, the uploading of credit data adopts a single data independent operation mode, and all materials need to be uploaded by the customer manager at a specific entrance according to the type of the material. For example, for the two common types of bank statement and financial report, the two types of data need to be uploaded at different entrances of the system. When a large amount of data is involved, it is very easy to cause misuploading and wrong uploading, and the operation efficiency is very low. SUMMARY

[0003] The present application aims to provide a credit data mixed batch uploading method to improve the operation efficiency and reduce the possibility of misuploading and wrong uploading. Another object of the present application is to provide a server for credit data processing. Another object of the present application is to provide a credit management system which receives and classifies credit data uploaded by a user based on a unified entrance using the credit data mixed batch uploading method and uses the classified credit data for subsequent analysis.

[0004] A credit data mixed batch uploading method comprises the following steps: receiving credit data submitted in batches by a first user through an uploading entrance, wherein the credit data comprises at least two types of credit files; calling a target large model; providing the credit data to the target large model and inputting a first prompt word to make the target large model return a main category of each credit file; the first prompt word comprises a plurality of predefined main categories and a feature description of each main category; receiving the main category of each credit file returned by the target large model; calling the target large model again; providing each credit file and its main category to the target large model and inputting a second prompt word to make the target large model return a subcategory of each credit file; the second prompt word comprises a predefined subcategory of each main category and a classification constraint of each subcategory; receiving a sub-category of each credit file returned by the target large model; storing each credit file in a corresponding file directory according to the main category and the sub-category.

[0005] In the credit information mixed batch uploading method described above, based on the customer selected by the first user, each credit file is stored in the corresponding file directory in the storage space allocated to the corresponding customer according to the sub-category classification result; Based on the customer identity authentication information of the second user, the credit files uploaded by the second user are classified and stored in the corresponding file directory of the corresponding customer.

[0006] In the credit information mixed batch uploading method described above, the method further comprises: extracting a serial file of a target email from a target mailbox based on a preset rule; The target email is sent to the target mailbox by the serial bank indicated by the second user; Identify the target email and the corresponding customer based on preconfigured information; Store the serial file in the corresponding file directory of the corresponding customer.

[0007] In the credit information mixed batch uploading method described above, the method further comprises: Receiving a file key uploaded by the second user; Extracting an encrypted serial file of a target email from a target mailbox based on a preset rule; The target email is sent to the target mailbox by the serial bank indicated by the second user; Identify the target email and the corresponding customer based on preconfigured information; Decrypt the encrypted serial file using the file key uploaded by the corresponding customer; Store the serial file in the corresponding file directory of the corresponding customer.

[0008] In the credit information mixed batch uploading method described above, the preset rule is to poll the target mailbox according to the set time to find new emails, and when new emails are found, to determine whether the new emails are target emails based on preconfigured information; In response to the new email being a target email, extract the serial file of the new email, and determine the corresponding customer of the target email according to the email content.

[0009] In the credit information mixed batch uploading method described above, the first prompt word is used to indicate that the target model returns the main category of each credit file according to the feature description of each main category; The second prompt word includes a first dictionary and a second dictionary; The first dictionary contains the classification constraints of each sub-category of each main category. The second dictionary contains a predefined set of subcategories for each main category; The second prompt word is used to instruct the target model to return the sub-category of each credit document according to the first dictionary and the second dictionary.

[0010] In the above-mentioned method for batch uploading credit data, the target large model is obtained by performing two-stage fine-tuning training on the pre-trained large model. Furthermore, the two-stage fine-tuning training method includes: The large model was fine-tuned in the first stage using credit data and main classification labels. After the first stage of fine-tuning training, the large model is then fine-tuned in the second stage using credit data that has been classified according to the main category and the sub-classification labels of each credit document.

[0011] In the above-mentioned method for batch uploading credit information, the predefined main categories include any combination of the following: certificates, real estate ownership certificates, asset certificates, liabilities, financial statements, business contracts, business materials, inquiry authorization, transaction records, employment certificates, invoices, tax returns, and other categories. Each main category is predefined with multiple subcategories.

[0012] A server, comprising: The credit data batch receiving module is used to receive credit data submitted in batches by the first user through the upload portal. The credit data includes at least two types of credit documents. The large model invocation module is used to invoke the target large model twice to obtain the sub-category of each credit file; The prompt word module is used to provide the credit information to the target large model and input a first prompt word to make it return the main category of each credit document; and to provide each credit document and its main category to the target large model and input a second prompt word to make it return the sub-category of each credit document; The first prompt word contains multiple predefined main categories and feature descriptions for each main category; The second prompt word includes a predefined subcategory for each main category and a classification constraint for each subcategory; The categorized storage module is used to store each credit file into its corresponding file directory according to its main category and subcategory.

[0013] The servers mentioned above also include: The email recognition module is used to connect to the target email address and identify the target email and its associated customer based on the configuration information. The target email was sent to the target mailbox by a bank account instructed by the second user. A key obtaining module is configured to receive a file key uploaded by the second user. A decryption module is configured to decrypt the encrypted log file extracted from the target mail using the file key uploaded by the corresponding client. A mail extracting module is configured to extract the log file of the target mail and store the unencrypted log file or the decrypted log file in the corresponding file directory of the client.

[0014] A credit management system includes a credit analysis module, and further includes the server and the like. A first client is configured to be used by the first user and has a unified upload portal for the first user to batch submit credit data. The credit analysis module is configured to perform credit risk analysis on a target client based on credit files in the file directory of the target client.

[0015] In the credit management system, the server is included, and further includes the following. A second client is configured to be used by the second user and has a plurality of special upload portals for various types of credit files. The second client further has a file key upload portal for the second user to upload a file key.

[0016] The credit management system has the following advantages: the traditional single data corresponding to a single portal dispersed upload mode is broken through, and the unified upload portal supports batch submission of multiple types of credit files, so that the credit manager can one-click upload without pre-classification, and the cumbersome problem of one-by-one matching portal is solved. Meanwhile, the automatic classification and classified storage of mixed batch data are realized by calling a large model, especially twice calling a large model and a specific form of prompt word, compared with the traditional data collection mode, the operation time of manual input system is greatly reduced, the operation efficiency is effectively improved, and the high-precision classification effect is ensured. The credit management system combines the three data collection methods of batch upload by the credit manager, classified upload by the client and automatic acquisition of mail, improves the data collection efficiency and convenience through multi-channel data collection, and reduces the operation burden of all parties. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A method flowchart of the credit data mixed batch upload method in the first embodiment of the present application is shown. Figure 2 A work flowchart of the credit management system in the first embodiment of the present application is shown. Figure 3 A system block diagram of the credit management system in the first embodiment of the present application is shown. Figure 4 A work flowchart of the credit management system in the second embodiment of the present application is shown. Figure 5This is a schematic diagram illustrating the multi-client credit file upload and categorized storage to a hierarchical file directory in Embodiment 2 of the present invention; Figure 6 This is a system block diagram of the credit management system in Embodiment 3 of the present invention; Figure 7 This is a flowchart of the method for processing unencrypted target emails in Embodiment 3 of the present invention; Figure 8 This is a flowchart of the method for processing encrypted target emails in Embodiment 3 of the present invention.

[0018] Figure labeling: Server 1; Credit data batch receiving module 11; Large model calling module 12; Prompt word module 13; Classification storage module 14; Email recognition module 15; Key acquisition module 16; Decryption module 17; Email extraction module 18; First client 2; Second client 3; Target large model 4; Target mailbox 5. Detailed Implementation

[0019] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0020] Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for batch uploading mixed credit information, which includes: Receive credit information about a customer submitted in batches by the first user through a unified upload portal. The credit information includes at least two types of credit documents, and each type of credit document may have one or more documents. Call the target large model; Provide the credit data to the target large model and input the first prompt word to make it return the main category of each credit file; The first prompt word contains multiple predefined main categories and feature descriptions for each main category; Receive the main category of each credit file returned by the target large model; Call the target large model again; Each credit file and its main category are provided to the target large model, and a second prompt word is input to make it return the sub-category of each credit file; The second prompt includes predefined subcategories for each main category and classification constraints for each subcategory; Receive the sub-category of each credit file returned by the target large model; Each credit document is stored in the corresponding file directory within the storage space allocated to the corresponding customer, based on its main category and subcategory.

[0021] The aforementioned credit documents can be of various file types, such as images and PDFs. For multi-page PDFs, they will be split into single-page files for processing.

[0022] Preferably, after receiving the above credit information, the type of the credit file is judged. If the file type is a picture, image quality detection and processing are performed. The completeness of the picture (obstruction, corner, edge, etc.), the clarity (blur, reflection, etc.) are detected. If there is a problem, the picture is completed, enhanced, corrected, watermarked, and the seal is extracted to improve the accuracy of subsequent recognition.

[0023] Specifically, the target large model is obtained by second-order fine-tuning training of a pre-trained large model. In this embodiment, a general multi-modal large model, i.e., Qianwen large model, is used. The existing classified information is used to fine-tune the Qianwen large model to obtain the target large model required for use. Other similar multi-modal large models can be used when put into use. The target large model should not be limited to the pre-trained Qianwen large model. The fine-tuning training is as follows: The first stage fine-tuning training of the large model is performed using the credit information and the main classification label, so that the large model has the ability to classify the main categories of each credit file in the credit information.

[0024] After the first stage fine-tuning training, the second stage fine-tuning training of the large model is performed using the credit information classified according to the main categories and the sub-classification label of each credit file, so that the large model has the ability to classify the sub-categories of each credit file in the credit information.

[0025] The target large model obtained by the above fine-tuning training achieves high-precision classification effect in combination with the way of calling the large model twice for classification and the specific form of prompt words.

[0026] Specifically, the first prompt word is used to instruct the target model to return the main category of each credit file according to the feature description of each main category. The specific prompt word provided in this embodiment is as follows: "Image material classification, please classify the pictures according to the following category definitions: Certificate class: standardized special certificates, including business license, tax registration certificate (not including tax return), company charter (not contract), organization code certificate, agency credit code certificate, company registration or change registration, driving license, driving license, motor vehicle registration certificate, opening license, taxi operation license, special industry qualification certificate, and personal ID (including ID national emblem, ID portrait, ID front and back, ID photo), household register, customer head portrait, network verification photo, marriage certificate, divorce certificate, school certificate, residence certificate. (Non-certificate photos such as various lists are not included in this category); Real estate right class: including house property certificate, land use certificate, land certificate, real estate certificate, house certificate, house property network query result, rural self-built house ownership certificate, and house purchase contract; Proof of assets: insurance policy, car purchase invoice, deposit, acceptance bill, inventory (inventory photos, inventory list, inventory list, etc.), various investments, intangible assets (intellectual property, reputation, brand value, data assets), production equipment (including equipment photos, equipment list and equipment ownership certificate), various asset use rights (farmland, forest, sea area), live breeding, ship certificate, other assets and other materials; Liabilities: including loan documents, loan cards, and loan contracts, and other liabilities, excluding bank statements and account details; Financial statements: only include profit table (including profit and loss table), cash flow table, balance sheet, financial indicators report, financial statements, audit report materials, and table headers must have these titles, and also include account books, ledgers, accounts payable, pre-paid accounts, accounts receivable, pre-paid accounts Non-standardized financial statements; Business contracts: business-related contracts, including sales contracts, lease contracts, procurement contracts, purchase and sale contracts, agreements, agreements, agreement documents, and other materials; Business materials: enterprise classification, tax payment certificate, employee salary sheet (including employee salary sheet), employee personnel list (including employee roster), water and electricity bill (including water and electricity bill), warehouse in and out sheet, company paid social security and provident fund certificate (not individual social security and provident fund), customs import and export data certificate, customs export goods declaration and query record; construction project list, in addition to, partnership agreement, capital verification report, etc. Query authorization: enterprise credit inquiry authorization, Zhejiang Provincial Taxation Authorization, Shanghai Big Data Authorization, post-loan query authorization, and other external data authorization, enterprise credit report and other categories; Transaction flow: various bank transaction flows, including bank account details or bank deposit details, and various bank detail data tables, including account number, transaction time, counterpart account, counterpart name, transaction amount, balance, and summary information, single transfer record, balance screenshot; Proof of work: proof of work, labor contract, job description, provident fund monthly payment certificate, social security certificate, pay slip, income certificate, etc., among which social security and provident fund certificates are personal, not company-paid employee payment certificates; Invoices: various invoices, such as housing purchase invoices, car purchase invoices, water and electricity invoices, transportation invoices, rental invoices, customs declaration fee invoices, goods and services (purchase and sale invoices), and other invoices; Tax return: various tax returns, such as value-added tax and additional tax return, value-added tax return, enterprise income tax annual report, enterprise income tax quarterly report, etc. Standard tax return, pay attention to table header information; Other: None of the above categories, such as credit reference defense, court defense, other defense, screenshot of Qichacha, seal card, enterprise credit report, personal credit report, relevant materials of the Credit Reference Council, special business approval form of centralized operation platform, on-site investigation photos of client managers (indoor and outdoor photos, including people and background).

[0027] Focus of classification basis: 1. Picture header information, 2. Picture main information, 3. Picture tail information, 4. Except for other categories, other categories do not contain people photos, 5. Certificate category must be strictly required, do not put other into enterprise certificate category, 6. Purchase contract cannot be placed in real estate right category.

[0028] According to the definition of each category above, the pictures are classified as follows:

certificate category, real estate right category, asset proof category, liability category, financial statement category, business contract category, business material category, query authorization category, transaction flow category, work proof category, invoice category, tax return form, other category, blank page

[0029] A certain client manager collects the following credit information at customer A: customer ID card (2, P1, P2) obtained by mobile phone, marriage certificate (1, P3), business license (1, P4), purchase and sale contract (10, P5~P14), real estate certificate (1, P15), picture form; financial statements (PDF1), Y1 bank flow file (PDF2), PDF form. The client manager selects customer A in the system and uploads these credit files to the system through the unified upload portal. After calling the target large model and inputting the above prompt words, the system gets the main category of each credit file returned by the large model, as follows: {"P1":"certificate category", "P2":"certificate category", "P3":"certificate category", "P4":"certificate category", "P5":"business contract category", "P6":"business contract category", "P7":"business contract category", "P8":"business contract category", "P9":"business contract category", "P10":"business contract category", "P11":"business contract category", "P12":"business contract category", "P13":"business contract category", "P14":"business contract category", "P15":"real estate right category", "PDF1":"financial statement category", "PDF2":"transaction flow category"}; The second prompt includes a first dictionary and a second dictionary. The first dictionary contains classification constraints for each subcategory of each main category, and the second dictionary contains a predefined set of subcategories for each main category. The second prompt is used to instruct the target model to return the subcategory of each credit document according to the first dictionary and the second dictionary.

[0030] In this embodiment, dictionary one and dictionary two are as follows: Dictionary 1: "{"Document Category": "Standardized special-purpose documents, including ID cards (national emblem side, portrait side, front and back integrated (excluding other images with portraits, such as marriage certificates, divorce certificates, etc.)), household registration books, marriage certificates (with the red seal of the Civil Affairs Department), divorce certificates, student status certificates, residence permits, vehicle registration certificates, motor vehicle registration certificates, driver's licenses, taxi operating permits, original business licenses (must be titled "Business License"), duplicate business licenses (must be titled "Business License" with "Duplicate" below, otherwise they are not considered business licenses), organization code certificates (titled "Organization Code Certificate of the People's Republic of China"), duplicate organization code certificates (titled "Organization Code Certificate of the People's Republic of China" with "Duplicate" below). These document categories must be standard format documents; otherwise, they are not classified as such, such as tax information, enterprise credit reports, and Qichacha (a business information platform)." "Real Estate Ownership": "Real estate ownership includes property ownership certificates, land use certificates, land certificates, real estate certificates, property ownership certificates, real estate website search results, ownership certificates for self-built rural houses, etc.; purchase contracts: excluding mortgage appraisal reports and agreements, which must contain the words 'commercial housing'; other categories are excluded." "Asset Proof Category": "Inventory: Inventory photos, detailed inventory list, stock list, etc.; Intangible Assets: Including intellectual property rights, reputation, goodwill, brand value, and data assets; Ownership Certificates for Major Production Equipment: In addition to photos showing a large number of equipment, equipment lists are also acceptable; Equity Investments: Excluding chat screenshots and other images; Ship certificates and fixed asset lists are considered other asset proof categories." "Liabilities": "Loan vouchers: including loan contracts, IOUs and other related vouchers. Other corporate liabilities besides loan vouchers and loan cards are considered liabilities." "Financial Statements": "Including income statement (including profit and loss statement), cash flow statement, balance sheet, and audit report. The title of the image includes the names of these reports. If the account is a single item, it is accounts receivable, accounts payable, advances from customers, and advances from customers. If it includes accounts payable, accounts receivable, advances from customers, and advances from customers, it is a ledger. Otherwise, it is other financial statements." "Business Contracts": "Purchase and Sale Contracts: Purchase contracts, engineering contracts, power supply contracts, service contracts, notices of award, purchase orders, and agreements belong to purchase and sale contracts; Lease Contracts: These are contracts strictly related to leasing." "Business materials": "Employee list: company employee list including name list; tax payment: title including tax payment photo; water and electricity bill: water and electricity bill and proof; customs import and export data: customs export goods declaration and query record; company social security payment (participation) proof: company's proof of employee participation, not personal participation; other business situation data: including list of ongoing or completed engineering projects, payroll, company's social security payment proof, etc. Materials related to company operation, on-site photos including indoor and outdoor photos with people, on-site investigation photos of account managers (indoor and outdoor photos or photos with people)"; "Query authorization": "Credit inquiry authorization distinguishing between individuals and enterprises, distinguished by title, Zhejiang Provincial Taxation Authorization, Shanghai Big Data Authorization: content includes Shanghai Public Data Open Inquiry Authorization and other external data authorization. Photos must have the title of the authorization category, otherwise it is other query authorization such as bill business acceptance and business application"; "Transaction flow": "Various transaction flows, bank transaction flow: including account number, transaction time, opposite account, opposite name, transaction amount, balance, summary, etc. Information, and multiple transaction records. If the flow note, summary, and remarks are stock transfer, it is external investment. If the note, summary, and remarks are related to water and electricity fees, it is a water and electricity fee voucher. If the note, summary, and remarks are wages, it is a payroll. Otherwise, it is other flow, such as non-standard bank flow, single transfer record, and balance screenshot"; "Invoice": "Purchase, purchase, water and electricity, transportation, rental, customs declaration fee, commodity and service (purchase and sale invoice), customs declaration fee invoice, otherwise it is other invoice; water and electricity invoice: including water supply and electricity supply information; rental invoice: project name is business rental or rent, etc. The priority of invoice classification is project name, followed by seller company name"; "Work proof": "Work proof, labor contract, job proof, public accumulation fund monthly payment proof, social security certificate, payroll, income proof, etc. Otherwise, it is other work proof"; "Tax return": "Various tax returns, such as value-added tax and additional tax return, value-added tax return, enterprise income tax annual return, enterprise income tax quarterly return, etc. Standard tax return, otherwise it is other tax proof"; "Other": "Qichacha screenshot: Qichacha photo screenshot must contain Qichacha, on-site photo: indoor and outdoor photos with people: customer manager on-site investigation photo (indoor and outdoor photo, including people and background), storefront photo: no people, including building, sign photo, credit report: including personal and enterprise credit report, otherwise it is other"; "Blank page": "Blank page: no words or information, otherwise it is other category"}

[0031] Dictionary 2: "{"Document Classes":["ID Card","Household Registration Book","Marriage Certificate","Divorce Certificate","Student Status Certificate","Residence Permit","Vehicle Registration Certificate","Motor Vehicle Registration Certificate","Driver's License","Taxi Operating Permit","Original Business License","Duplicate Business License","Original Organization Code Certificate","Duplicate Organization Code Certificate","Tax Registration Certificate","Organization Credit Code Certificate","Bank Account Opening Permit","Articles of Association - Non-Contractual","Special Industry Qualification Certificate","Company Registration or Change Registration","Other Document Classes",]; "Employment Certificates": ["Employment Certificate", "Employment Contract", "Employment Certificate", "Proof of Monthly Housing Provident Fund Contribution", "Social Security Certificate", "Payslip", "Income Certificate", "Other Employment Certificates"]; "Asset Certificates": ["Ownership Certificates of Major Production Equipment", "Deposit Certificates", "Accepted Bills of Exchange", "Inventory", "Rights to Use Rights of Rights Assets (Farmland, Forest, Sea Area)", "Long-Term Investments", "Intangible Assets", "Equity Investments", "Live Animals (Aquaculture)", "Insurance Policies", "Other Asset Certificates"]; "Liabilities": ["Loan Certificates", "Loan Cards", "Other Corporate Liabilities"]; "Real Estate Titles": ["Real Estate Titles", "Purchase Contracts", "Other Real Estate Titles"]; "Financial Statements": ["Balance Sheet", "Income Statement", "Cash Flow Statement", "Audit Report"] "Accounts Receivable", "Prepaid Expenses", "Accounts Payable", "Advances from Customers", "Ledgers", "Other Financial Statements"; "Business Contracts": ["Lease Contracts", "Purchase and Sale Contracts", "Other Business Contracts"]; "Operating Materials": ["Inbound Slip", "Outbound Slip", "Enterprise Classification", "Employee Roster", "Various Tax Returns", "Tax Payment Certificate", "Water Usage Receipts", "Electricity Usage Receipts", "Customs Import and Export Data Certificates", "On-site Photos - Indoor and Outdoor Photos Including People", "Company Social Security Payment (Participation) Certificate", "Capital Verification Report", "Accounts Receivable", "Prepaid Accounts", "Accounts Payable", "Advanced Revenue", "Ledgers", "Other Operating Information"]; "Inquiry Authorization Type": ["Real Estate Website Inquiry Authorization Letter", "Enterprise Credit Information Authorization Letter", "Personal Credit Information Authorization Letter", "Post-Loan Risk Management Inquiry Authorization Letter", "Zhejiang Provincial State Taxation Bureau Authorization Letter", "Shanghai Big Data Authorization Letter", "Other External Data Authorization Letter", "Other Inquiry Authorization Type"]; "Transaction slip class":["bank transaction slip","foreign equity investment","payroll","other slip class"]; "Tax return":["Various tax returns","Tax proof"]; "invoice":["purchase invoice","car purchase invoice","electricity invoice","water bill","transportation invoice","rental invoice","purchase and sale invoice","customs invoice","other invoice"]; "Other":["Other","Qichacha screenshot","On-site photos-Indoor and outdoor photos with people","Credit Reference Bureau materials","Centralized operation platform special business approval form","Credit report","Shop front photo"]; "Blank page":["Blank page","Other"]}”.

[0032] Prompt: "Classify various image materials according to the following category definitions, select the most appropriate one from these types based on the following category definitions {Dictionary 2}, and strictly follow the description of {Dictionary 1} as the classification basis; The focus of the classification basis is 1. The head information of the picture, the highest weight, 2. The main body information of the picture, 3. The tail information of the picture, 4. The title of the business license must have a business license, 5. The photo is not upright, rotate and correct it after recognition. According to the definition of each category, if the picture has no content or too little information, it is a blank page, select one from Dictionary 2 as the answer, do not output any content other than the options, output in JSON format, format as "Picture type":"xxx", output briefly, do not output any reason".

[0033] For the above A customer example, the target large model outputs {"P1":"identity card","P2":"identity card","P3":"marriage certificate","P4":"business license","P5":"purchase and sale contract","P6":"purchase and sale contract","P7":"purchase and sale contract","P8":"purchase and sale contract","P9":"purchase and sale contract","P10":"purchase and sale contract","P11":"purchase and sale contract","P12":"purchase and sale contract","P13":"purchase and sale contract","P14":"purchase and sale contract","P15":"real estate rights class","PDF1-1":"balance sheet","PDF1-2":"profit table","PDF1-3":"cash flow table","PDF2":"bank transaction slip"}.

[0034] Based on the above processing, the files P1, P2 will be stored in the "ID card" folder under the "certificate type" folder of the A customer; the file P3 will be stored in the "marriage certificate" folder under the "certificate type" folder; the file P4 will be stored in the "business license" folder under the "certificate type" folder; the files P5-P14 will be stored in the "purchase and sale contract" folder of the "business contract type" folder; P15 will be stored in the "real estate right" folder of the "real estate right type" folder; PDF1 is split into three parts, PDF1-1, PDF1-2, PDF1-3, which are stored in the "asset and liability table", "profit table", "cash flow table" folders of the "financial report type" folder respectively; PDF2 is stored in the "bank transaction flow" folder of the "transaction flow type" folder.

[0035] After the above storage is preferably completed, an artificial calibration task is initiated, and the classification result is notified to the customer manager or the centralized operation platform for confirmation. After the confirmation is completed, the files can be subjected to regular analysis and processing, such as material verification processing, risk analysis of the corresponding customer based on the material, risk monitoring, etc.

[0036] When in use, the specific content of the above prompt words can be appropriately modified.

[0037] The present scheme proposes the idea of unified upload portal in view of the current cumbersome credit material uploading problem. The customer manager can upload all materials at one key, without the need for artificial classification and separate portal uploading, which significantly improves the material uploading efficiency. And the high-precision automatic classification effect is realized by using twice large model calling, specific form prompt words and automatic classification storage mode based on large model output. As for the above prompt words, the classification accuracy can reach more than 95% through verification, which is a very obvious improvement compared with the one-time classification effect of the large model. The present scheme overcomes the hallucination problem and business logic errors of the large model by limiting the output range of the prompt words input by twice calling the large model, and by the first dictionary, the second dictionary and the main category classification feature description and the subcategory classification constraint.

[0038] Further, as shown in Figure 2 and Figure 3 , the present embodiment also provides a credit management system, comprising a credit analysis module, a server, a first client, wherein, the server is used to realize the above-mentioned credit material mixed batch uploading method, comprising: a credit material batch receiving module 11 for receiving the credit materials submitted by the first user through the upload portal in batches; a large model calling module 12 for calling a target large model 4 twice to obtain the subcategory of each credit file; The prompt word module 13 is configured to provide the credit information to the target large model 4, input a first prompt word, and return a main category of each credit file; and provide each credit file and the main category to the target large model 4, input a second prompt word, and return a subcategory of each credit file. The first prompt word comprises a plurality of predefined main categories and a feature description of each main category. The second prompt word comprises a predefined subcategory of each main category and a classification constraint of each subcategory. The classification storage module 14 is configured to store each credit file in a corresponding file directory according to the main category and the subcategory.

[0039] The first client 2 is used by a first user, i.e., a client manager in this embodiment, and has a unified upload entrance for the first user to submit credit information in batches. The credit analysis module analyzes the credit risk of a target client based on the credit files in the file directory of the target client, such as credit risk assessment and monitoring.

[0040] Embodiment Two This embodiment is similar to Embodiment One, except that, as shown in FIG. 2, the credit management system of this embodiment further provides a second client 3 for a client, i.e., a second user in this embodiment, to use, and the second client 3 has a plurality of special upload entrances for various types of credit files. Figure 4

[0041] This embodiment provides special upload entrances for the second client 3 for the seven common types of credit files, i.e., the seven types of credit files that the clients are more familiar with, such as an ID card, a social security certificate, a public accumulation fund certificate, an individual tax certificate, a bank transaction statement, a real estate right certificate, and a vehicle driving license, which correspond to the first to seventh upload entrances, respectively. The clients manually classify the credit files and upload them to different entrances, respectively. The credit files uploaded through different entrances will be directly stored in the corresponding file directories, as shown in FIG. 3. Figure 5

[0042] The storage rules of the credit files uploaded through the entrances are as follows: The first upload entrance (ID card) is a certificate type, i.e., an ID card. The second upload entrance (social security certificate) is a work certificate type, i.e., a social security certificate. The third upload entrance (public accumulation fund certificate) is a work certificate type, i.e., a public accumulation fund monthly payment certificate. The fourth upload entrance (individual tax certificate) is a work certificate type, i.e., other work certificates. The fifth upload entrance (bank transaction statement) is a transaction flow type, i.e., a bank transaction flow. The sixth upload entrance (real estate right certificate) is a real estate right type, i.e., a real estate right.​​ The seventh upload entry (vehicle driving license), the certificate type is driving license.

[0043] In this embodiment, the customer and the customer manager upload credit part credit data respectively. The customer manually classifies the data uploaded by the customer after completing identity authentication in the system, and then uploads the data. The system stores the corresponding files into the corresponding file directory according to the entry uploaded by the customer. The customer manager selects the customer and uploads all the credit data collected about the current customer by one key. After the credit data is classified by the method described in embodiment one, the data is stored in the corresponding file directory of the current customer. Different channel data collection and intelligent combination of different channel data are realized.

[0044] Embodiment three This embodiment is similar to embodiment two, and the difference is that, as shown in the figure, in the credit management system of this embodiment, the server further includes, Figure 6 a mail identification module 15, configured to connect to the target mailbox 5, identify the target mail and the corresponding customer based on the configuration information, and send the target mail to the target mailbox 5 by the second user; a key acquisition module 16, configured to receive the file key uploaded by the second user, and the second client 3 has a file key upload entry for the second user to upload the file key; a decryption module 17, configured to use the file key uploaded by the corresponding customer to decrypt the encrypted flow file extracted from the target mail; a mail extraction module 18, configured to extract the flow file of the target mail, and store the unencrypted flow file or the decrypted flow file into the corresponding file directory of the corresponding customer.

[0045] As shown in the figure, for the unencrypted target mail, the processing method is as follows: Figure 7 Poll the target mailbox according to the set time to find new mail; When new mail is found, determine whether the new mail is the target mail based on the pre-configuration information; In response to the new mail being the target mail, extract the flow file of the new mail, and determine the corresponding customer of the target mail according to the mail content; Store the flow file into the corresponding file directory of the corresponding customer, that is, the transaction flow type-bank transaction flow.

[0046] As shown in the figure, for the encrypted target mail, the processing method is as follows: Figure 8 Poll the target mailbox according to the set time to find new mail; When new mail is found, determine whether the new mail is the target mail based on the pre-configuration information; In response to the new mail being the target mail, extract the flow file of the new mail, and determine the corresponding customer of the target mail according to the mail content; ​​In response to the new mail being the target mail, the serial file of the new mail is extracted, and the client to which the target mail belongs is determined according to the mail content; receiving the file key uploaded by the second user; decrypting the encrypted serial file using the file key uploaded by the corresponding client; storing the serial file in the corresponding file directory of the corresponding client, i.e., the transaction serial file class-bank transaction serial file.

[0047] In the embodiment, the system polls the target mailbox 5 every 3 minutes to determine whether there is new mail.

[0048] The target mail is sent to the target mailbox by the client through the mobile banking of the bank providing the serial file.

[0049] Based on the preconfigured information, it is determined whether the new mail is the target mail. Specifically, according to the preconfigured serial acquisition table of each institution, the official mail address, mail format, serial file format, and whether there is a password of each bank are identified. The mail address, mail format, and serial file format are compared with the received mail to determine whether they are consistent. If they are consistent, the serial file of the attachment in the mail is automatically downloaded. For the encrypted serial file, the file key uploaded by the second user is used to decrypt and upload to the corresponding file directory (under the "transaction serial file class-bank transaction serial file" folder of the client). The serial files collected through the above-mentioned method do not need to be manually or automatically classified by the system.

[0050] The embodiment proposes a direct transmission and storage method of serial files. When the client manager instructs the client to request the serial file from the serial bank, the client can complete the collection of the serial file by providing the mailbox number of the target mailbox, without the client downloading and uploading to the system and a series of complex operation processes, and without the client manager collecting and uploading the system privately, further improving the efficiency of collecting data in the front end of credit approval. Moreover, the above-mentioned direct transmission and storage method can effectively avoid the problem of serial file fraud, and ensure the authenticity of the serial file.

[0051] The specific embodiments described herein are merely illustrative of the spirit of the present application. Those skilled in the art of the present application can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, without deviating from the spirit of the present application or exceeding the scope defined by the appended claims.

[0052] Although the terms server 1; credit information batch receiving module 11; large model calling module 12; prompt word module 13; classified storage module 14; mail identification module 15; key obtaining module 16; decryption module 17; mail extraction module 18; first client 2; second client 3; target large model 4; target mailbox 5 and the like are used more frequently herein, the possibility of using other terms is not excluded. The use of these terms is merely for the convenience of describing and explaining the essence of the present application; any kind of additional limitation by interpreting them is contrary to the spirit of the present application.

Claims

1. A method for batch uploading mixed credit data, characterized in that, The method comprises the following steps: receiving credit data submitted by a first user in bulk through an upload portal, the credit data comprising at least two types of credit files; calling a target large model; providing the credit data to the target large model and inputting a first prompt word to make the target large model return a main category of each credit file; the first prompt word comprising a plurality of predefined main categories and a feature description of each main category; receiving the main category of each credit file returned by the target large model; calling the target large model again; providing each credit file and its main category to the target large model and inputting a second prompt word to make the target large model return a subcategory of each credit file; the second prompt word comprising a predefined subcategory of each main category and a classification constraint of each subcategory; receiving the subcategory of each credit file returned by the target large model; storing each credit file in a corresponding file directory according to the main category and the subcategory.

2. The credit data bulk upload method of claim 1, wherein, storing each credit file in a corresponding file directory in a storage space assigned to a corresponding customer according to the subcategory classification result based on a customer selected by the first user; classifying and storing credit files uploaded by a second user in a corresponding file directory of a corresponding customer based on customer identity authentication information of the second user.

3. The credit data bulk upload method of claim 2, wherein, The method further comprises the following steps: extracting a log file of a target email from a target mailbox based on a preset rule; the target email being sent to the target mailbox by the second user instructing the log bank; identifying the target email and a corresponding customer based on preconfigured information; storing the log file in a corresponding file directory of the corresponding customer.

4. The credit data mixed batch uploading method according to claim 2, wherein, The method further comprises the following steps: receiving a file key uploaded by the second user; extracting an encrypted log file of a target email from a target mailbox based on a preset rule; the target email being sent to the target mailbox by the second user instructing the log bank; identifying the target email and a corresponding customer based on preconfigured information; decrypting the encrypted log file using the file key uploaded by the corresponding customer; storing the log file in a corresponding file directory of the corresponding customer.

5. The credit information mixed batch uploading method according to claim 3 or 4, characterized in that, The preset rule is to poll the target mailbox at a set time to find new emails and, when new emails are found, to determine whether the new emails are target emails based on preconfigured information; in response to the new emails being target emails, extracting a log file of the new emails and determining the corresponding customer of the target emails according to the content of the emails.

6. The credit data bulk upload method of claim 1, wherein, The first prompt word is used to instruct the target model to return the main category of each credit file according to the feature description of each main category; The second prompt word comprises a first dictionary and a second dictionary; The first dictionary comprises a classification constraint of each subcategory of each main category; The second dictionary comprises a predefined subcategory set of each main category; The second prompt word is used to instruct the target model to return the subcategory of each credit file according to the first dictionary and the second dictionary.

7. The credit data bulk upload method of claim 1, wherein, The target large model is obtained by performing two-stage fine-tuning training on a pre-trained large model; and the two-stage fine-tuning training method comprises the following steps: performing first-stage fine-tuning training on the large model using credit data and main classification labels; After the first-stage fine-tuning training, the second-stage fine-tuning training is performed on the large model using the credit data that has been distinguished according to the main categories and the sub-category labels of each credit file.

8. The credit data bulk upload method of claim 1, wherein, The predefined main categories include any one or more of the following: a certificate category, a real estate right category, an asset proof category, a liability category, a financial statement category, an operating contract category, an operating material category, an inquiry authorization category, a transaction log category, a work proof category, an invoice category, a tax return form, and an other category. Each main category is predefined with multiple sub-categories.

9. A server, characterized by The system comprises: a credit data batch receiving module configured to receive credit data submitted in batches by a first user through a unified uploading portal, wherein the credit data comprises at least two types of credit files; a large model calling module configured to call a target large model twice to obtain the sub-category of each credit file; a prompt word module configured to provide the credit data to the target large model and input a first prompt word to make the target large model return the main category of each credit file, and provide each credit file and its main category to the target large model and input a second prompt word to make the target large model return the sub-category of each credit file; the first prompt word comprises multiple predefined main categories and a feature description of each main category; the second prompt word comprises the predefined sub-categories of each main category and a classification constraint of each sub-category; a classification storage module configured to store each credit file in a corresponding file directory according to the main category and the sub-category thereof.

10. The server of claim 9, wherein, The system further comprises: a mail identification module configured to connect to a target mailbox and identify a target mail and a corresponding customer based on configuration information; the target mail is sent to the target mailbox by a second user through the flow bank; a key acquisition module configured to receive a file key uploaded by the second user; a decryption module configured to decrypt an encrypted transaction log file extracted from the target mail using the file key uploaded by the corresponding customer; a mail extraction module configured to extract the transaction log file of the target mail and store the unencrypted transaction log file or the decrypted transaction log file in a corresponding file directory of the corresponding customer.

11. A credit management system comprising a credit analysis module, characterized by The system further comprises the server of claim 9, and a first client configured to be used by the first user and having a unified uploading portal for the first user to submit credit data in batches; the credit analysis module is configured to perform credit risk analysis on a target customer based on the credit files in the file directories of the target customer.

12. The credit management system according to claim 11, characterized by, The system further comprises the server of claim 10, and a second client configured to be used by the second user and having multiple special uploading portals for different types of credit files; the second client further has a file key uploading portal for the second user to upload a file key.

Images