Intelligent auditing system for registration of an electronic procurement platform provider

Abstract

The application relates to the technical field of qualification auditing, in particular to an intelligent auditing system for supplier registration of an electronic procurement platform, which comprises the following modules: a registration information input module, which is used for inputting the basic information of a supplier's enterprise to obtain information data; a data uploading and identifying module, which is used for uploading and identifying the unstructured data of the supplier to generate data; a docking module, which is used for docking an official system; a rule auditing module, which is used for constructing an auditing rule library and performing rule auditing on the data of the data uploading and identifying module to obtain a rule auditing result; a comparison and verification module, which compares the official record information with the information data and the data to obtain a verification result; and a judgment module, which obtains a final auditing result according to the verification result and the verification result. The application can accurately verify the registration data of the supplier, unify the auditing standard, reduce the auditing errors caused by human negligence, and improve the accuracy and compliance of the supplier access auditing.

Description

Technical Field

[0001] This invention relates to the field of qualification verification technology, and in particular to an intelligent verification system for supplier registration on an electronic procurement platform. Background Technology

[0002] Against the backdrop of the digitalization and intelligent development of e-procurement, e-procurement platforms have become the core carriers of enterprise procurement. Supplier registration and verification is a crucial step in the platform's supplier access management, directly impacting the compliance, security, and efficiency of procurement transactions. Currently, the supplier registration and verification process on mainstream e-procurement platforms primarily relies on manual reviewers to verify and validate the business licenses, qualification certificates, business scopes, and enterprise information submitted by suppliers one by one.

[0003] The existing technology has the following drawbacks:

[0004] Manual review relies on the professionalism and meticulousness of the reviewers, and is prone to errors in information verification due to human negligence, which may lead to suppliers who do not meet the access standards being approved, thus creating risks in the platform's procurement transactions.

[0005] The entire review process is done manually, requiring manual verification and information comparison of each registration document. The review efficiency is extremely low, especially during peak periods of supplier registration, which can lead to a backlog of reviews and extend the supplier access period.

[0006] Different reviewers have subjective differences in their review standards, which can easily lead to inconsistent review results for the same type of materials. This results in inconsistent review standards and reduces the standardization of platform management.

[0007] Humans cannot connect to official data sources such as industry and commerce bureaus and tax bureaus in real time, making it difficult to fully verify the authenticity of enterprise information submitted by suppliers, identify false registration information, and create loopholes for suppliers to falsify their qualifications.

[0008] Traditional OCR recognition uses a "non-discriminatory" recognition strategy, allocating the same recognition resources to core qualifications and auxiliary materials, as well as core fields and reference fields. This can easily lead to problems such as insufficient recognition accuracy of core information and redundant recognition of auxiliary information.

[0009] The matching of supplier product categories with registration information often relies on single keyword comparison, which cannot accurately identify the suitability of the supplier's actual business scope with the product categories supplied. This easily leads to audit loopholes such as "mismatch between qualifications and business", and fails to meet the refined and personalized audit requirements of electronic procurement platforms for supplier access. Summary of the Invention

[0010] To address the aforementioned issues, this invention provides an intelligent verification system for supplier registration on an electronic procurement platform. This system can accurately verify supplier registration information, standardize verification criteria, reduce verification errors caused by human negligence, and improve the accuracy and compliance of supplier access verification.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] An intelligent verification system for supplier registration on an electronic procurement platform includes:

[0013] The registration information entry module is used to enter the supplier's basic corporate information in order to obtain information data;

[0014] The data upload and recognition module is used for uploading and recognizing unstructured data from suppliers to generate data.

[0015] The interface module is used for connecting to official systems to retrieve the company's official registration information;

[0016] The rule review module is used to review the construction of the rule base. The rule base has corresponding review rules set according to different industries, and performs rule review on the data of the data upload and recognition module to obtain the rule review results.

[0017] The comparison and verification module is used to acquire data from the registration information entry module, the data upload and identification module, and the docking module, and compares the official filing information with the information data and the data data to obtain the verification result.

[0018] The determination module is used to acquire data from the rule review module and the comparison and verification module, so as to obtain the final review result based on the verification result and the verification result.

[0019] Furthermore, the basic enterprise information includes the enterprise name, unified social credit code, registered address, legal representative, and contact information. The registration information entry module has a built-in real-time format verification unit, and the verification rules of the real-time format verification unit are based on the relevant national standards preset, so as to perform real-time verification after the basic enterprise information is entered, and obtain information data after the real-time verification is passed.

[0020] Furthermore, the data upload and identification module includes:

[0021] The data acquisition submodule is used for uploading unstructured data, which includes basic data and registration data. The basic data includes the industry to which the data belongs and a description of the relevant business. The registration data includes a business license, industry qualification certificate, tax registration certificate, bank account opening certificate, and the legal representative's ID card.

[0022] The format verification submodule is used to acquire data from the data acquisition submodule for verifying the registration data.

[0023] The dynamic weight optimization submodule is used to verify the acquisition of the registration data, and the dynamic weight optimization submodule dynamically adjusts the material type weight and field weight of the OCR recognition of the registration data according to the industry and the relevant business description of the industry.

[0024] The OCR recognition submodule is used for OCR recognition of the registration data to obtain structured data.

[0025] Furthermore, the verification dimensions of the format verification submodule include file format compliance, image resolution, data clarity, file size, and qualification validity period. The format verification submodule is also configured with corresponding license verification formats for different industries to verify licenses based on the supplier's industry.

[0026] Furthermore, in the dynamic adjustment of material type weights, the dynamic weight optimization submodule divides the registration materials into different levels of materials based on the core nature of the registration materials and the differences in industry qualification requirements. The material levels of the registration materials include core qualification materials, key qualification materials, and auxiliary materials. The dynamic weight optimization submodule sets corresponding basic weights for each level of materials and calculates the material type weights by using the basic weights of the material types and the industry adaptation coefficient.

[0027] In the dynamic weight optimization submodule, the registration information is divided into different levels of fields according to the importance of the review during the dynamic adjustment of field weights. The field levels of the registration information include core fields, secondary fields, and reference fields. The dynamic weight optimization submodule sets corresponding basic weights for each field level and calculates the comprehensive weight of the field by using the basic weights of the field levels, the weights of the material types, and the industry field weight coefficients.

[0028] After the OCR recognition submodule completes the recognition, it obtains the original recognition confidence of the core field, and calculates the OCR weighted confidence of the corresponding core field by using the original recognition confidence and the comprehensive weight of the field.

[0029] Furthermore, the rule review module constructs the review rule base using the Drools rule engine, and the review rule base includes basic rules, weighted linkage rules, and industry-adaptive rules to review the basic rules, weighted linkage rules, and industry-adaptive rules.

[0030] The review of the basic rules is used to review the completeness of the information in the registration materials, in order to determine whether the completeness review is passed;

[0031] The weighted linkage rule is set with an industry-specific threshold, and the review of the weighted linkage rule is conducted by comparing the OCR weighted confidence score with the industry-specific threshold to determine whether the weighted confidence score review is passed.

[0032] The industry adaptation rules set corresponding matching thresholds based on the supplier's industry attributes, qualification and licensing scope, and business scope. The industry adaptation rules are reviewed by comparing the material type weight with the matching threshold to determine whether the industry adaptation review is passed.

[0033] Furthermore, the comparison and verification module includes:

[0034] The first comparison submodule is used to compare the unstructured data with the data. The first comparison submodule compares the unstructured data with the core fields of the data using a precise comparison algorithm to obtain the core field comparison result. The first comparison submodule also compares the unstructured data with the secondary fields of the data using a fuzzy comparison algorithm to obtain the secondary field comparison result.

[0035] The second comparison submodule is used to compare the data with the official filing information. The second comparison submodule compares all fields of the data with the official filing information using a precise comparison algorithm to obtain the filing comparison result.

[0036] The comparison result quantification submodule is used to calculate the comprehensive comparison score based on the comparison results, the comprehensive weight of the fields, and the industry comparison weight coefficient, and to obtain the verification result based on the comprehensive comparison score.

[0037] Furthermore, it also includes a supplier industry-specific matching module and a differentiated review module.

[0038] The supplier industry precise matching module is used for data acquisition by the data upload and recognition module. The supplier industry precise matching module calculates the semantic matching score, qualification coverage score, and industry relevance score of the supplier through the industry semantic matching model, the industry qualification and license matching model, and the business scope industry association matching model. The module also calculates a comprehensive matching score based on the semantic matching score, the qualification coverage score, and the industry relevance score.

[0039] The differentiated audit module is used for data acquisition by the supplier industry precision matching module. The differentiated audit module sets corresponding core field weighted confidence thresholds, comprehensive adaptation score thresholds, and industry comparison thresholds for different industries. The differentiated audit module is connected to the rule audit module to associate the core field weighted confidence thresholds with the corresponding industry thresholds.

[0040] Furthermore, the industry semantic matching model generates semantic vectors after constructing a semantic library, and calculates the similarity between the semantic vectors and the core semantic vectors of each industry using a cosine similarity algorithm to obtain a semantic matching score, and selects the industry with the highest similarity as the judgment industry for suppliers.

[0041] The industry qualification and licensing matching model constructs a mapping table between industries and qualifications, and uses a string covering algorithm to calculate the qualification and licensing scope of the data and the degree of coverage of industry filing information with the corresponding industry access requirements to obtain a qualification coverage score.

[0042] The business scope industry association matching model identifies related entities between the semantic vector and the determined industry, and calculates the industry association score based on the results of the related entity identification.

[0043] Further, the determination module compares the comprehensive comparison score with the corresponding industry comparison threshold to obtain a first result; the determination module compares the comprehensive adaptation score with the comprehensive adaptation score threshold to obtain a second result, and obtains the final review result based on the rule review result, the first result and the second result.

[0044] The beneficial effects of this invention are:

[0045] The registration information entry module and data upload and recognition module enable intelligent OCR recognition of structured information and unstructured data. Combined with the automated rule matching of the rule review module, this replaces the traditional manual data verification and rule judgment work, improving review efficiency by over 80%, effectively solving the backlog problem and significantly shortening the supplier onboarding cycle. The integration module enables real-time connection with official data sources such as industry and commerce bureaus and tax bureaus. Through the information comparison and verification module, a dual comparison is performed between the entered information, extracted information, and official information, verifying the authenticity of supplier information from the data source. This completely solves the problem that manual review cannot fully verify official information, effectively identifying false registration information and reducing the risk of qualification fraud in platform procurement transactions. This invention solidifies supplier admission standards into the rule review module, enabling all suppliers to undergo automated review according to unified rules. This eliminates subjective differences in manual review, standardizes and normalizes review standards, and enhances the uniformity and professionalism of platform supplier management. By replacing manual verification with a comparison and verification module, it avoids review errors caused by human negligence, improves the accuracy of supplier admission review, ensures the qualification quality of suppliers on the platform, and lays the foundation for the compliance of subsequent procurement transactions. Attached Figure Description

[0046] Figure 1 This is a block diagram of a preferred embodiment of the present invention.

[0047] Figure 2 This is a flowchart of the review process for a preferred embodiment of the present invention.

[0048] 1-Registration Information Input Module, 2-Document Upload and Recognition Module, 21-Document Acquisition Submodule, 22-Format Verification Submodule, 23-Dynamic Weight Optimization Submodule, 24-OCR Recognition Submodule, 3-Interconnection Module, 4-Rule Review Module, 5-Comparison and Verification Module, 51-First Comparison Submodule, 52-Second Comparison Submodule, 53-Comparison Result Quantification Submodule, 6-Judgment Module, 7-Supplier Industry Precise Matching Module, 8-Verification Review Module. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0051] Please also see Figure 1 and Figure 2 The intelligent verification system for supplier registration on an electronic procurement platform according to a preferred embodiment of the present invention includes: a registration information input module 1, a data upload and identification module 2, a docking module 3, a rule verification module 4, a comparison and verification module 5, and a judgment module 6.

[0052] Registration Information Entry Module 1 is used for entering basic enterprise information of suppliers to obtain information data. Registration Information Entry Module 1 provides a standardized registration information entry interface for supplier information entry.

[0053] The basic enterprise information includes the enterprise name, unified social credit code, registered address, legal representative, and contact information. The registration information entry module 1 has a built-in real-time format verification unit. The verification rules of this unit are based on preset national standards, allowing for real-time verification after the basic enterprise information is entered. Information data is obtained only after the real-time verification is successful. The real-time format verification unit can perform real-time verification of information with strict format requirements, such as the unified social credit code, mobile phone number, and bank account number. The verification rules are based on preset national standards.

[0054] The data upload and recognition module 2 is used for uploading and recognizing unstructured data from suppliers to generate data.

[0055] Data upload and recognition module 2 includes:

[0056] The data acquisition submodule 21 is used for uploading unstructured data, which includes basic data and registration data. The basic data includes the industry and related business descriptions, while the registration data includes business license, industry qualification certificate, tax registration certificate, bank account opening certificate, and legal representative's ID card.

[0057] In this embodiment, the data acquisition submodule 21 supports suppliers to upload unstructured registration data such as business licenses, industry qualification certificates, tax registration certificates, bank account opening certificates, and legal representative ID cards, and supports formats such as JPG, PNG, and PDF.

[0058] The format verification submodule 22 is used to acquire data from the data acquisition submodule 21 for verification of registration data.

[0059] The verification dimensions of the format verification submodule 22 include file format compliance, image resolution, document clarity, file size, and qualification validity period. The format verification submodule 22 is also set with corresponding license verification formats according to different industries to verify licenses based on the supplier's industry.

[0060] The format verification submodule 22 in this embodiment performs preliminary screening of uploaded materials. Verification dimensions include file format compliance, image resolution (not less than 300 DPI), data clarity (determined through image contrast and blur algorithms), file size (single file not exceeding 10MB), and qualification validity period (preliminarily identifying the certificate validity period field to determine if it has expired). Simultaneously, considering the characteristics of the supplier's industry, industry-specific document verification is added (e.g., the medical industry requires verification of the "Medical Device Business License" format, and the financial industry requires verification of the "Financial License" format). If verification fails, specific error messages are returned to the supplier.

[0061] The dynamic weight optimization submodule 23 is used to obtain the registration data after verification. The dynamic weight optimization submodule 23 dynamically adjusts the material type weight and field weight of the OCR recognition of the registration data according to the industry and related business descriptions.

[0062] In the dynamic adjustment of material type weights, the dynamic weight optimization submodule 23 divides the registration materials into different levels of materials based on the core nature of the registration materials and the differences in industry qualification requirements. The material levels of the registration materials include core qualification materials, key qualification materials, and auxiliary materials. The dynamic weight optimization submodule 23 sets corresponding basic weights for each level of materials and calculates the material type weights by using the basic weights of the material types and the industry adaptation coefficient.

[0063] In this embodiment, the material level of the registration materials is:

[0064] Core qualification materials: Business license, legal representative's ID card (required for all industries, high risk of forgery); supplement with industry-specific core qualifications (e.g., medical industry: Medical Device Business License; financial industry: Financial License; energy industry: Energy Business License; digital economy industry: Value-added Telecommunications Business License).

[0065] Key qualification materials: Industry franchise license, tax registration certificate (adapted to industry characteristics, such as adding "Medical Institution Practice License" for the medical industry, and "Internet Culture Business License" for the digital economy industry);

[0066] Supporting documents: Bank account opening certificate, corporate credit report (for verification purposes, not essential; requirements for supporting documents vary slightly depending on the industry, such as the financial industry which requires additional proof of funds).

[0067] The formula for calculating the basic weight of material type is:

[0068] Material type base weight = Core value assigned to this type of material ÷ Sum of core values ​​assigned to the three types of materials

[0069] Among them, the three types of materials represent core, key, and auxiliary materials, respectively. The core score is assigned to the core score of the material type. Based on the necessity of the review and the characteristics of the industry, the basic score is 6 points for core qualification materials, 3 points for key qualification materials, and 1 point for auxiliary materials. In this embodiment, the total score is 10. It can be fine-tuned according to the characteristics of the industry. For example, in the financial industry, the core score of core qualification materials is assigned 7 points, and the core score of auxiliary materials is assigned 0.8 points.

[0070] The formula for adjusting the weight of material type is:

[0071] Material type weight = Material type base weight × Industry compatibility coefficient

[0072] The industry fit coefficient is set based on differences in industry qualification requirements, with a value ranging from 0.8 to 1.3. This ensures that the weights are dynamically adjusted according to the actual industry needs of suppliers. For example, the coefficient for core materials in the medical industry is 1.2, for the financial industry it is 1.3, and for general industries it is 1.0. If the total weights after adjustment are not equal to 1, they are normalized.

[0073] Normalized material type weight = Adjusted material type weight ÷ Sum of adjusted weights for the three material types

[0074] In the dynamic weight optimization submodule 23, the registration information is divided into different levels of fields according to the importance of the review. The field levels of the registration information include core fields, secondary fields and reference fields. The dynamic weight optimization submodule 23 sets the corresponding field level basic weight for each field level, and calculates the comprehensive field weight by using the field level basic weight, material type weight and industry field weight coefficient.

[0075] In this embodiment, the field levels of the registration information are as follows:

[0076] Core fields: Common core fields for all industries (such as the unified social credit code and company name on a business license); industry-specific core fields (such as the license number and scope of permission for a Medical Device Business License in the medical industry; and the institution code and business scope for a Financial License in the financial industry).

[0077] Secondary fields: general secondary fields applicable to all industries (such as the registered address and legal representative on a business license); industry-specific secondary fields (such as the energy operating area in the energy industry and the business coverage area in the digital economy industry).

[0078] Reference fields: such as the establishment date of the business license, the issuing authority of various qualification certificates, and the weight of reference fields for different industries can be slightly adjusted (e.g., the weight of the issuing authority field is slightly higher in the financial industry than in other industries).

[0079] The formula for calculating the basic weight of a field hierarchy is:

[0080] Field hierarchy base weight = Preset fixed weight value of the field at this hierarchy

[0081] The three levels of fields represent core, secondary, and reference fields, respectively. The preset fixed weight values ​​are 0.6 for core fields, 0.3 for secondary fields, and 0.1 for reference fields, with a total weight of 1. This conforms to the review priority logic and can be fine-tuned according to industry characteristics. For example, in the medical industry, the preset fixed weight value for core fields is 0.65, and the preset fixed weight value for reference fields is 0.05.

[0082] The formula for calculating the overall weight of the fields is:

[0083] Overall Field Weight = Normalized Material Type Weight × Field Hierarchy Base Weight × Industry Field Weight Coefficient

[0084] The industry field weight coefficient is set according to industry characteristics. For example, the weight coefficient of the core industry field in the financial industry is 1.1, while that of ordinary industries is 1.0, which further strengthens the weight allocation for industry differentiation identification.

[0085] OCR recognition submodule 24 is used for OCR recognition of registered data to obtain structured data.

[0086] After the OCR recognition submodule 24 completes the recognition, it obtains the original recognition confidence of the core field and calculates the OCR weighted confidence of the corresponding core field by using the original recognition confidence and the comprehensive weight of the field.

[0087] In this embodiment, recognition resources are allocated based on the overall weight of the fields: fields with higher weights utilize higher-precision recognition models, such as deep learning character recognition models, increase the number of feature extractions, and extend the recognition computation time. Considering industry characteristics, additional recognition verification dimensions are added for industry-specific core fields, such as the scope of permission in the medical industry and the business scope in the financial industry, to improve recognition accuracy. After recognition is complete, the original recognition confidence score for each field is output, ranging from 0 to 1, and a weighted confidence score is calculated based on the overall weight of the fields, as shown in the following formula:

[0088] OCR weighted confidence score = Original recognition confidence score of this field × Overall weight of this field

[0089] The OCR weighted confidence score serves as the criterion for determining the validity of the recognition results. Fields with higher weights require stricter weighted confidence scores. Furthermore, the weighted confidence score thresholds for industry-specific core fields can be adjusted based on industry characteristics; for example, the threshold for core fields specific to the financial industry is ≥0.96, and for the medical industry, it is ≥0.95. After recognition, standardized structured data is obtained.

[0090] Module 3 is used for connecting to official systems to retrieve the company's official registration information.

[0091] Module 3 establishes secure data interfaces with official data sources such as the National Enterprise Credit Information Publicity System, the State Administration for Industry and Commerce, the State Taxation Administration, the People's Bank of China's Enterprise Credit Information System, and licensing and filing systems of various industry regulatory authorities, including the drug regulatory system in the medical industry, the banking and insurance regulatory system in the financial industry, the energy regulatory system in the energy industry, and the cyberspace administration system in the digital economy industry. It adopts HTTPS encryption protocol + interface authentication mechanism (API Key + timestamp signature) to ensure the security and legality of data transmission.

[0092] The integration module 3 has a built-in data retrieval queue. Based on request priority (e.g., requests for core field comparison and industry compatibility verification are given priority), it retrieves official registration information from the relevant industry regulatory authority, taking into account the supplier's industry. It also sets a retrieval frequency limit, such as no more than 5 requests per supplier per minute, to avoid congestion or blocking of the official interface due to high-frequency requests. The integration module 3 then transmits the retrieved official registration information, including the company's business scope, qualification and licensing registration information, industry regulatory authority registration records, and industry-specific compliance information, back to the information comparison and verification module 5 in real time.

[0093] Rule review module 4 is used to build the review rule base. The review rule base has corresponding review rules set according to different industries, and performs rule review on the data of the data upload and recognition module 2 to obtain the rule review results.

[0094] The rule review module 4 uses the Drools rule engine to build a review rule library, which includes basic rules, weighted linkage rules, and industry-adaptive rules for review.

[0095] The basic rules review is used to verify the completeness of the information filled in the registration materials in order to determine whether the completeness review is passed;

[0096] The weighted linkage rules are set with corresponding thresholds for the industry, and the review of the weighted linkage rules is carried out by comparing the OCR weighted confidence score with the corresponding threshold for the industry to determine whether the weighted confidence score review is passed.

[0097] The industry-specific matching rules set corresponding matching thresholds based on the supplier's industry attributes, qualification and licensing scope, and the industry-specific matching rule review process compares the material type weight with the matching threshold to determine whether the industry-specific matching review is passed.

[0098] In this embodiment, the audit rule base is divided into three levels, combining a weighting system, supplier industry characteristics, and industry-differentiated audit standards to achieve accurate auditing of suppliers in different industries:

[0099] Basic rules: Information completeness ≥ 100%, data format compliance, core materials complete, and industry-specific core material requirements defined in light of industry characteristics;

[0100] Weighted linkage rules: The weighted confidence score of the core field OCR is greater than or equal to the corresponding threshold of the industry, such as ≥0.96 for the financial industry, ≥0.95 for the medical industry, ≥0.95 for general industries, and the average weighted confidence score of a single material is greater than or equal to 0.90.

[0101] Industry Adaptation Rules: Combining the matching results of the supplier industry precise matching and differentiated audit module 8 and the industry differentiated audit standards, the adaptability of the supplier's industry attributes and qualification and license scope and business scope is greater than or equal to the matching threshold, such as ≥0.92 for the medical and financial industries, and ≥0.90 for the energy and digital economy industries. At the same time, it meets the specific compliance requirements of the corresponding industry, such as the medical industry needing to have medical device business registration, and the financial industry needing to have fund custody qualifications.

[0102] The rule review module 4 receives the format verification results and weighted confidence data from the registration information entry module 1 and the data upload and recognition module 2. It then performs automated rule matching and review according to the rule base and the review standards of the corresponding industry, and generates a "pass / fail" rule review result.

[0103] The comparison and verification module 5 is used to acquire data from the registration information entry module 1, the data upload and identification module 2, and the docking module 3, and compares the official filing information with the information data and data to obtain the verification results.

[0104] The comparison and verification module 5 includes:

[0105] The first comparison submodule 51 is used to compare unstructured data with data. The first comparison submodule 51 compares the core fields of unstructured data with data using a precise comparison algorithm to obtain the core field comparison results. The first comparison submodule 51 also compares the secondary fields of unstructured data with data using a fuzzy comparison algorithm to obtain the secondary field comparison results.

[0106] In this embodiment, in the first comparison submodule 51, the core fields (field comprehensive weight ≥ 0.3) adopt a precise comparison algorithm, which does not allow any character differences; the secondary fields (0.1 ≤ field comprehensive weight < 0.3) adopt a fuzzy comparison algorithm, which allows differences in conventional expressions such as "road / avenue" and "number / building" in the address; combined with industry characteristics, the industry-specific fields (such as the scope of permission in the medical industry and the business scope in the financial industry) adopt precise comparison, which does not allow deviations.

[0107] The second comparison submodule 52 is used to compare the data with the official filing information. The second comparison submodule 52 compares all fields of the data with the official filing information through a precise comparison algorithm to obtain the filing comparison result.

[0108] In the second comparison submodule 52, all fields are compared precisely, and core fields that fail the comparison are directly marked as "false information". In combination with industry characteristics, the information is compared with the filing information of the industry authorities, such as the digital economy industry comparing the filing of value-added telecommunications services, and the energy industry comparing the filing of energy operations.

[0109] The comparison result quantification submodule 53 is used to calculate the comprehensive comparison score based on the comparison results, the comprehensive weight of the fields, and the industry comparison weight coefficient, and to obtain the verification result based on the comprehensive comparison score.

[0110] Quantification formula for comparison results:

[0111] Overall comparison score = (Sum of the products of all field comparison results and their corresponding field weights and industry comparison weight coefficients) ÷ Sum of all field weights

[0112] Among them, the comprehensive comparison score is the comprehensive comparison score, which ranges from 0 to 1; the single field comparison result is the single field comparison result, with a match being 1 and a non-match being 0; the single field comprehensive weight is the single field comprehensive weight; the total number of fields participating in the comparison is the total number of fields participating in the comparison; and the industry comparison weight coefficient is the industry comparison weight coefficient, such as 1.1 for the medical and financial industries and 1.0 for other industries, to enhance the comparison accuracy of key industries.

[0113] This embodiment also includes a supplier industry precise matching module 7 and a differentiated review module 8.

[0114] The supplier industry precise matching module 7 is used to acquire data from the data upload and recognition module 2. The supplier industry precise matching module 7 calculates the semantic matching score, qualification coverage score, and industry relevance score of the supplier through the industry semantic matching model, the industry qualification and license matching model, and the business scope industry association matching model. It also calculates the comprehensive matching score based on the semantic matching score, qualification coverage score, and industry relevance score.

[0115] The industry semantic matching model generates semantic vectors after constructing a semantic library. The model then uses a cosine similarity algorithm to calculate the similarity between the semantic vectors and the core semantic vectors of each industry to obtain a semantic matching score. Finally, the industry with the highest similarity is selected as the industry for supplier determination.

[0116] In the industry semantic matching model, based on word segmentation analysis, a full semantic library of the electronic procurement platform is constructed, which covers mainstream industries such as healthcare, energy, digital economy, and finance. It includes core industry terms, business scope descriptions, industry-specific identifiers, and a semantic library of supplier registration information.

[0117] Industry semantic library: Collects core terms from various industries such as healthcare, energy, digital economy, and finance (e.g., healthcare: medical devices, medical consumables, diagnostic equipment; finance: credit, custody, payment settlement; energy: electricity, coal, new energy; digital economy: big data, cloud computing, value-added telecommunications), industry business scope descriptions, and industry-specific qualification names, with each industry corresponding to a unique core semantic vector;

[0118] Supplier registration data semantic library: It performs word segmentation, deduplication, and entity recognition (such as recognizing industry-specific entities such as "medical device operation", "electricity sales", "big data services", and "credit business") on the business scope, qualification and license scope, and industry business description text extracted by OCR, and generates semantic vectors.

[0119] The cosine similarity algorithm is used to calculate the similarity between the semantic vector of supplier information and the core semantic vector of each industry. The industry with the highest similarity is selected as the supplier's determining industry. At the same time, the semantic matching score of the industry is calculated. The formula is as follows:

[0120] Semantic matching score = Dot product of supplier data semantic vector and industry core semantic vector ÷ (Magnitude of supplier data semantic vector × Magnitude of industry core semantic vector)

[0121] The dot product of the supplier information semantic vector and the industry core semantic vector is the dot product of the two vectors. The magnitude of the supplier information semantic vector and the magnitude of the industry core semantic vector are the magnitudes of the two vectors, respectively. The semantic matching score ranges from 0 to 1. The closer to 1, the higher the semantic matching degree. If the highest similarity is lower than 0.6, it is judged as "industry affiliation is unclear" and manual review is required.

[0122] The industry qualification and licensing matching model constructs a mapping table between industries and qualifications, and uses a string covering algorithm to calculate the scope of qualification and licensing information and the degree of coverage of industry filing information with the corresponding industry access requirements to obtain a qualification coverage score.

[0123] The industry qualification and licensing matching model combines the specific qualification requirements of different industries such as healthcare, energy, digital economy, and finance to verify whether the qualification and licensing scope of OCR data extraction and industry filing information meet the access qualification requirements of the corresponding industry and whether they cover the industry business scope claimed by the supplier. It is a hard verification standard for industry adaptability and the core manifestation of differentiated audit.

[0124] Implementation steps and formulas:

[0125] Industry-Qualification Mapping Table Construction:

[0126] A pre-defined industry-qualification mapping table (configured by the platform management terminal and supports dynamic updates) clearly defines the required qualification types, licensing scope, and industry-specific compliance requirements for each industry, distinguishing the qualification differences between different industries:

[0127] Healthcare industry: The mapped qualification types are "Medical Device Business License" and "Medical Institution Practice License". The scope of the license must include the operation / diagnosis and treatment services of the corresponding category of medical devices. The specific compliance requirement is to have medical industry filing.

[0128] Financial industry: The mapping qualification types are "Financial License" and "Fund Custody Qualification Certificate". The scope of the license must include the corresponding financial business (such as credit and payment). The specific compliance requirement is to be registered with the financial regulatory authorities.

[0129] Energy industry: The mapping qualification types are "Energy Operation License" and "Safety Production License". The scope of the license must include the operation of the corresponding energy category (such as electricity and coal). The specific compliance requirement is to have energy regulatory filing.

[0130] Digital economy industry: The mapping qualification types are "Value-added Telecommunications Business Operating License" and "Internet Culture Operating License". The scope of the license must include the corresponding digital business (such as big data and cloud computing). The specific compliance requirement is to have filed with the Cyberspace Administration of China.

[0131] Industry qualification coverage calculation:

[0132] Using a string overlay algorithm and an industry-qualification mapping table, the degree to which the supplier's submitted qualification and licensing scope and industry filing information cover the corresponding industry access requirements is calculated. The formula is as follows:

[0133] Industry qualification coverage score = (Number of matching keywords / Number of semantic entities) ÷ (Number of keywords required to be covered / Number of semantic entities)

[0134] The number of matched keywords / semantic entities refers to the number of keywords / semantic entities in the supplier's qualification and licensing scope and industry filing information that match the corresponding industry access requirements (qualification type, licensing scope, and specific compliance requirements). The number of keywords / semantic entities that must be covered is the number of access requirements keywords / semantic entities that must be covered in this industry (based on the preset industry-qualification mapping table). The industry qualification coverage score ranges from 0 to 1, where 1 indicates full compliance with industry qualification requirements, less than 1 indicates partial compliance, and 0 indicates non-compliance.

[0135] The business scope industry association matching model identifies related entities by matching semantic vectors with the industry, and calculates the industry association score based on the results of the related entity identification.

[0136] The business scope industry association matching model is an auxiliary matching dimension. It can reflect the semantic relevance between the supplier's business scope and the industry being judged, exclude suppliers that "operate outside the industry" or "operate without relevance", and help improve the accuracy of industry matching. At the same time, it can refine the association judgment criteria by combining the characteristics of the business scope of different industries (for example, the business scope of the medical industry must include core terms such as "medical" and "device", and the financial industry must include core terms such as "finance" and "credit").

[0137] Implementation steps and formulas:

[0138] Industry-related entity identification: This involves identifying related entities by analyzing the semantic vectors of supplier data and the core semantic vectors used to determine the industry. Based on the characteristics of different industries, it identifies the semantic relationship between the two (direct relationship, indirect relationship, no relationship).

[0139] Directly related: The business scope includes core industry terms (such as "medical device operation" in the medical industry and "electricity sales" in the energy industry).

[0140] Indirectly related: The business scope includes industry-related terms (such as "sales of medical consumables" in the medical industry and "information technology services" in the digital economy industry).

[0141] No connection: The business scope has no connection with the core terms or related terms of the industry (e.g., claiming to be in the medical industry, but the business scope only includes "office supplies sales").

[0142] Industry Relevance Calculation: Based on the results of related entity identification, and combined with industry characteristics, a correlation weight coefficient is set to calculate the semantic relevance between the business scope and the determined industry. The formula is as follows: Industry Relevance Score = (Number of Semantically Related Entities ÷ Total Number of Entities in the Business Scope) × Correlation Weight Coefficient. Where, the number of semantically related entities is the number of semantically related entities, the total number of entities in the business scope is the total number of entities in the business scope, and the correlation weight coefficient is the correlation weight coefficient (direct relevance is 1.0, indirect relevance is 0.6, and no relevance is 0). At the same time, it is fine-tuned in combination with industry characteristics (e.g., indirect relevance in the financial and medical industries is 0.5, with stricter requirements). The industry relevance score ranges from 0 to 1.

[0143] In the comprehensive matching score calculation, the scores of the three matching models are combined with the review weights of different industries. The comprehensive matching score between the supplier's industry attributes and registration information is calculated by weighted summation, as shown in the following formula:

[0144] Overall Adaptability Score = Semantic Matching Score × Semantic Matching Dimension Weight + Industry Qualification Coverage Score × Qualification and Permit Matching Dimension Weight + Industry Relevance Score × Business Scope Relevance Matching Dimension Weight

[0145] The three dimensions are weighted according to matching dimensions, and are set differently based on industry characteristics to ensure stricter compatibility verification for key industries: Medical and financial industries: semantic matching dimension weight is 0.3, qualification and license matching dimension weight is 0.5, and business scope association matching dimension weight is 0.2 (emphasizing qualification and license matching); Energy and digital economy industries: semantic matching dimension weight is 0.35, qualification and license matching dimension weight is 0.45, and business scope association matching dimension weight is 0.2; General industries: semantic matching dimension weight is 0.4, qualification and license matching dimension weight is 0.4, and business scope association matching dimension weight is 0.2.

[0146] The differentiated audit module 8 is used to acquire data from the supplier industry precise matching module 7. The differentiated audit module 8 sets corresponding core field weighted confidence thresholds, comprehensive adaptation score thresholds, and industry comparison thresholds for different industries. The differentiated audit module 8 is connected to the rule audit module 4 to associate the core field weighted confidence thresholds with the corresponding industry thresholds.

[0147] The differentiated audit module 8 assigns corresponding differentiated audit standards to suppliers in different industries based on the results of the supplier industry precision matching module 7, and simultaneously pushes them to the intelligent rule audit module 4 and the information comparison and verification module 5 to realize the industry-specific differentiated audit throughout the entire process. The core differentiated settings are as follows:

[0148] Medical industry: The core qualification materials have been expanded to include the "Medical Device Business License" and the "Medical Institution Practice License". The weighted confidence threshold of the core fields must be ≥0.95 and the comprehensive fit score threshold must be ≥0.92. Additional comparison with the drug regulatory system filing information is required. Suppliers without medical qualifications are prohibited from entering the market.

[0149] Financial industry: Core qualification materials now include the "Financial License" and "Fund Custody Qualification Certificate". The weighted confidence threshold for core fields must be ≥0.96, and the comprehensive matching score threshold must be ≥0.92. Additional comparison with the filing information in the China Banking and Insurance Regulatory Commission system is required, with a focus on verifying fund qualifications.

[0150] Energy industry: Core qualification materials include "Energy Operation License" and "Safety Production License". The weighted confidence threshold of core fields must be ≥0.95 and the comprehensive matching score threshold must be ≥0.90. Additional comparison with the filing information in the energy regulatory system is required, with a focus on verifying safety production qualifications.

[0151] Digital economy industry: Core qualification materials include "Value-added Telecommunications Business Operation License" and "Internet Culture Operation License". The weighted confidence threshold of core fields must be ≥0.95 and the comprehensive adaptation score threshold must be ≥0.90. Additional comparison with the filing information in the Internet Information Office system is required, with a focus on verifying the compliance of the business scope.

[0152] The judgment module 6 is used to acquire data from the rule review module 4 and the comparison and verification module 5, so as to obtain the final review result based on the verification result and the verification result.

[0153] The judgment module 6 compares the comprehensive comparison score with the corresponding industry comparison threshold to obtain the first result; the judgment module 6 compares the comprehensive adaptation score with the comprehensive adaptation score threshold to obtain the second result; and the final review result is obtained based on the rule review result, the first result and the second result.

[0154] In this embodiment, when the review result, the first result, and the second result are all passed, the final review result is indicated as passed and recorded as 1:

[0155] If the rule review is approved, the comprehensive comparison score is greater than or equal to the corresponding industry comparison threshold, and the comprehensive adaptation score is greater than or equal to the corresponding industry adaptation threshold. The comprehensive comparison score is greater than or equal to the corresponding industry comparison threshold, which means that the comprehensive comparison score meets the standard (the corresponding industry comparison threshold is the industry comparison threshold, ≥0.98 for medical and financial industries, ≥0.97 for energy and digital economy industries, and ≥0.96 for general industries). The comprehensive adaptation score is greater than or equal to the corresponding industry adaptation threshold, which means that the industry comprehensive adaptation score meets the standard (corresponding industry adaptation threshold).

[0156] If the review results are unsuccessful, both the first and second results will be unsuccessful, and the final review result will be "rejected," recorded as 2.

[0157] If the rule review fails, the overall comparison score is less than the corresponding industry comparison threshold, and the overall fit score is less than the corresponding industry fit threshold. Among them, failure of core fields includes false core information and missing core qualifications. An overall comparison score less than the corresponding industry minimum comparison threshold indicates that the overall comparison difference is too large (the corresponding industry minimum comparison threshold is the industry minimum comparison threshold, and all industries are ≥0.90). An overall fit score less than the corresponding industry minimum fit threshold indicates that the industry fit is extremely poor (the corresponding industry minimum fit threshold is the industry minimum fit threshold, and all industries are ≥0.70).

[0158] If any of the audit results, first result, or second result fails, the final audit result in the table will be marked as "returned" and recorded as 3.

[0159] If the rule review fails, the comprehensive comparison score is less than or equal to the corresponding industry comparison threshold, or the comprehensive compatibility score is less than or equal to the corresponding industry compatibility threshold, the final review result will be a return for modification. Non-core issues include inconsistencies in the comparison of secondary fields, vague supporting materials, failure to meet industry compatibility standards but not meeting the rejection criteria, and failure to meet the weighted confidence level of core fields. At the same time, the specific requirements for return for modification will be clarified based on industry characteristics. For example, if the medical industry returns for modification, medical device business registration must be supplemented, and if the financial industry returns for modification, proof of funds must be supplemented.

[0160] In this embodiment, the judgment module 6 obtains the reasons for rejection and return through the rule review module 4, the comparison and verification module 5, the supplier industry precise matching module 7, and the differentiation review module 8, and then feeds them back to the supplier.

[0161] This embodiment also includes an audit log storage module that establishes real-time data connections with all functional modules. It employs a hybrid storage architecture of "blockchain + relational database" to ensure the immutability and efficient retrieval of information throughout the entire audit process. The module records information including: supplier ID, original registration information (including industry and business description), hash value of uploaded materials (to prevent tampering), audit data from each module (including weight calculation process, OCR confidence, comparison score, industry matching results, and implementation status of differentiated audit standards), audit time, auditor (for manual review), and final audit conclusion. All log information is indexed according to timestamp + supplier ID + industry category, supporting rapid retrieval by audit time, supplier name, audit result, industry category, etc., facilitating subsequent industry classification statistics and compliance audits. The core formula for storage (material hash value calculation): Material hash value = SHA256 algorithm calculation result (the calculation content is material content + timestamp + supplier ID + industry category). The hash algorithm ensures the uniqueness and immutability of the materials and the audit process, while also associating with the industry category for quick querying of industry-related logs.

[0162] This implementation uses a B / S architecture-based management backend for platform review and management personnel, with a three-tiered permission system: ordinary reviewers, senior reviewers, and administrators. It automates and standardizes the review of supplier registration materials, improving review efficiency; verifies the authenticity of information by connecting to official data sources, reducing the risk of fraudulent registrations; optimizes OCR recognition logic through a dynamic weighting mechanism, prioritizing the accuracy of core information recognition; and constructs a multi-dimensional matching method based on supplier-specific characteristics to accurately verify supplier registration materials, unify review standards, reduce review errors caused by human error, and improve the accuracy and compliance of supplier access reviews.

Claims

1. An intelligent verification system for supplier registration on an electronic procurement platform, characterized in that, include: The registration information entry module is used to enter the supplier's basic corporate information in order to obtain information data; The data upload and recognition module is used for uploading and recognizing unstructured data from suppliers to generate data. The interface module is used for connecting to official systems to retrieve the company's official registration information; The rule review module is used to review the construction of the rule base. The rule base has corresponding review rules set according to different industries, and performs rule review on the data of the data upload and recognition module to obtain the rule review results. The comparison and verification module is used to acquire data from the registration information entry module, the data upload and identification module, and the docking module, and compares the official filing information with the information data and the data data to obtain the verification result. The determination module is used to acquire data from the rule review module and the comparison and verification module, so as to obtain the final review result based on the verification result and the verification result.

2. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 1, characterized in that: The basic enterprise information includes the enterprise name, unified social credit code, registered address, legal representative, and contact information. The registration information entry module has a built-in real-time format verification unit, and the verification rules of the real-time format verification unit are based on the relevant national standards to perform real-time verification after the basic enterprise information is entered, and obtain information data after the real-time verification is passed.

3. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 1, characterized in that: The data upload and recognition module includes: The data acquisition submodule is used for uploading unstructured data, which includes basic data and registration data. The basic data includes the industry to which the data belongs and a description of the relevant business. The registration data includes a business license, industry qualification certificate, tax registration certificate, bank account opening certificate, and the legal representative's ID card. The format verification submodule is used to acquire data from the data acquisition submodule for verifying the registration data. The dynamic weight optimization submodule is used to verify the acquisition of the registration data, and the dynamic weight optimization submodule dynamically adjusts the material type weight and field weight of the OCR recognition of the registration data according to the industry and the relevant business description of the industry. The OCR recognition submodule is used for OCR recognition of the registration data to obtain structured data.

4. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 3, characterized in that: The verification dimensions of the format verification submodule include file format compliance, image resolution, document clarity, file size, and qualification validity period. Furthermore, the format verification submodule is configured with corresponding license verification formats for different industries to verify licenses based on the supplier's industry.

5. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 3, characterized in that: In the dynamic adjustment of material type weights, the dynamic weight optimization submodule divides the registration materials into different levels of materials based on the core nature of the registration materials and the differences in industry qualification requirements. The material levels of the registration materials include core qualification materials, key qualification materials, and auxiliary materials. The dynamic weight optimization submodule sets corresponding basic weights for each level of materials and calculates the material type weights by using the basic weights of the material types and the industry adaptation coefficient. In the dynamic weight optimization submodule, the registration information is divided into different levels of fields according to the importance of the review during the dynamic adjustment of field weights. The field levels of the registration information include core fields, secondary fields, and reference fields. The dynamic weight optimization submodule sets corresponding basic weights for each field level and calculates the comprehensive weight of the field by using the basic weights of the field levels, the weights of the material types, and the industry field weight coefficients. After the OCR recognition submodule completes the recognition, it obtains the original recognition confidence of the core field, and calculates the OCR weighted confidence of the corresponding core field by using the original recognition confidence and the comprehensive weight of the field.

6. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 5, characterized in that: The rule review module constructs the review rule library using the Drools rule engine. This review rule library includes basic rules, weighted linkage rules, and industry-adaptive rules for reviewing these rules. The review of the basic rules is used to review the completeness of the information in the registration materials in order to determine whether the completeness review is passed; The weighted linkage rule is set with an industry-specific threshold, and the review of the weighted linkage rule is conducted by comparing the OCR weighted confidence score with the industry-specific threshold to determine whether the weighted confidence score review is passed. The industry adaptation rules set corresponding matching thresholds based on the supplier's industry attributes, qualification and licensing scope, and business scope. The industry adaptation rules are reviewed by comparing the material type weight with the matching threshold to determine whether the industry adaptation review is passed.

7. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 6, characterized in that: The comparison and verification module includes: The first comparison submodule is used to compare the unstructured data with the data. The first comparison submodule compares the unstructured data with the core fields of the data using a precise comparison algorithm to obtain the core field comparison result. The first comparison submodule also compares the unstructured data with the secondary fields of the data using a fuzzy comparison algorithm to obtain the secondary field comparison result. The second comparison submodule is used to compare the data with the official filing information. The second comparison submodule compares all fields of the data with the official filing information using a precise comparison algorithm to obtain the filing comparison result. The comparison result quantification submodule is used to calculate the comprehensive comparison score based on the comparison results, the comprehensive weight of the fields, and the industry comparison weight coefficient, and to obtain the verification result based on the comprehensive comparison score.

8. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 7, characterized in that: It also includes a supplier industry-specific matching module and a differentiated review module. The supplier industry precise matching module is used for data acquisition by the data upload and recognition module. The supplier industry precise matching module calculates the semantic matching score, qualification coverage score, and industry relevance score of the supplier through the industry semantic matching model, the industry qualification and license matching model, and the business scope industry association matching model. The module also calculates a comprehensive matching score based on the semantic matching score, the qualification coverage score, and the industry relevance score. The differentiated audit module is used for data acquisition by the supplier industry precision matching module. The differentiated audit module sets corresponding core field weighted confidence thresholds, comprehensive adaptation score thresholds, and industry comparison thresholds for different industries. The differentiated audit module is connected to the rule audit module to associate the core field weighted confidence thresholds with the corresponding industry thresholds.

9. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 8, characterized in that: The industry semantic matching model generates semantic vectors after constructing a semantic library. The industry semantic matching model calculates the similarity between the semantic vectors and the core semantic vectors of each industry using a cosine similarity algorithm to obtain a semantic matching score. The industry with the highest similarity is selected as the industry for supplier determination. The industry qualification and licensing matching model constructs a mapping table between industries and qualifications, and uses a string covering algorithm to calculate the qualification and licensing scope of the data and the degree of coverage of industry filing information with the corresponding industry access requirements to obtain a qualification coverage score. The business scope industry association matching model identifies related entities between the semantic vector and the determined industry, and calculates the industry association score based on the results of the related entity identification.

10. The intelligent verification system for supplier registration on an electronic procurement platform according to claim 9, characterized in that: The determination module compares the comprehensive comparison score with the corresponding industry comparison threshold to obtain a first result; The determination module compares the comprehensive adaptation score with the comprehensive adaptation score threshold to obtain a second result, and obtains the final review result based on the rule review result, the first result, and the second result.

Images