Loan product recommendation method and related equipment

Abstract

The invention discloses a loan product recommendation method and related equipment, and the method comprises the steps: calling a user input module to obtain the enterprise basic information of a loan enterprise, calling an industry and commerce information retrieval module to obtain the industry and commerce registration information of the loan enterprise, calling a vector knowledge base material retrieval module, and obtaining the loan qualification information of the loan enterprise. And calling the large model module to process the to-be-processed information, and outputting a loan product recommendation report. According to the method, matching between the information of the loan enterprise and the information of the loan product can be assisted, so that the appropriate loan product can be efficiently and accurately matched for the loan enterprise, and compared with a recommendation process which is completely manually executed by a worker of a financial institution, the method has higher execution efficiency and accuracy, and the user experience is improved. The standardization of the loan product recommendation process is improved, the workload of workers is reduced, and the loan time required to be waited by a loan enterprise is shortened, so that the financial efficiency is improved. The method is widely applied to the technical field of financial information.

Description

Technical Field

[0001] This invention relates to the field of financial information technology, and in particular to a method and related equipment for recommending loan products. Background Technology

[0002] Banks and other financial institutions provide loan services to corporate clients. Because different corporate clients vary in size, qualifications, industry, and financial status, financial institutions typically offer a variety of loan products to meet the diverse needs of different clients and provide precise financial services. Therefore, when financial institutions conduct their work, staff need to be familiar with the eligibility criteria of various loan products and understand the information of corporate clients. They then compare and match this information with the various loan products to determine if the institution has a suitable loan product for the client, and if multiple suitable products are available, which one is the most appropriate. This process involves complex and extensive information collection, organization, and comparison. In traditional financial institution operations, this process is performed manually, requiring a high level of experience from staff, resulting in a heavy workload, low efficiency, and a high risk of errors. From the perspective of corporate clients, the long approval time further reduces financial efficiency. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, the present invention aims to provide a loan product recommendation method and related equipment.

[0004] On one hand, embodiments of the present invention include a loan product recommendation method, which includes the following steps: Call the user input module to obtain the basic information of the loan company; The business registration information of the loan company is retrieved by calling the business registration information retrieval module. The vector knowledge base material retrieval module is invoked to obtain the loan qualification information of the loan enterprise; The large model module is invoked to process the information to be processed and output a loan product recommendation report; wherein, the information to be processed includes at least one of the enterprise basic information, the business registration information and the loan qualification information.

[0005] Furthermore, the step of calling the user input module to obtain the basic information of the loan company includes: The user input module displays the customer interaction interface; The customer interaction interface receives input information from the lending company. The input information is converted into a first prompt word; The first prompt word is input into the first large language model run by the user input module; Receive the structured output of the first large language model to obtain the basic information of the enterprise.

[0006] Furthermore, the step of calling the business registration information retrieval module to obtain the business registration information of the loan enterprise includes: Extract the company name information from the aforementioned basic company information; Convert the company name information into a second prompt word; The second prompt word is input into the second language model run by the business information retrieval module; Receive the structured output of the second language model to obtain the business registration information.

[0007] Furthermore, the vector knowledge base material retrieval module includes a loan product access submodule, a whitelist submodule, and an enterprise tag submodule; the step of calling the vector knowledge base material retrieval module to obtain the loan qualification information of the loan enterprise includes: Run the vector knowledge base; A knowledge space is created in the vector knowledge base; the knowledge space stores loan product knowledge, whitelist knowledge, and enterprise tag knowledge; Generate a third prompt word; the third prompt word includes keywords used to trigger a search for knowledge about the loan product; The third prompt word is input into the third language model running in the loan product access submodule, and the structured output of the third language model is received to obtain loan product access information; Generate a fourth prompt word; the fourth prompt word includes keywords used to trigger a search of the whitelist knowledge; The fourth prompt word is input into the fourth language model running in the whitelist submodule, and the structured output of the fourth language model is received to obtain whitelist information; Generate a fifth prompt word; the fifth prompt word includes keywords used to trigger a search for the enterprise's tag knowledge; The fifth prompt word is input into the fifth language model running in the enterprise tag submodule, and the structured output of the fifth language model is received to obtain enterprise tag information; The loan qualification information is composed of the loan product access information, the whitelist information, and the enterprise label information.

[0008] Furthermore, the step of calling the vector knowledge base material retrieval module to obtain the loan qualification information of the loan enterprise also includes: Retrieve file information uploaded by the backend department; The knowledge space is updated based on the document information.

[0009] Furthermore, the large model module runs a general-purpose financial large model and an inference-based financial large model; the large model module is invoked to process the information to be processed and output a loan product recommendation report, including: Based on the information to be processed, a sixth prompt word is generated; Based on the content of the information to be processed, a financial big model is determined; the financial big model is one of the general financial big model and the reasoning financial big model; The sixth prompt word is input into the financial big data model, the structured output of the financial big data model is received, and the loan product recommendation report is obtained.

[0010] Furthermore, determining the financial big data model based on the content of the information to be processed includes: When the loan qualification information is not included in the information to be processed, the general financial big model is selected as the financial big model. When the information to be processed includes the loan qualification information, the reasoning-based financial big model is selected as the financial big model.

[0011] Furthermore, the loan product recommendation method also includes: After obtaining the prompt word, and before processing the prompt word, the prompt word is anonymized and restricted before being sent to the lending company; wherein the prompt word is at least one of the first prompt word, the second prompt word, the third prompt word, the fourth prompt word, the fifth prompt word, and the sixth prompt word; Obtain the edited prompt returned by the loan company; Validate the returned prompt words; If the verification passes, the original prompt word is replaced with the returned prompt word.

[0012] On the other hand, embodiments of the present invention also include a computer device, including a memory and a processor, the memory for storing at least one program, and the processor for loading at least one program to execute the loan product recommendation method in the embodiments.

[0013] On the other hand, embodiments of the present invention also include a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the loan product recommendation method in the embodiments.

[0014] The beneficial effects of this invention are as follows: The loan product recommendation method in the embodiments can assist financial institution staff in matching the information of loan enterprises with the information of loan products legally sold by the financial institution, thereby efficiently and accurately matching suitable loan products for loan enterprises. This assists financial institution staff in recommending loan products to loan enterprises. Compared with the recommendation process that is entirely performed manually by financial institution staff, it has higher execution efficiency and accuracy, which is conducive to improving the standardization of the loan product recommendation process, reducing the workload of staff, and reducing the waiting time for loan enterprises, thereby improving financial efficiency. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a system that can implement a loan product recommendation method in the embodiment; Figure 2 This is a schematic diagram illustrating the steps of the loan product recommendation method in the embodiment; Figure 3 This is a flowchart illustrating the loan product recommendation method implemented in this embodiment. Figure 4 This is a schematic diagram of the customer interaction interface displayed on the front end of the user input module in the embodiment. Detailed Implementation

[0016] This embodiment provides a method for recommending loan products. (Refer to...) Figure 1 Modules such as user input, business information retrieval, vector knowledge base material retrieval, and large model modules can be set up on the servers of financial institutions. These modules can be hardware modules, software modules, or a combination of hardware and software.

[0017] pass Figure 1 The modules shown can execute loan product recommendation methods. Specifically, these methods can be implemented by financial institutions when a lending company wants to apply for a loan from a financial institution, or when a financial institution wants to market its loan business to a lending company. Each step in the loan product recommendation method can be executed by the financial institution's server.

[0018] Reference Figure 2 The method for recommending loan products includes the following steps: S1. Call the user input module to obtain the basic information of the loan company; S2. Call the business registration information retrieval module to obtain the business registration information of the loan company; S3. Call the vector knowledge base material retrieval module to obtain the loan qualification information of the loan company; S4. Call the large model module to process the information to be processed and output a loan product recommendation report.

[0019] In this embodiment, when executing the loan product recommendation method, the workflow of the user input module, business information retrieval module, vector knowledge base material retrieval module, and large model module is as follows: Figure 3 As shown.

[0020] Reference Figure 3 When executing step S1, which involves calling the user input module to obtain the basic information of the loan company, the following steps can be performed: S101. Display the customer interaction interface through the user input module; S102. Receive input information from loan companies through the customer interaction interface; S103. Convert the input information into the first prompt word; S104. Input the first prompt word into the first language model run by the user input module; S105. Receive the structured output of the first language model to obtain basic enterprise information.

[0021] In step S101, the user input module displays the following information on the front end: Figure 4 The customer interaction interface shown is accessible via the internet and can be sent to the lending company's client. In step S102, the lending company's staff inputs the following information into the input boxes on the customer interaction interface: company name, size category (large, medium, small, and micro), desired loan amount (unit: RMB 10,000), whether the company can provide a guarantee from its controlling shareholder or a co-borrower, number of other banks offering credit lines to the company, amount of credit lines from other banks (unit: RMB 10,000), and other qualifications the company possesses (e.g., independent intellectual property rights, the company owner or major shareholders receiving high-level talent titles awarded by national / local governments, undertaking major national / provincial / autonomous region / municipal science and technology projects, demonstration of the first (set) of major technical equipment, and technology transfer projects from renowned universities and research institutes). The customer interaction interface then uploads the basic company information input by the lending company to the server.

[0022] In step S103, the user input module converts the input information obtained in step S102 into a first prompt word. In this embodiment, an example of the content of the first prompt word is as follows:

【#Role You are a user input extractor for China Construction Bank's intelligent decision-making engine for small and micro loans. Known for its rigorous logic and adherence to standards, you extract only customer input and output it strictly according to the format after comprehensive analysis, without making any associations or expansions.

[0023] #Task You need to fill out the [Company Tag Form] based on the company information entered by the user. 1. {Company Name} 2. {The four categories of enterprise size} (The four categories of enterprise size) 3. {Expected Loan Amount (Unit: RMB 10,000)} (Expected Loan Amount) 4. {Can the company provide a guarantee from its actual controller or a co-borrower?} 5. {Enterprise's credit line from other banks (unit: RMB 10,000)} (Enterprise's credit line from other banks) 6. {Number of enterprises receiving credit from other banks} (Number of enterprises receiving credit from other banks) 7. Does the enterprise possess one of the following qualifications: [having independent intellectual property rights; the enterprise owner or major shareholder having received a high-level talent title awarded by the national / local government; undertaking a major national / provincial / autonomous region / municipal science and technology project; demonstrating the first (set) of major technical equipment; or a technology transfer project of a well-known university or research institute]? #Processing Input 1. For the parameters of [Company Name], [Unified Credit Code], and [Expected Loan Amount], **directly fill in the values ​​into the structured output**.

[0024] 2. For the parameter "Can the company provide a guarantor or co-borrower?", understand and analyze the user input, and fill in the result as "[Yes]" or "[No]" in the structured output.

[0025] 3. Regarding the parameter "Other Qualifications of the Enterprise," analyze the user input. If the user selects "Yes," it indicates that the enterprise possesses one of the following qualifications: [having independent intellectual property rights; the enterprise owner or major shareholder receiving a high-level talent title awarded by the national / local government; undertaking major national / provincial / autonomous region / municipal science and technology projects; demonstration of the first (set) of major technical equipment; or technology transfer projects from renowned universities and research institutes]. In this case, fill in "**The enterprise possesses other qualifications**" in the structured output. Otherwise, fill in "[No]" in the structured output for the corresponding parameter. # Structured Output Please use Markdown format, fill in the corresponding values ​​according to the following headings and points, and output: --- **Basic Corporate Label List** 1. [Company Name] (Company Name) 2. [The Four Categories of Enterprise Size] (The four categories of enterprise size) 3. [Expected Loan Amount (Unit: RMB 10,000)] 4. [Can the company provide a guarantee from its actual controller or a co-borrower?] 5. [Enterprise's Credit Line from Other Banks (Unit: RMB 10,000)] (Enterprise's Credit Line from Other Banks) 6. [Number of Enterprises Granted Credit by Other Banks] (Number of Enterprises Granted Credit by Other Banks) 7. Does the company possess one of the following qualifications: [having independent intellectual property rights; the company owner or major shareholder having received a high-level talent title awarded by the national / local government; undertaking a major national / provincial / autonomous region / municipal science and technology project; demonstrating the first (set) of major technical equipment; or a technology transfer project from a well-known university or research institute]? --- In step S104, the user input module inputs the first prompt word to the first large language model for processing. The first large language model is a trained large language model. Since the first prompt word contains instructions for structured output using Markdown format, the first large language model will process the first prompt word. In step S105, the output will include basic enterprise information in structured format, such as the enterprise name, the enterprise's four-category size, and the expected loan amount (unit: RMB 10,000).

[0026] Reference Figure 3 In this embodiment, when executing step S2, which is to call the business registration information retrieval module to obtain the business registration information of the loan company, the following steps can be performed: S201. Extract company name information from basic company information; S202. Convert company name information into a second prompt word; S203. Input the second prompt word into the second language model run by the business information retrieval module; S204. Receive the structured output of the second language model and obtain business registration information.

[0027] The principle behind execution steps S201-S204 is as follows: Since the basic enterprise information output by the user input module includes an enterprise profile, and the business information retrieval module happens to include the enterprise name, industry scope, industry category, major industry category, business scope, shareholder and investor information, key management personnel information, and external investment information, the enterprise profile can be obtained simply by matching the enterprise name, making the final output results appear more professional and objective.

[0028] In step S201, the company name information is extracted from the basic company information output by the user input module. Then, step S202 is executed to convert the company name information into a second prompt word. In this embodiment, an example of the content of the second prompt word is as follows:

【

Require

[0029] 2. Determine whether the company information and the company in the whitelist are completely identical to the provided company name: {company name}. If they are not completely identical (including similar ones), output "[This company does not exist, please verify]", terminate the judgment, and do not output any other content.

[0030] 3. If it is the same company, then compile and summarize the relevant information of the company [company information in the knowledge space (including company name, industry scope, company technology qualifications, whether it belongs to the whitelist of science and technology innovation platform loan companies), company business registration information (including company name, industry category, company technology qualifications, etc.), company shareholder and investor information, company key management personnel information, and company external investment information]. 4. Fill in the company information into the table according to the output format. [Company Information] 1. Company Information {Company Information} 1. Science and Technology Innovation Platform Loan - Enterprise Whitelist Tag {Science and Technology Innovation Platform Loan - Enterprise Whitelist Tag Information} 2. Enterprise registration information {Company Registration Information} 3. Information on the company's shareholders and investors {Information on Company Shareholders and Investors} 4. Information on key management personnel of the enterprise {Information on Key Management Personnel of the Enterprise} 5. Information on Enterprises' Overseas Investments {Information on Corporate Outbound Investment} [Structured Output] Output Format: Please use Markdown format, fill in the corresponding values ​​according to the following headings and points, and output: --- **Company Information** 1. **Basic Company Information** -Company Name - Valid or already obtained technology qualifications of the enterprise -Does the company have access to science and technology innovation platform loans?-Company whitelist label 2. **Enterprise Registration Information** - Industry category of the company -Enterprise Technology Qualification -Scope of business operations 3. **Information on Company Shareholders and Investors** -Information on corporate shareholders and investors 4. **Information on Key Management Personnel of the Enterprise** -Information on key management personnel of the enterprise 5. **Information on Enterprises' Overseas Investments** -Information on corporate overseas investment 】 In step S203, the business registration information retrieval module inputs the aforementioned second prompt word into the second large language model for processing. The second large language model is a trained large language model. Specifically, an independent large language model can be used as the second large language model, or the same large language model as the first large language model can be used as the second large language model. Since the second prompt word contains instructions for structured output using Markdown format, the second large language model will process the second prompt word. In step S204, it outputs structured business registration information containing the company name, the industry category name of the company, information on the company's shareholders and investors, and information on the company's key management personnel.

[0031] Reference Figure 3 In this embodiment, the vector knowledge base material retrieval module includes sub-modules such as a loan product access sub-module, a whitelist sub-module, and an enterprise tag sub-module. Based on these sub-modules, when executing step S3, which is to call the vector knowledge base material retrieval module to obtain the loan qualification information of the loan enterprise, the following steps can be performed: S301. Run the vector knowledge base; S302. Create a knowledge space in the vector knowledge base; the knowledge space stores loan product knowledge, whitelist knowledge, and enterprise tag knowledge; S303. Generate a third prompt word; S304. Input the third prompt word into the third language model running in the loan product access submodule, receive the structured output of the third language model, and obtain loan product access information; S305. Generate the fourth prompt word; S306. Input the fourth prompt word into the fourth language model running in the whitelist submodule, receive the structured output of the fourth language model, and obtain the whitelist information; S307. Generate the fifth prompt word; S308. Input the fifth prompt word into the fifth language model running in the enterprise tag submodule, receive the structured output of the fifth language model, and obtain enterprise tag information; S309. Loan qualification information is composed of loan product access information, whitelist information, and enterprise label information.

[0032] In steps S301-S302, the vector knowledge base material retrieval module runs and maintains a vector knowledge base. The vector knowledge base includes a knowledge space, which stores knowledge such as loan product knowledge, whitelist knowledge, and enterprise tag knowledge. Specifically, loan product knowledge is the search object of the loan product access submodule, whitelist knowledge is the search object of the whitelist submodule, and enterprise tag knowledge is the search object of the enterprise tag submodule.

[0033] Steps S303-S304 are the steps in the loan product access submodule that search for loan product knowledge. In step S303, the loan product access submodule can generate the following third-party suggestion keywords:

Task

[0034] 2. Compile the basic information of this loan product based on the input.

[0035] 3. Fill in the admission criteria into the table according to the output format.

Product Information

Detailed Application Process

Detailed Application Process

Detailed Application Process

Detailed Application Process

[0036] Steps S305-S306 are the steps taken by the whitelist submodule to search for whitelist knowledge. In step S305, the whitelist submodule can generate the following fourth suggestion word: [Search within the whitelist of companies lending through science and technology innovation platforms to see if {company name} has a whitelist tag.] The aforementioned fourth prompt word includes keywords used to trigger searches for whitelisted information such as the whitelist of science and technology innovation platform loan companies and whitelist tags. In step S306, the aforementioned fourth prompt word is input into the fourth major language model. Specifically, an independent major language model can be used as the fourth major language model, or the same major language model as the first, second, and third major language models can be used as the fourth major language model. The fourth major language model will process the fourth prompt word, and in step S304, the whitelist information will be output.

[0037] Steps S307-S308 are the steps in which the enterprise tag submodule searches for enterprise tag knowledge. In step S307, the enterprise tag submodule can generate the following fifth suggestion word: [Retrieve and recall enterprise tags for {company name}] The aforementioned fifth prompt word includes keywords used to trigger the enterprise tag submodule to search for enterprise tag knowledge, such as enterprise tags. In step S308, the aforementioned fifth prompt word is input into the fifth major language model. Specifically, an independent major language model can be used as the fifth major language model, or the same major language model as the first, second, third, and fourth major language models can be used as the fifth major language model. The fifth major language model will process the fifth prompt word, and in step S308, the enterprise tag information will be output.

[0038] In step S309, the loan product access information obtained by the loan product access submodule in step S304, the whitelist information obtained by the whitelist submodule in step S306, and the enterprise tag information obtained by the enterprise tag submodule in step S308 are combined to form the loan qualification information output by the vector knowledge base material retrieval module.

[0039] In this embodiment, the knowledge space in the vector knowledge base can be dynamically maintained. Specifically, since some selected loan products require enterprises to have technology enterprise qualifications, and the Science and Technology Innovation Platform Loan only allows whitelisted customers to access the platform, the vector knowledge base material retrieval module can periodically read the file information uploaded by the science and technology innovation department and other back-end departments. This file information may include a list of new-generation technology enterprises and whitelisted customers for the Science and Technology Innovation Platform Loan. In addition, since the access conditions for loan products may also change, fixing the access conditions would require taking the tool offline, modifying it, and then re-uploading it every time the access conditions for loan products are modified, which is cumbersome. Therefore, the vector knowledge base material retrieval module can directly read the latest file information from the science and technology innovation department's back-end and update the loan product knowledge, whitelist knowledge, and enterprise tag knowledge stored in the knowledge space of the vector knowledge base according to the file information. If the access conditions for loan products are updated later, the back-end department only needs to modify the file information and re-upload it. Therefore, by setting up a vector knowledge base and its knowledge space, the relative independence of the back-end department and the front-end business department can be achieved. The maintenance of the system is completed by the back-end department, allowing the staff of the front-end business department to focus on communication with loan companies and other aspects, thereby improving work efficiency.

[0040] Reference Figure 3 In this embodiment, the large model module can run two types of large models: a general-purpose financial large model and a reasoning-based financial large model. These two large models also belong to the category of large language models. Based on these large models, when executing step S4, which involves calling the large model module to process the information to be processed and outputting a loan product recommendation report, the following steps can be performed: S401. Generate the sixth prompt word based on the information to be processed; S402. Determine the large-scale financial model based on the content of the information to be processed; S403. Input the sixth prompt word into the financial big data model, receive the structured output of the financial big data model, and obtain a loan product recommendation report.

[0041] In step S401, the previously recalled information, which includes at least one of the following pending information: basic enterprise information, business registration information, and loan qualification information, is used to generate a sixth prompt word. In this embodiment, the sixth prompt word can be referred to as the final recommendation decision tree prompt word. An example of the content of the sixth prompt word is as follows:

【#Role You are a smart decision-making engine for small and micro loans at China Construction Bank, known for your rigorous logic and adherence to standards.

Task

[0042] 2. Technological qualifications only include those for specialized, refined, and innovative SMEs / specialized, refined, and innovative "little giants" / high-tech enterprises / technology-based SMEs / innovative SMEs / manufacturing single champions / national technology innovation demonstration enterprises. 3. Finally, output according to the output format requirements, and do not output any other irrelevant content.

[0043] [Decision-making information] 1. [Company-related information]: {Company-related information} 2. [Company Qualification Information]: {Company Qualification Information} 3. [Loan Product Information]: {Loan Product Information} [Decision Logic Tree] 1. **Existence Check:** Determines if the company exists. If the company does not exist, output "[Unable to find information for this company, please verify]" and terminate the check. If the company exists, then proceed with condition matching and decision-making. 2. **Company Size Assessment:** This function assesses the company's size across four categories. If the company is classified as **Medium** or **Large**, it directly outputs "[This assistant only recommends products to small and micro-sized technology companies]" and terminates the assessment. If the company is classified as **Micro** or **Small**, then condition matching and decision-making are performed. 3. **Compile the eligibility criteria for loan products**: Compile the eligibility criteria for the four loan products: [Growth Path, Good New Loan, Good Science Loan, and Science and Technology Innovation Platform Loan] [maximum loan amount / maximum credit limit, whether a guarantee from the actual controller of the enterprise is required (including other requirements for similar guarantees, such as co-borrowing), requirements for credit balance with other banks, whether technology qualifications are required, whether there are requirements for registered capital, and other eligibility criteria that need to be met].

[0044] 4. *Compile basic information about the loan product: Compile and summarize the basic information of this loan product [interest rate, application method, application process, product features and advantages].

[0045] 3. **Condition Matching**: Matches loan products to eligibility criteria based on [company-related information] and [company qualification information].

[0046] 4. **Admission Criteria Matching**: -Analyze the eligibility criteria and relevant information and qualifications of the companies involved in the four loan products: Growth Path, Good New Loan, Good Science Loan, and Science and Technology Innovation Platform Loan. - Determine if the business loan amount exceeds the maximum loan amount of the loan product; loan products exceeding the limit will not be recommended. - Determine if the company's qualifications meet the eligibility criteria for loan products; loan products that do not meet the criteria will not be recommended. -Based on a comprehensive analysis of the company's loan amount and creditworthiness, all eligible loan products will be recommended and displayed. 5. **Product Comparison**: (If multiple products are recommended, a comparison, conclusion, and recommendations between the recommended loan products will be displayed after each product recommendation; if only one product is recommended, this module will display the corresponding introduction of that product.) (- **Interest Rate Comparison**) What are the interest rates for the recommended loan products? in conclusion: - **Comparison of Processing Procedures**: What are the recommended application methods and procedures for loan products? Do the recommended loan products require a guarantee from the company's actual controller (including other requirements similar to guarantees, such as joint borrowing)? What are the detailed application requirements for the recommended loan products? What are the detailed application procedures for the recommended loan products? What is the applicable scope of the recommended loan products? Which one is simpler? in conclusion: - **Other differences:** in conclusion: )

Application Guide

[0047] After completing step S402 to determine the financial big data model to be used, in step S403, the big data model module inputs the sixth prompt word into the financial big data model, receives the structured output of the financial big data model, and obtains a loan product recommendation report. The loan product recommendation report mainly includes company introduction, loan product recommendations, and loan product comparisons. Specifically, the loan product recommendations include the loan product name, introduction, reasons for recommendation, application guidelines, and reasons for not recommending other loan products. The loan product comparisons include interest rate comparisons, application process comparisons, and other differences.

[0048] In this embodiment, by using a computer to execute the loan product recommendation method, financial institution staff can be assisted in matching the information of loan companies with the information of loan products legally sold by the financial institution. This allows for efficient and accurate matching of suitable loan products to loan companies, thus assisting financial institution staff in recommending loan products to them. Compared to a recommendation process entirely performed manually by financial institution staff, this method has higher execution efficiency and accuracy. It also helps to improve the standardization of the loan product recommendation process, reduce the workload of staff, and shorten the waiting time for loan companies, thereby improving financial efficiency.

[0049] Before developing this workflow—the loan product recommendation method—financial institution staff (such as frontline account managers) needed to search for eligibility criteria in multiple technology loan product documents within the bank, match the eligibility criteria of each loan product according to the customer's situation, and manually compare the criteria among suitable loan products to determine which product best suited the customer's needs. Furthermore, the workload of frontline account managers was quite complex, and they might not be able to keep track of changes in loan product eligibility criteria or whether the customer still met the criteria after the changes, requiring manual verification. By implementing this loan product recommendation method, a workflow integrating intelligent recommendation, multi-dimensional analysis, and precise matching capabilities was achieved. By deeply integrating intelligent algorithms into the customer service process, inputting the qualification characteristics and financing needs of technology-based micro and small enterprises, and performing multi-dimensional intelligent analysis, the entire process of "enterprise profiling - product screening - solution adaptation" is automated. A precise recommendation report for technology loan products is generated with one click, and the latest loan document details are accessible. While ensuring business compliance, this significantly reduces the workload of frontline account managers in searching and matching eligibility criteria across products and conducting cross-product comparison analysis, effectively releasing frontline productivity and fully demonstrating the practical value of financial digital transformation.

[0050] In this embodiment, the prompt words can be processed immediately after they are obtained. For example, after obtaining the first prompt word in step S103, step S104 is immediately executed to input the first prompt word into the first large language model for processing; after obtaining the second prompt word in step S202, step S203 is immediately executed to input the second prompt word into the second large language model for processing; after obtaining the third, fourth, and fifth prompt words in steps S303, S305, and S307, steps S304, S306, and S308 are immediately executed to input these prompt words into the third, fourth, and fifth large language models for processing; after obtaining the sixth prompt word in step S401, step S403 is immediately executed to input the sixth prompt word into the financial large model for processing.

[0051] In this embodiment, after obtaining the prompt words, but before inputting them into the corresponding large language model for processing, the prompt words can be anonymized and have their editing restricted before being sent to the lending company. For example, taking the first prompt word as an example, the anonymization process includes masking specific numerical values ​​and file source information in the first prompt word, while the editing restriction process includes setting some content in the first prompt word to be viewable only and not editable, while setting other content to be editable. When the lending company receives the first prompt word after the anonymization and editing restriction process, the lending company cannot see specific numerical values ​​and file source information in the first prompt word, but can see the general framework of the first prompt word. The lending company can choose to edit the first prompt word, but due to the editing restriction process, the lending company can only edit the parts that the financial institution allows to be edited.

[0052] After the lending company edits the received first prompt (or confirms the content directly without editing), it returns the edited first prompt to the financial institution's server. The financial institution's server verifies the first prompt returned by the lending company, checking the completeness of the non-editable parts and the compliance of the editable parts. If the verification passes, for example, if the non-editable parts of a prompt are complete and the editable parts are compliant, the financial institution's server replaces the original first prompt (the one sent by the financial institution to the lending company) with the first prompt returned by the lending company. In step S104, the first prompt returned by the lending company is input into the first language model for processing.

[0053] In this embodiment, after obtaining the prompt words, sending them to the lending company for review and editing before returning them, and processing the returned prompt words, allows the lending company to understand the progress of the loan processing and participate in the loan product recommendation process. For example, the lending company can edit the prompt words based on its latest information, enabling the loan product recommendation process to track changes in the lending company's situation in real time, thus ensuring that the final recommended loan product is better suited to the lending company's latest circumstances. Furthermore, by desensitizing and restricting the editing of the prompt words before sending them, trade secrets and legally required confidential information during the loan product recommendation process can be protected from disclosure, and the arbitrariness of prompt word editing can be reduced, thereby ensuring financial security.

[0054] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "upper," "lower," "left," and "right" used in this disclosure are only relative to the relative positional relationships of the components of this disclosure in the accompanying drawings. The singular forms "a," "an," and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. Moreover, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this embodiment specification is only for describing particular embodiments and is not intended to limit the invention. The term "and / or" as used in this embodiment includes any combination of one or more of the associated listed items.

[0055] It should be understood that although the terms first, second, third, etc., may be used to describe various elements in this disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. The use of any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided in this embodiment is intended only to better illustrate embodiments of the invention and, unless otherwise required, does not impose a limitation on the scope of the invention.

[0056] It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including a non-transitory computer-readable storage medium configured with a computer program, wherein such a storage medium causes the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0057] Furthermore, the procedures described in this embodiment can be performed in any suitable order unless otherwise indicated by this embodiment or otherwise obviously contradict the context. The procedures (or variations and / or combinations thereof) described in this embodiment can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. A computer program includes a plurality of instructions executable by one or more processors.

[0058] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention of this embodiment includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques of the invention, the invention also includes the computer itself.

[0059] A computer program can be applied to input data to perform the functions of this embodiment, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including a specific visual depiction of physical and tangible objects generated on the display.

[0060] The above are merely preferred embodiments of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. A method for recommending loan products, characterized in that, The loan product recommendation methods include: Call the user input module to obtain the basic information of the loan company; The business registration information of the loan company is retrieved by calling the business registration information retrieval module. The vector knowledge base material retrieval module is invoked to obtain the loan qualification information of the loan enterprise; The large model module is invoked to process the information to be processed and output a loan product recommendation report; wherein, the information to be processed includes at least one of the enterprise basic information, the business registration information and the loan qualification information.

2. The loan product recommendation method according to claim 1, characterized in that, The user input module is invoked to obtain the basic information of the loan company, including: The user input module displays the customer interaction interface; The customer interaction interface receives input information from the lending company. The input information is converted into a first prompt word; The first prompt word is input into the first large language model run by the user input module; Receive the structured output of the first large language model to obtain the basic information of the enterprise.

3. The loan product recommendation method according to claim 1, characterized in that, The step of calling the business registration information retrieval module to obtain the business registration information of the loan enterprise includes: Extract the company name information from the aforementioned basic company information; Convert the company name information into a second prompt word; The second prompt word is input into the second language model run by the business information retrieval module; Receive the structured output of the second language model to obtain the business registration information.

4. The loan product recommendation method according to claim 1, characterized in that, The vector knowledge base material retrieval module includes a loan product access submodule, a whitelist submodule, and an enterprise tag submodule; the step of calling the vector knowledge base material retrieval module to obtain the loan qualification information of the loan enterprise includes: Run the vector knowledge base; A knowledge space is created in the vector knowledge base; the knowledge space stores loan product knowledge, whitelist knowledge, and enterprise tag knowledge; Generate a third prompt word; the third prompt word includes keywords used to trigger a search for knowledge about the loan product; The third prompt word is input into the third language model running in the loan product access submodule, and the structured output of the third language model is received to obtain loan product access information; Generate a fourth prompt word; the fourth prompt word includes keywords used to trigger a search of the whitelist knowledge; The fourth prompt word is input into the fourth language model running in the whitelist submodule, and the structured output of the fourth language model is received to obtain whitelist information; Generate a fifth prompt word; the fifth prompt word includes keywords used to trigger a search for the enterprise's tag knowledge; The fifth prompt word is input into the fifth language model running in the enterprise tag submodule, and the structured output of the fifth language model is received to obtain enterprise tag information; The loan qualification information is composed of the loan product access information, the whitelist information, and the enterprise label information.

5. The loan product recommendation method according to claim 4, characterized in that, The method of calling the vector knowledge base material retrieval module to obtain the loan qualification information of the loan enterprise also includes: Retrieve file information uploaded by the backend department; The knowledge space is updated based on the document information.

6. The loan product recommendation method according to claim 1, characterized in that, The large model module runs a general-purpose financial large model and an inference-type financial large model; The large model module is invoked to process the information to be processed and output a loan product recommendation report, including: Based on the information to be processed, a sixth prompt word is generated; Based on the content of the information to be processed, a financial big model is determined; the financial big model is one of the general financial big model and the reasoning financial big model; The sixth prompt word is input into the financial big data model, the structured output of the financial big data model is received, and the loan product recommendation report is obtained.

7. The loan product recommendation method according to claim 6, characterized in that, The step of determining the financial big data model based on the content of the information to be processed includes: When the loan qualification information is not included in the information to be processed, the general financial big model is selected as the financial big model. When the information to be processed includes the loan qualification information, the reasoning-based financial big model is selected as the financial big model.

8. The loan product recommendation method according to any one of claims 1-7, characterized in that, The loan product recommendation method also includes: After obtaining the prompt word, and before processing the prompt word, the prompt word is anonymized and restricted before being sent to the lending company; wherein the prompt word is at least one of the first prompt word, the second prompt word, the third prompt word, the fourth prompt word, the fifth prompt word, and the sixth prompt word; Obtain the edited prompt returned by the loan company; Validate the returned prompt words; If the verification passes, the original prompt word is replaced with the returned prompt word.

9. A computer device, characterized in that, It includes a memory and a processor, the memory being used to store at least one program, and the processor being used to load at least one program to perform the loan product recommendation method according to any one of claims 1-8.

10. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the loan product recommendation method according to any one of claims 1-8.

Images