Enterprise service matching method based on large language model

Abstract

The application discloses an enterprise service matching method based on a large language model, and particularly relates to the technical field of enterprise service management, and comprises the following steps: collecting basic information, operation information, financial information, technical information, historical service information and service demand information input in natural language of an enterprise. The application utilizes a large language model to understand multi-source information of the enterprise and natural language demand, realizes automatic classification, condition identification and accurate matching of services such as project declaration, scientific and technological services and accounting agency, comprehensively scores through combination of semantic similarity, qualification conditions, service capacity, timeliness, region, budget and historical evaluation, improves the accuracy, reliability and applicability of the recommended result, enhances the explainability of the result through output of matching basis, supplementary materials and risk prompts, continuously optimizes the matching model through feedback information, and improves data security and business traceability in combination of permission control, data desensitization and log recording.

Description

Technical Field

[0001] This invention relates to the field of enterprise service management technology, and more specifically, to an enterprise service matching method based on a large language model. Background Technology

[0002] When businesses engage in project applications, technology services, or bookkeeping, they typically need to select suitable services and service providers based on their size, industry, operating status, technological achievements, financial situation, and specific needs. Currently, service matching for businesses mainly relies on manual consultation, keyword searches, or fixed rule filtering. Because business information is scattered across business registration, finance, project, technological achievement, and historical service records, and because business needs are usually described in natural language, existing methods struggle to accurately understand the actual needs of businesses and handle synonyms, implicit conditions, and combinations of multiple services.

[0003] In addition, project applications involve application conditions and material requirements, science and technology services involve technical fields, R&D needs and service capabilities, and bookkeeping services involve enterprise size, number of invoices, financial and tax status and filing cycle. The matching conditions for different service types vary greatly. Using a single keyword or static rules for matching can easily lead to problems such as inaccurate recommendation results, opaque matching basis, and failure to update in a timely manner after changes in service conditions.

[0004] Therefore, there is an urgent need for an enterprise service matching method based on a large language model to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, embodiments of the present invention provide an enterprise service matching method based on a large language model. This invention utilizes a large language model to understand multi-source information and natural language needs of enterprises, enabling automatic classification, conditional identification, and accurate matching of services such as project applications, technology services, and bookkeeping. By combining semantic similarity, qualification conditions, service capabilities, timeliness, region, budget, and historical evaluations for comprehensive scoring, the accuracy, reliability, and applicability of the recommendation results are improved. The interpretability of the results is enhanced by outputting matching criteria, supplementary materials, and risk warnings. The matching model is continuously optimized through feedback information, and data security and business traceability are improved by combining access control, data anonymization, and log recording, thereby solving the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an enterprise service matching method based on a large language model, comprising the following steps: S1. Collect basic information, operational information, financial information, technical information, historical service information, and service request information input by the enterprise in natural language. S2. Perform data cleaning, format unification, sensitive information processing, entity recognition, and semantic segmentation on the enterprise information and service demand information to extract enterprise characteristics, demand characteristics, constraints, service period, and expected goals; S3. Based on the preset enterprise service classification system, the service requirements are divided into one or more of project application services, technology services, and bookkeeping services, and the corresponding demand intent and service conditions are generated through a large language model. S4. Establish a service resource database, which includes service items, application conditions, service content, professional capabilities of service organizations or personnel, service areas, service prices, service periods, required materials, historical service evaluations, and availability status. S5. Using a large language model, perform semantic vectorization or semantic association analysis on the structured representation of the enterprise characteristics, demand intentions, service conditions, and service resources in the service resource library to obtain a set of candidate service resources. S6. Perform condition filtering and comprehensive scoring on the candidate service resource set. The comprehensive score is determined at least based on enterprise adaptability, service capability matching degree, condition satisfaction degree, timeliness matching degree, regional matching degree and historical service effect. S7. Sort the candidate service resources according to the comprehensive score and generate matching results including recommended services, matching basis, required supplementary materials, applicable conditions, risk warnings and expected service results; S8. Receive confirmation, modification, rejection or evaluation information from enterprises or business personnel regarding the matching results, and use the feedback information to update the demand characteristics, service resource weights or matching model parameters; S9. Implement access control for enterprise information, service demand information, and service resource information; control the scope of data access based on user roles, affiliated organizations, responsible customers, and service categories; perform anonymization or encryption on identity verification information, bank account information, tax information, and financial data; and log the input, output, call time, caller, and matching results of the large language model.

[0007] In a preferred embodiment, the enterprise characteristics include the enterprise's industry, years of establishment, registered capital, operating status, number of employees, operating revenue, R&D investment, number of intellectual property rights, project experience, scale of financial processing, number of invoices, tax payment status, and service budget; the demand characteristics include service category, service objectives, expected completion time, service area, budget range, material preparation status, and special requirements.

[0008] In a preferred embodiment, generating a structured representation of demand intent and service conditions through a large language model includes inputting natural language service demands into the large language model; having the large language model identify service objects, service actions, service objectives, time requirements, necessary conditions, exclusion conditions, and uncertain conditions; outputting structured demand data according to preset field templates; performing field integrity checks, format checks, and confidence assessments on the structured demand data; and generating supplementary queries for missing fields when key fields are missing or the confidence level is below a preset threshold.

[0009] In a preferred embodiment, the service resources in the service resource library are described by unified service tags, which include one or more of the following: service category tags, industry tags, technology field tags, enterprise size tags, region tags, qualification tags, material tags, timeliness tags, and capability tags; and the service descriptions, historical cases, and business capability descriptions of service organizations or service personnel are converted into the unified service tags and corresponding semantic vectors through a large language model.

[0010] In a preferred embodiment, for project application services, the condition filtering and comprehensive scoring are based at least on the enterprise's industry, years of operation, intellectual property status, R&D investment, project foundation, application region, application time, completeness of application materials, and fulfillment of application conditions; for technology services, the criteria are based at least on the enterprise's technical field, technical needs, R&D stage, technological achievements, professional capabilities of service personnel, service cases, and delivery cycle; for bookkeeping services, the criteria are based at least on the enterprise's size, number of invoices, complexity of accounting, tax type, completeness of financial data, service cycle, and financial and tax service capabilities.

[0011] In a preferred embodiment, the comprehensive score of the candidate service resources is determined as follows: semantic similarity is calculated between enterprise characteristics and service resource characteristics to obtain a semantic matching score; the necessary conditions in the enterprise's needs and the service conditions in the service resources are compared item by item to obtain a condition satisfaction score; a business constraint score is obtained based on the service period, service area, budget range, and service resource availability; a service quality score is obtained based on historical service evaluations, service completion rate, customer feedback, and anomaly records; and the semantic matching score, condition satisfaction score, business constraint score, and service quality score are weighted according to preset weights to obtain a comprehensive score.

[0012] In a preferred embodiment, before generating matching results, interpretability analysis is performed on candidate service resources to determine key enterprise characteristics, demand conditions, and service resource characteristics that affect the comprehensive score, and the key characteristics are converted into matching criteria in natural language form; when enterprise information is incomplete, service conditions are uncertain, or candidate service resources do not meet the necessary conditions, corresponding risk warnings are generated in the matching results.

[0013] In a preferred embodiment, when an enterprise has simultaneous needs for project application, technology services, and bookkeeping services, the method generates a combined service plan based on the priority relationship, material dependency relationship, service period, and service resource capacity among the multiple needs, and determines the execution order, associated materials, responsible parties, and phased output results for each service item.

[0014] In a preferred embodiment, the feedback information includes at least the adoption status of recommendation results, service contract status, service completion status, customer evaluation, service anomalies, and actual service effects; and the enterprise feature weights, service resource quality scores, service category matching rules, and candidate resource ranking weights are adjusted based on the feedback information.

[0015] The technical effects and advantages of this invention are as follows: This invention utilizes a large language model to understand the natural language needs of enterprises and multi-source business information, enabling intelligent classification, condition recognition, and accurate matching of services such as project applications, technology services, and bookkeeping. It improves the accuracy and applicability of matching results by combining semantic analysis with structured condition verification; enhances the reliability of recommendation results through comprehensive scoring based on factors such as service capabilities, timeliness, region, budget, and historical evaluations; strengthens the interpretability of results through matching criteria and risk warnings; continuously optimizes the matching model through feedback information, improving the system's adaptability to different enterprises and service scenarios; and enhances the security and traceability of enterprise operational and financial data through access control, data anonymization, and log recording. Detailed Implementation

[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] This invention provides an enterprise service matching method based on a large language model, comprising the following steps: S1. Collect basic information, operational information, financial information, technical information, historical service information, and service request information input by the enterprise in natural language. S2. Perform data cleaning, format unification, sensitive information processing, entity recognition, and semantic segmentation on the enterprise information and service demand information to extract enterprise characteristics, demand characteristics, constraints, service period, and expected goals; S3. Based on the preset enterprise service classification system, the service requirements are divided into one or more of project application services, technology services, and bookkeeping services, and the corresponding demand intent and service conditions are generated through a large language model. S4. Establish a service resource database, which includes service items, application conditions, service content, professional capabilities of service organizations or personnel, service areas, service prices, service periods, required materials, historical service evaluations, and availability status. S5. Using a large language model, perform semantic vectorization or semantic association analysis on the structured representation of the enterprise characteristics, demand intentions, service conditions, and service resources in the service resource library to obtain a set of candidate service resources. S6. Perform condition filtering and comprehensive scoring on the candidate service resource set. The comprehensive score is determined at least based on enterprise adaptability, service capability matching degree, condition satisfaction degree, timeliness matching degree, regional matching degree and historical service effect. S7. Sort the candidate service resources according to the comprehensive score and generate matching results including recommended services, matching basis, required supplementary materials, applicable conditions, risk warnings and expected service results; S8. Receive confirmation, modification, rejection or evaluation information from enterprises or business personnel regarding the matching results, and use the feedback information to update the demand characteristics, service resource weights or matching model parameters; S9. Implement access control for enterprise information, service demand information, and service resource information; control the scope of data access based on user roles, affiliated organizations, responsible customers, and service categories; perform anonymization or encryption on identity verification information, bank account information, tax information, and financial data; and log the input, output, call time, caller, and matching results of the large language model.

[0018] The enterprise characteristics include the industry to which the enterprise belongs, the number of years since its establishment, the registered capital, the operating status, the size of its workforce, its operating revenue, its R&D investment, the number of intellectual property rights, its project experience, the scale of its financial processing, the number of invoices, its tax payment status, and its service budget; the demand characteristics include the service category, service objectives, expected completion time, service area, budget range, material preparation status, and special requirements.

[0019] The process of generating a structured representation of demand intent and service conditions through a large language model includes: inputting natural language service demands into the large language model; having the large language model identify service objects, service actions, service goals, time requirements, necessary conditions, exclusion conditions, and uncertain conditions; outputting structured demand data according to preset field templates; performing field integrity checks, format checks, and confidence assessments on the structured demand data; and generating supplementary queries for missing fields when key fields are missing or the confidence level is below a preset threshold.

[0020] The service resources in the service resource library are described using unified service tags, which include one or more of the following: service category tags, industry tags, technology field tags, enterprise size tags, region tags, qualification tags, material tags, timeliness tags, and capability tags. The service descriptions, historical cases, and business capability descriptions of service organizations or service personnel are converted into the unified service tags and corresponding semantic vectors through a large language model.

[0021] For project application services, the filtering and comprehensive scoring should be based at least on the enterprise's industry, years of operation, intellectual property status, R&D investment, project foundation, application region, application time, completeness of application materials, and fulfillment of application conditions. For technology services, the criteria should be based at least on the enterprise's technology field, technology needs, R&D stage, technological achievements, professional capabilities of service personnel, service cases, and delivery cycle. For bookkeeping services, the criteria should be based at least on the enterprise's size, number of invoices, complexity of accounting, tax type, completeness of financial data, service cycle, and financial and tax service capabilities.

[0022] The comprehensive score of the candidate service resources is determined as follows: Semantic similarity is calculated between enterprise characteristics and service resource characteristics to obtain a semantic matching score; each necessary condition in the enterprise's needs and service condition in the service resources is compared to obtain a condition satisfaction score; a business constraint score is obtained based on service period, service area, budget range, and service resource availability; a service quality score is obtained based on historical service evaluations, service completion rate, customer feedback, and anomaly records; and the semantic matching score, condition satisfaction score, business constraint score, and service quality score are weighted according to preset weights to obtain a comprehensive score.

[0023] Before generating matching results, interpretability analysis is performed on candidate service resources to determine key enterprise characteristics, demand conditions, and service resource characteristics that affect the comprehensive score, and these key characteristics are converted into matching criteria in natural language form. When enterprise information is incomplete, service conditions are uncertain, or candidate service resources do not meet the necessary conditions, corresponding risk warnings are generated in the matching results.

[0024] When an enterprise has simultaneous needs for project application, technology services, and bookkeeping services, the method generates a combined service plan based on the priority relationship between multiple needs, material dependencies, service periods, and service resource capacity, and determines the execution order, related materials, responsible parties, and phased output results for each service item.

[0025] The feedback information includes at least the adoption status of recommendation results, service contract status, service completion status, customer evaluation, service anomalies, and actual service effects; the enterprise feature weights, service resource quality scores, service category matching rules, and candidate resource ranking weights are adjusted based on the feedback information.

[0026] This invention allows clients to input service requests. The system collects information such as the enterprise's unified social credit code, industry, establishment date, staff size, R&D investment, intellectual property, financial scale, and historical service records. The system then segments and identifies entities within the input. The request understanding module inputs these natural language requests into a large language model. The large language model identifies three service intentions: technology services, project application services, and bookkeeping services. Technology service requests include automated detection technology, R&D support, and technical service objectives; project application requests include applicable project selection, application time, and enterprise qualification verification; and bookkeeping requests include the number of invoices per month, financial processing scale, and continuous service period. The system converts the identification results into unified structured data and generates supplementary queries for missing information, such as the enterprise's location, R&D stage, current taxpayer type, and expected service budget.

[0027] The service resource library pre-stores project application items, technology service items, and accounting service resources. For project application items, service resources include project name, applicable industry, application area, application conditions, application deadline, required materials, and past approval status. For technology service resources, service resources include technical field, service content, service personnel's professional specialization, past cases, delivery cycle, and service price. For accounting service resources, service resources include types of enterprises that can be served, invoice processing capabilities, accounting personnel configuration, service area, fee standards, tax declaration capabilities, and customer reviews.

[0028] The candidate recall module first performs preliminary filtering based on service category and company location, then uses a large language model to conduct semantic association analysis on the company's needs and service resource descriptions. For example, it identifies different expressions such as "automated inspection technology requires R&D support" and "industrial vision inspection technology R&D, inspection equipment algorithm optimization, R&D solution design" as having similar service intentions, thereby recalling relevant technology service resources. For project application and bookkeeping services, the system performs necessary condition checks based on the application conditions and financial and tax service conditions, excluding service resources that do not meet key conditions or are currently unavailable.

[0029] The matching and scoring module comprehensively evaluates candidate service resources. Taking project application services as an example, the system compares the degree of matching between the company's establishment year, industry category, R&D investment, intellectual property rights, and application region with the project application requirements; taking technology services as an example, the system compares the degree of matching between the company's technical needs and the professional capabilities of service personnel, historical cases, and delivery cycle; taking bookkeeping services as an example, the system compares the degree of matching between the company's number of invoices, tax types, and financial processing complexity and the service agency's capacity. The system also combines service price, service period, historical evaluations, and service completion rate to generate the final ranking results.

[0030] The results interpretation module outputs a list of recommended services and provides matching criteria for each recommendation. For example, it might explain that a project application matches the company's industry, R&D investment, and intellectual property conditions, but still lacks the previous year's financial statements; it might explain that a technology service provider has relevant technical case studies and the expected delivery cycle meets the company's requirements; or it might explain that an accounting agency has the capacity to process more than 300 invoices per month, but the company needs to supplement its taxpayer type and historical accounting information.

[0031] For combined needs, the system links project application, technology services, and accounting services. For example, the accounting service provider first completes the financial statements and tax information, then the technology service personnel supplement the technical solutions and R&D supporting materials, and finally the project application service personnel organize and submit the application materials according to the application guidelines. The system generates independent service item identifiers for each service and establishes the relationship between enterprise identifiers, service item identifiers, material identifiers, and business status.

[0032] Once a company adopts the recommended results and completes the service, the system collects feedback data such as contract signing status, service completion status, customer reviews, actual delivery cycle, and service exceptions. The feedback learning module adjusts the service resource quality score and matching weight based on this data. For example, if a service provider has a high semantic matching degree but repeatedly experiences delayed delivery, its timeliness matching weight will be reduced; if a technology service provider has high customer satisfaction in a specific technical field, its recommendation priority in that field will be increased.

[0033] Regarding data security, the system restricts the scope of enterprise information access based on user roles and business permissions. Project applicants can only access enterprise information related to the application process, technology service personnel can only access information necessary to complete technical services, and accounting agents can only access authorized financial and invoice information. For sensitive data such as identity verification, bank accounts, tax information, and financial amounts, the system performs anonymization and encrypted transmission, and records operations such as model calls, data queries, recommendation generation, result confirmation, and feedback evaluation to ensure the traceability of the matching process.

[0034] This invention utilizes a large language model to understand the natural language needs of enterprises and multi-source business information, enabling intelligent classification, condition recognition, and accurate matching of services such as project applications, technology services, and bookkeeping. It improves the accuracy and applicability of matching results by combining semantic analysis with structured condition verification; enhances the reliability of recommendation results through comprehensive scoring based on factors such as service capabilities, timeliness, region, budget, and historical evaluations; strengthens the interpretability of results through matching criteria and risk warnings; continuously optimizes the matching model through feedback information, improving the system's adaptability to different enterprises and service scenarios; and enhances the security and traceability of enterprise operational and financial data through access control, data anonymization, and log recording.

[0035] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for matching enterprise services based on a large language model, characterized in that, Includes the following steps: S1. Collect basic information, operational information, financial information, technical information, historical service information, and service request information input by the enterprise in natural language. S2. Perform data cleaning, format unification, sensitive information processing, entity recognition, and semantic segmentation on the enterprise information and service demand information to extract enterprise characteristics, demand characteristics, constraints, service period, and expected goals; S3. Based on the preset enterprise service classification system, the service requirements are divided into one or more of project application services, technology services, and bookkeeping services, and the corresponding demand intent and service conditions are generated through a large language model. S4. Establish a service resource database, which includes service items, application conditions, service content, professional capabilities of service organizations or personnel, service areas, service prices, service periods, required materials, historical service evaluations, and availability status. S5. Using a large language model, perform semantic vectorization or semantic association analysis on the structured representation of the enterprise characteristics, demand intentions, service conditions, and service resources in the service resource library to obtain a set of candidate service resources. S6. Perform condition filtering and comprehensive scoring on the candidate service resource set. The comprehensive score is determined at least based on enterprise adaptability, service capability matching degree, condition satisfaction degree, timeliness matching degree, regional matching degree and historical service effect. S7. Sort the candidate service resources according to the comprehensive score and generate matching results including recommended services, matching basis, required supplementary materials, applicable conditions, risk warnings and expected service results; S8. Receive confirmation, modification, rejection or evaluation information from enterprises or business personnel regarding the matching results, and use the feedback information to update the demand characteristics, service resource weights or matching model parameters; S9. Implement access control for enterprise information, service demand information, and service resource information; control the scope of data access based on user roles, affiliated organizations, responsible customers, and service categories; perform anonymization or encryption on identity verification information, bank account information, tax information, and financial data; and log the input, output, call time, caller, and matching results of the large language model.

2. The enterprise service matching method based on a large language model according to claim 1, characterized in that: The enterprise characteristics include the industry to which the enterprise belongs, the number of years since its establishment, the registered capital, the operating status, the size of its workforce, its operating revenue, its R&D investment, the number of intellectual property rights, its project experience, the scale of its financial processing, the number of invoices, its tax payment status, and its service budget; the demand characteristics include the service category, service objectives, expected completion time, service area, budget range, material preparation status, and special requirements.

3. The enterprise service matching method based on a large language model according to claim 1, characterized in that: The process of generating a structured representation of demand intent and service conditions through a large language model includes: inputting natural language service demands into the large language model; having the large language model identify service objects, service actions, service goals, time requirements, necessary conditions, exclusion conditions, and uncertain conditions; outputting structured demand data according to preset field templates; performing field integrity checks, format checks, and confidence assessments on the structured demand data; and generating supplementary queries for missing fields when key fields are missing or the confidence level is below a preset threshold.

4. The enterprise service matching method based on a large language model according to claim 1, characterized in that: The service resources in the service resource library are described using unified service tags, which include one or more of the following: service category tags, industry tags, technology field tags, enterprise size tags, region tags, qualification tags, material tags, timeliness tags, and capability tags. The service descriptions, historical cases, and business capability descriptions of service organizations or service personnel are converted into the unified service tags and corresponding semantic vectors through a large language model.

5. The enterprise service matching method based on a large language model according to claim 1, characterized in that: For project application services, the filtering and comprehensive scoring should be based at least on the enterprise's industry, years of operation, intellectual property status, R&D investment, project foundation, application region, application time, completeness of application materials, and fulfillment of application conditions. For technology services, the criteria should be based at least on the enterprise's technology field, technology needs, R&D stage, technological achievements, professional capabilities of service personnel, service cases, and delivery cycle. For bookkeeping services, the criteria should be based at least on the enterprise's size, number of invoices, complexity of accounting, tax type, completeness of financial data, service cycle, and financial and tax service capabilities.

6. The enterprise service matching method based on a large language model according to claim 1, characterized in that: The comprehensive score of the candidate service resources is determined as follows: the semantic similarity between enterprise characteristics and service resource characteristics is calculated to obtain a semantic matching score; the necessary conditions in the enterprise's needs and the service conditions in the service resources are compared item by item to obtain a condition satisfaction score. Based on the service period, service area, budget scope, and the availability of service resources, a business constraint score is obtained. A service quality score is obtained based on historical service evaluations, service completion rates, customer feedback, and anomaly records. The semantic matching score, condition fulfillment score, business constraint score, and service quality score are weighted according to preset weights to obtain a comprehensive score.

7. The enterprise service matching method based on a large language model according to claim 1, characterized in that: Before generating matching results, interpretability analysis is performed on candidate service resources to determine key enterprise characteristics, demand conditions, and service resource characteristics that affect the comprehensive score, and these key characteristics are converted into matching criteria in natural language form. When enterprise information is incomplete, service conditions are uncertain, or candidate service resources do not meet the necessary conditions, corresponding risk warnings are generated in the matching results.

8. The enterprise service matching method based on a large language model according to claim 1, characterized in that: When an enterprise has simultaneous needs for project application, technology services, and bookkeeping services, the method generates a combined service plan based on the priority relationship between multiple needs, material dependencies, service periods, and service resource capacity, and determines the execution order, related materials, responsible parties, and phased output results for each service item.

9. The enterprise service matching method based on a large language model according to claim 1, characterized in that: The feedback information includes at least the adoption status of recommendation results, service contract status, service completion status, customer evaluation, service anomalies, and actual service effects; the enterprise feature weights, service resource quality scores, service category matching rules, and candidate resource ranking weights are adjusted based on the feedback information.