Supplier evaluation and selection method based on large model semantic parsing and scenario adaptation

By constructing a supplier business database and a procurement expertise base, and utilizing a large language model for semantic parsing and scenario-adaptive evaluation, the problems of automation and interpretability of supplier selection methods are solved, achieving compliant, intelligent, and interpretable supplier evaluation and selection.

CN123117967APending Publication Date: 2026-11-17NO 15 INST OF CHINA ELECTRONICS TECH GRP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611087973.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing supplier selection methods have low automation levels, making it difficult to cope with the exponential growth of supplier data and the fragmentation of procurement tasks. They cannot identify implicit demands, the evaluation results deviate from the actual business preferences, and the decisions are unexplainable.

Method used

We construct a supplier business database and a procurement knowledge base, use a large language model for semantic normalization, parse supplier constraints and identify target scenario types, and achieve multi-dimensional feature scoring and ranking through dynamic weight adjustment, providing explainable recommendation reasons.

Benefits of technology

It achieves compliance, traceability, and explainability in the supplier selection process, enabling quantitative ranking and tracing back to the original data, and meeting the procurement needs of dynamic adaptation in multiple scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117967A_ABST
    Figure CN123117967A_ABST
Patent Text Reader

Abstract

This application provides a supplier evaluation and selection method based on large-scale model semantic parsing and scenario adaptation, including: constructing a supplier business database and a procurement professional knowledge base; inputting procurement requirements expressed in natural language into a large language model, performing semantic normalization processing using the procurement professional knowledge base, parsing supplier constraints and identifying target scenario types; performing product category filtering and compliance filtering on all suppliers according to the parsed first constraint to obtain a candidate supplier pool; for each supplier in the candidate supplier pool, obtaining multi-dimensional feature data and performing quantitative scoring according to a second constraint, then obtaining the target dynamic weights corresponding to the target scenario type and performing weighted summation to obtain a comprehensive matching score for each supplier; and outputting a qualified supplier list and traceable recommendation reasons after sorting the comprehensive matching scores. Through a deep understanding of procurement requirements, scenario-adaptive dynamic weights, and recommendation reasons, the accuracy of supplier evaluation is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and supply chain management technology, and in particular to a supplier evaluation and selection method based on large model semantic parsing and scenario adaptation. Background Technology

[0002] In a supply chain management system, supplier evaluation and selection directly affect a company's procurement costs, product quality, and supply chain security. As the procurement environment becomes increasingly complex, supplier selection has evolved from a single price-oriented approach to a comprehensive consideration of multiple dimensions, including quality, delivery, R&D collaboration, and operational risks.

[0003] Currently, supplier selection primarily relies on two technical approaches. One is a rule-based screening mechanism based on structured data, utilizing ERP or supplier management systems. Procurement personnel manually review hard criteria such as qualifications and delivery dates, then rank suppliers using form matching or fixed-weight scoring. This method has low automation and rigid rules, making it difficult to address the exponential growth of supplier data and the fragmented, high-frequency nature of procurement tasks. The other approach relies on intelligent recommendation methods based on machine learning models. These methods establish feature maps by fitting historical transaction data to achieve scoring and ranking. While this method reduces manual intervention, it heavily depends on high-quality labeled datasets, and model training struggles to incorporate compliance red lines and implicit professional constraints in procurement operations, limiting its effectiveness for industrial application.

[0004] In summary, the challenges of supplier selection in practical applications lie in the following aspects: First, procurement needs are often expressed in unstructured forms such as technical specifications and communication minutes, described in natural language. Existing tools can only extract explicit keywords for matching, failing to perceive implicit requirements such as process adaptation experience and R&D collaboration capabilities, resulting in matching that is merely literal rather than substantive. Second, procurement scenarios are diverse. The emphasis on various evaluation dimensions differs significantly in different scenarios such as mass production, expedited orders, and R&D trial production. However, the evaluation mechanism uses a uniform static scheme for compliance evaluation, leading to evaluation results that deviate from the actual business preferences. Third, procurement behavior is subject to strong compliance constraints, requiring decisions to be traceable and explainable. However, existing intelligent methods are mostly black-box models, making it difficult to provide reasons for recommended selections.

[0005] Therefore, there is an urgent need for a supplier evaluation and selection scheme that can take into account a deep understanding of unstructured needs, dynamic adaptation to multiple scenarios, and compliance and explainability. Summary of the Invention

[0006] This application provides a supplier evaluation and selection method based on large model semantic parsing and scenario adaptation, which can achieve compliant record-keeping of the qualified supplier selection process and provide an efficient, intelligent and explainable solution for compliant procurement.

[0007] In a first aspect, embodiments of this application provide a supplier evaluation and selection method based on large-model semantic parsing and scenario adaptation, the method comprising: Construct a supplier business database and a procurement professional knowledge base. The supplier business database is used to store business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports supplier selection tasks in procurement scenarios. The procurement requirements expressed in natural language are input into a large language model. The large language model is used to perform semantic normalization on the procurement requirements through the procurement professional knowledge base. The supplier constraints are parsed from the semantic normalization results, and the target scenario type to which the procurement requirements belong is identified. Based on the first constraint in the supplier constraints, product category filtering and compliance filtering are performed on all suppliers to obtain a candidate supplier pool. For each supplier in the candidate supplier pool, according to the second constraint in the supplier constraint conditions, the multi-dimensional feature data corresponding to each supplier is obtained from the supplier business database and quantitatively scored. Obtain the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight value corresponding to each dimension feature. Different scenario types correspond to different dynamic weights. The quantitative scoring results of multiple dimensions corresponding to each supplier are weighted and summed with the weight values ​​in the target dynamic weight to obtain the comprehensive matching score for each supplier. The comprehensive matching scores are sorted according to their values, and a list of qualified suppliers is output, along with a recommendation reason for each qualified supplier. The recommendation reason includes the multi-dimensional feature data and comprehensive matching score of each supplier.

[0008] Secondly, embodiments of this application provide a supplier evaluation and selection device based on large model semantic parsing and scenario adaptation, the device comprising: The module is used to build a supplier business database and a procurement professional knowledge base. The supplier business database is used to store business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports supplier selection tasks in procurement scenarios. The processing module is used to input the acquired procurement requirements expressed in natural language into a large language model, and use the large language model to perform semantic normalization processing on the procurement requirements through the procurement professional knowledge base. It then parses the supplier constraints and identifies the target scenario type to which the procurement requirements belong from the semantic normalization processing result. Based on the first constraint in the supplier constraints, it performs product category filtering and compliance filtering on all suppliers to obtain a candidate supplier pool. For each supplier in the candidate supplier pool, based on the second constraint in the supplier constraints, it retrieves multi-dimensional feature data corresponding to each supplier from the supplier business database and performs quantitative scoring. It retrieves the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight values ​​corresponding to each dimension feature, with different dynamic weights corresponding to different scenario types. Finally, it performs a weighted sum of the quantitative scoring results of multiple dimensions for each supplier and the weight values ​​in the target dynamic weight to obtain a comprehensive matching score for each supplier. The output module is used to sort the comprehensive matching scores according to the scores and output the sorted list of qualified suppliers, as well as the recommendation reasons for each qualified supplier. The recommendation reasons include the multi-dimensional feature data and comprehensive matching scores of each supplier.

[0009] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a communication interface; wherein, the memory stores a computer program, and when the computer program is executed by the processor, the processor can at least implement the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in the first aspect.

[0010] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor of an electronic device, enables the processor to at least implement the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in the first aspect.

[0011] Fifthly, embodiments of this application provide a computer program product, including: a computer program or instructions, which, when executed by a processor of an electronic device, enable the processor to at least implement the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in the first aspect.

[0012] The solution provided in this application first constructs a supplier business database and a procurement professional knowledge base to provide complete data and knowledge support for subsequent semantic parsing, compliance filtering, quantitative scoring, and weight configuration. The supplier business database stores the business capability data of supplier companies, while the procurement professional knowledge base stores domain knowledge supporting supplier selection tasks in procurement scenarios. In the supplier selection scenario, procurement requirements expressed in natural language are input into a large language model. This model uses the procurement professional knowledge base to perform semantic normalization processing on the procurement requirements, eliminating colloquial ambiguities and terminological inconsistencies in natural language expressions. This accurately parses out supplier constraints and identifies the target scenario type to which the procurement requirements belong. Furthermore, based on the parsed first constraint, product category filtering and compliance filtering are performed on all suppliers to obtain a candidate supplier pool. Suppliers that do not meet basic access qualifications are eliminated in advance, ensuring that all suppliers entering subsequent evaluations meet the procurement compliance baseline. For each supplier in the candidate supplier pool, multi-dimensional feature data is retrieved from the supplier business database according to the second constraint and quantitatively scored. Target dynamic weights corresponding to the target scenario type are obtained from the procurement professional knowledge base. The quantitative scores of each supplier across multiple dimensions are weighted and summed with the weights in the target dynamic weights to obtain a comprehensive matching score for each supplier. Since different scenario types correspond to different dynamic weights, the comprehensive matching score obtained after weighted summation can adaptively adjust the proportion of each evaluation dimension according to the actual focus of the procurement task, achieving precise alignment between the evaluation results and actual business preferences. Finally, the comprehensive matching scores are sorted by score, and a list of qualified suppliers and a recommendation reason for each supplier are output. This recommendation reason includes the multi-dimensional feature data and comprehensive matching score for each supplier, making the recommendation results not only quantitatively sortable but also traceable to the original data source of each score, effectively overcoming the problem that traditional black-box models cannot explain the decision-making basis. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating a supplier evaluation and selection method based on large model semantic parsing and scene adaptation, provided for embodiments of this application; Figure 2 A general flowchart of a supplier evaluation and selection method provided for embodiments of this application; Figure 3A schematic diagram illustrating the construction of a supplier business database and a procurement expertise base, provided for embodiments of this application; Figure 4 A schematic diagram illustrating a scene-adaptive multi-dimensional feature index fusion calculation provided in an embodiment of this application; Figure 5 This is a schematic diagram illustrating a supplier sorting and selection method provided in an embodiment of this application. Figure 6 A schematic diagram of the structure of a supplier evaluation and selection device based on large model semantic parsing and scene adaptation provided in an embodiment of this application; Figure 7 To and Figure 6 The illustrated embodiment provides a schematic diagram of the electronic device corresponding to the supplier evaluation and selection device based on large model semantic parsing and scene adaptation. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards.

[0017] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0018] To address the problems of existing supplier selection methods, such as strong reliance on human intervention, rigid evaluation schemes, insufficient scenario adaptability, difficulty in identifying implicit needs, and poor decision interpretability, this application provides a supplier evaluation and selection method based on large-model semantic parsing and scenario adaptation, the concept of which is as follows: First, a supplier business database and a procurement knowledge base are constructed. The supplier business database stores supplier companies' business capability data, including structured data across multiple dimensions such as basic company information, cooperation history, production capacity, R&D capabilities, contract fulfillment services, and operational risks. The procurement knowledge base stores domain knowledge supporting supplier selection in procurement scenarios, including standard terminology for eliminating semantic ambiguity, evaluation rules for determining compliance, and scenario weight data for dynamically configuring weights. Based on this, the procurement requirements expressed in natural language are input into a Large Language Model (LLM). The LLM uses the procurement knowledge base to semantically normalize the procurement requirements, eliminating ambiguity caused by colloquial expressions and terminological differences. Supplier constraints are then parsed from the semantic normalization results, and the target scenario type of the procurement requirements is identified. Subsequently, based on the first constraint in the supplier constraints, product category filtering and compliance filtering are performed on all suppliers, eliminating suppliers that do not meet basic access qualifications, resulting in a candidate supplier pool. For each supplier in the candidate supplier pool, based on the second constraint in the supplier constraints, multi-dimensional feature data corresponding to each supplier is retrieved from the supplier business database and quantitatively scored. Simultaneously, target dynamic weights corresponding to the target scenario type are obtained from the procurement professional knowledge base. These target dynamic weights include the weight values ​​corresponding to each dimension of features, with different dynamic weights corresponding to different scenario types. The quantitative scores of each supplier across multiple dimensions are weighted and summed with the weight values ​​in the target dynamic weights to obtain the comprehensive matching score for each supplier. Finally, the comprehensive matching scores are sorted according to their scores, and a sorted list of qualified suppliers is output, along with a recommendation reason for each qualified supplier. This recommendation reason includes the multi-dimensional feature data and comprehensive matching score for each supplier, ensuring that the recommendation results are both quantitatively sorted and traceable to the original data, achieving compliant record-keeping and interpretable output of the supplier selection process.

[0019] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0020] The supplier evaluation and selection method based on large model semantic parsing and scene adaptation provided in this application can be executed by an electronic device, which can be a terminal device such as a PC, laptop, or smartphone, or a server. The server can be a physical server containing an independent host, a virtual server, a cloud server, or a server cluster.

[0021] Figure 1 A flowchart illustrating a supplier evaluation and selection method based on large model semantic parsing and scene adaptation, provided for embodiments of this application, is shown below. Figure 1As shown, it may include the following steps 101 to 102: 101. Construct a supplier business database and a procurement professional knowledge base. The supplier business database is used to store the business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports the supplier selection task in the procurement scenario.

[0022] 102. Input the procurement requirements expressed in natural language into the large language model, and use the large language model to perform semantic normalization processing on the procurement requirements through the procurement professional knowledge base. Then, parse the supplier constraints and identify the target scenario type to which the procurement requirements belong from the semantic normalization processing results.

[0023] 103. Based on the first constraint in the supplier constraints, perform product category filtering and compliance filtering on all suppliers to obtain a candidate supplier pool.

[0024] 104. For each supplier in the candidate supplier pool, according to the second constraint in the supplier constraint conditions, obtain the multi-dimensional feature data corresponding to each supplier from the supplier business database and perform quantitative scoring.

[0025] 105. Obtain the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight value corresponding to each dimension feature. Different scenario types correspond to different dynamic weights.

[0026] 106. The quantitative scores of multiple dimensions for each supplier are weighted and summed with the weight values ​​in the target dynamic weights to obtain the comprehensive matching score for each supplier.

[0027] 107. Sort the comprehensive matching scores according to the scores and output the sorted list of qualified suppliers, as well as the recommendation reasons for each qualified supplier. The recommendation reasons include the multi-dimensional feature data and comprehensive matching score of each supplier.

[0028] Figure 2 A general flowchart of a supplier evaluation and selection method provided for embodiments of this application is shown below. Figure 2 As shown, the supplier evaluation and selection scheme provided in this application embodiment is divided into three parts: The first part involves collecting data from multiple sources, integrating original business data from suppliers with professional knowledge in the procurement field, to establish a supplier business database and a procurement professional knowledge base. By building these two databases, data support will be provided for subsequent evaluation and ranking.

[0029] The second part constructs a multi-dimensional characteristic evaluation index system for suppliers, clarifies evaluation dimensions such as enterprise foundation, cooperation history, production and R&D, and performance risk control, and designs calculation methods for each dimension index under multiple scenarios to achieve quantitative mapping of supplier capabilities.

[0030] The third part uses a large language model to analyze procurement needs, combines scenario-adaptive weighting and progressive optimization logic, sorts suppliers by score, and outputs the selectable suppliers and the reasons for recommendation.

[0031] Next, for Figure 1 Steps 101 to 107 in the embodiments will be described in detail.

[0032] 101. Construct a supplier business database and a procurement professional knowledge base. The supplier business database is used to store the business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports the supplier selection task in the procurement scenario.

[0033] For the supplier business database, it can be a structured relational database used to store real, fixed, and statistically verifiable business capability data of supplier companies, such as procurement data, supplier management data, and historical project data. The data tables in the supplier business database are interconnected through company identifiers, supporting precise field-level retrieval and quantitative calculations. All data consists of structured fields actually collected by the company, thus ensuring the authenticity and traceability of subsequent feature calculations.

[0034] Figure 3 This application provides a schematic diagram illustrating the construction of a supplier business database and a procurement expertise base, as shown in the embodiments. Figure 3 As shown in the left figure, the supplier business database can include basic information tables, qualification and compliance information tables, production capacity information tables, R&D technology information tables, performance service information tables, cooperation history information tables, and risk information tables.

[0035] The system includes the following tables: Basic Information Table: Stores inherent static data such as the company's years of operation, registered capital, company size, business address, and main product categories. Compliance Information Table: Stores compliance information such as system certifications, special qualifications, credit ratings, administrative penalties, and blacklisting. Production Capacity Information Table: Stores production data such as production line automation level, equipment precision, maximum production capacity, tooling configuration, and expedited production capacity. R&D Technology Information Table: Stores R&D data such as the number of patents, the proportion of R&D personnel, laboratory capabilities, and experience with similar products. Contract Performance Information Table: Stores contract performance data such as historical on-time delivery rate, after-sales response time, anomaly handling records, and cooperation evaluations. Cooperation History Information Table: Stores cooperation data such as supplier's historical supply records, product categories supplied, cooperation frequency, and customer evaluations. Risk Information Table: Stores negative risk data such as litigation records, operational anomalies, debt-to-asset ratio, and risk of supply disruption.

[0036] The procurement expertise base stores domain knowledge that supports supplier selection in procurement scenarios. It serves as the core data foundation for assisting large language models in deeply understanding the semantics of procurement business and standardizing evaluation criteria.

[0037] like Figure 3 As shown in the right figure, a hierarchical procurement expertise base can be constructed by collecting industry standards, compliant procurement rules, algorithm scoring parameters, historical cases, etc., and then classifying them into standard terminology, evaluation rules, scenario weights, procurement common sense, and recommended cases. This procurement expertise base can include sub-bases such as a standard terminology database, an evaluation rule database, a scenario weight database, a procurement common sense database, and a historical case database. These sub-bases are interdependent and progressively supportive of each other from bottom to top, jointly constructing a complete procurement knowledge reasoning system.

[0038] Specifically, the standard terminology library stores professional terms, industry-standard names, aliases, and abbreviations mappings for the procurement industry and component sector. This ensures consistent semantic language when the large language model parses procurement requirements, eliminating semantic ambiguity caused by colloquial descriptions and aliases, and improving parsing accuracy. For example, diverse expressions such as "precision parts" and "precision-machined structural parts" are normalized into the standard term: "precision mechanical structural parts."

[0039] The evaluation rule library stores the scoring criteria, deduction details, rigid compliance access constraints, and veto conditions for each evaluation dimension. It provides standardized judgment benchmarks during the compliance filtering stage, ensuring the standardization and unbiasedness of the evaluation. For example, setting a supplier's history of dishonesty or significant administrative penalties as a veto condition directly eliminates suppliers with incomplete qualifications.

[0040] The scenario weight library stores weight allocation templates for various typical procurement scenarios. Each scenario has an independent set of weight values ​​for each evaluation dimension, and the sum of the weight values ​​for each evaluation dimension within the same scenario is always equal to 1. The scenario weight library provides the system with pre-configured weight criteria. After identifying the target scenario type in subsequent steps, the system can automatically retrieve the corresponding weight value set, achieving dynamic weighted evaluation without manual intervention. For example, in expedited procurement scenarios, a higher weight is preset for the fulfillment service dimension to align with the urgent delivery requirements; in customized R&D scenarios, a higher weight is preset for the R&D technology dimension to adapt to technology-capability-oriented procurement needs, and so on.

[0041] The procurement common sense base stores industry-standard procurement practices, including general procurement standards for components, industry supply specifications, and general screening criteria. This assists the large language model in determining basic supply compliance when parsing procurement requirements and supplements industry-standard judgment logic. For example, requirements such as rust prevention treatment and burr-free surfaces for metal components are used as basic common sense constraints to help identify supply compliance requirements.

[0042] The historical case library stores selected high-quality and qualified suppliers from past procurement cases, recommendation reason templates, and compliance evaluation templates. It provides standardized wording when generating recommendation reasons, ensuring their standardization, consistency, and explainability. For example, based on historical cases of similar precision component procurement, a standardized recommendation statement can be generated: complete qualifications, matching processes, and stable delivery.

[0043] In this implementation, the five sub-libraries are not independent of each other, but rather interconnected in a bottom-up, hierarchical, and progressively supportive manner: the standard terminology library unifies professional expressions, laying the semantic foundation for upper-level applications; the evaluation rule library provides judgment constraints to ensure the standardization of compliance filtering; the scenario weight library relies on the first two layers to achieve intelligent allocation of scenario weights; the procurement common sense library supplements industry-wide standards, helping to avoid loopholes in professional judgment; and the historical case library relies on the knowledge of the first four layers to generate standardized explanatory language. Through the coordinated efforts of these sub-libraries, they jointly provide end-to-end knowledge support for the semantic parsing, compliance filtering, quantitative scoring, and recommendation reason generation of the large language model.

[0044] 102. Input the procurement requirements expressed in natural language into the large language model, and use the large language model to perform semantic normalization processing on the procurement requirements through the procurement professional knowledge base. Then, parse the supplier constraints and identify the target scenario type to which the procurement requirements belong from the semantic normalization processing results.

[0045] In this embodiment, the procurement requirements are input in natural language, and their specific form can be technical specification text, product design summary, supply requirement description or transcript of oral communication entered by the procurement personnel.

[0046] In practical applications, due to issues such as colloquialisms, industry aliases, abbreviations, and inconsistent terminology in natural language expressions, directly performing keyword matching or rule parsing based on the original text will make it difficult to accurately understand the purchaser's true needs. Therefore, this application's embodiments utilize the natural language understanding capabilities of a large language model, combining it with a procurement expertise base to semantically normalize procurement requirements.

[0047] Specifically, semantic normalization processing includes: inputting the obtained natural language expression of the procurement requirements text into the large language model; when parsing the text, the large language model calls the standard terminology library in the procurement professional knowledge base to perform term mapping and ambiguity elimination on various expressions appearing in the procurement requirements text.

[0048] As mentioned above, the standard terminology database pre-stores professional terms, industry common names, aliases, and abbreviation mappings for the procurement industry and parts field. For example, diverse expressions such as precision parts, precision-machined structural parts, and high-precision machined parts are unified and standardized into the standard term: precision mechanical structural parts; diverse expressions such as delivery date, delivery period, and delivery time are unified and standardized into the standard term: delivery cycle.

[0049] Through the above terminology mapping process, semantic ambiguity caused by colloquial expressions and terminology differences can be eliminated, and procurement requirements can be transformed into unified and standardized expressions, providing a standardized input basis for subsequent constraint analysis and scenario recognition.

[0050] After semantic normalization, the large language model parses supplier constraints from the normalized standard representation. Specifically, the large language model calls upon the evaluation rule base and procurement common sense base in the procurement expertise base to determine the constraints used to evaluate supplier compliance.

[0051] The evaluation rule base stores scoring standards, deduction details, rigid compliance access red lines, and veto conditions for each evaluation dimension; the procurement common sense base stores general procurement standards for components, industry supply specifications, and general screening criteria. By combining these two knowledge sub-bases, the large language model can identify and extract two types of supplier constraints from the procurement requirement text: the first constraint is rigid compliance access conditions, such as required product categories, essential qualification requirements, blacklist exclusion requirements, and abnormal business operation exclusion requirements, used to determine whether the supplier possesses the basic access qualifications to participate in this procurement task; the second constraint is screening constraints related to matching the procurement requirements, such as delivery cycle requirements and supply batch requirements, used to limit the scope of extraction of multi-dimensional feature data from the supplier's business database.

[0052] Simultaneously, the large language model can also identify the target scenario type of the procurement requirement from the semantic normalization processing results. Specifically, the large language model extracts scenario determination information from the semantic normalization processing results, which includes at least one of the following: delivery urgency, product customization level, and procurement batch size. For example, when the procurement requirement includes expressions such as urgent, rush, or within 24 hours, the large language model extracts the delivery urgency level as high priority; when the procurement requirement includes expressions such as customized development, new mold opening, or non-standard design, the large language model extracts the product customization level as high level; when the procurement requirement includes expressions such as large batch, annual framework, or long-term supply, the large language model extracts the procurement batch size as large batch.

[0053] The large language model matches the extracted scenario judgment information with multiple preset scenario judgment conditions in the procurement professional knowledge base, and outputs successfully matched scenario tags. For example, scenario tags may include batch production scenarios oriented towards mass production stability, expedited procurement scenarios oriented towards delivery timeliness, and R&D customization scenarios oriented towards technical capabilities. They may also include regular inventory preparation scenarios and long-term supply scenarios. This application does not limit the specific scenario type classification, and in practical applications, it can be customized according to the characteristics of the enterprise's procurement business.

[0054] Through the above process, this step transforms unstructured natural language procurement requirements into structured output results, specifically including: standardized procurement requirement statements, first constraints, second constraints, and target scenario type labels, providing standardized input parameters for subsequent compliance filtering, quantitative scoring, and dynamic weight configuration.

[0055] 103. Based on the first constraint in the supplier constraints, perform product category filtering and compliance filtering on all suppliers to obtain a candidate supplier pool.

[0056] In this step, the first constraint is the rigid compliance access condition parsed from the procurement requirement text, which includes two levels: product category constraint and compliance access constraint.

[0057] The product category constraint is used to limit the range of product categories used in this initial supplier screening. Specifically, the category description of the required products is parsed from the procurement request text and used as the benchmark for category matching. The category conformity between each supplier's main products and the required products is calculated from the supplier's business database. The category matching range includes standard products, synonym products, subordinate products, superior products, and similar products of the required products. The number of superior products output cannot exceed 3, and the number of subordinate products output cannot exceed 4. Based on the category conformity calculation results, all suppliers are initially screened. A category conformity threshold is set, and suppliers with a similarity greater than the set threshold are marked as candidate suppliers and proceed to the next stage of matching calculation.

[0058] Compliance access constraints include, but are not limited to, mandatory qualification requirements, blacklist exclusion requirements, abnormal business operation exclusion requirements, credit record exclusion requirements, and geographical delivery constraints. Compliance access constraints are veto-qualifying; suppliers either meet them or do not, with no intermediate state. When performing compliance filtering, the first constraint condition is converted into a standardized query or judgment rule, which is then compared with the corresponding fields of each supplier in the supplier's business database to determine whether each supplier in the entire database meets all the first constraint conditions. For example, if the first constraint conditions include both ISO9001 certification and not being blacklisted, the system certification field and blacklist flag field in the supplier's business database's qualification and compliance information table are compared. Suppliers that meet both requirements are retained, while suppliers that do not meet either requirement are directly eliminated.

[0059] Product category filtering and compliance filtering can be executed in different orders, and this application does not limit this. For example, product category filtering and compliance filtering can be executed sequentially, with compliance filtering executed first and then product category filtering executed; or product category filtering can be executed first and then compliance filtering executed.

[0060] In practical applications, the execution order of the two processes can be flexibly configured based on the data volume and computational complexity. For example, compliance filtering can be performed first, followed by product category filtering. This means first removing suppliers who do not meet basic compliance eligibility criteria from the candidate pool, and then performing product category matching calculations on the remaining suppliers. This allows for reducing the data size before category matching, decreasing the computational load of category matching calculations, and improving processing efficiency. Alternatively, product category filtering can also serve as a pre-processing step for compliance filtering. Suppliers relevant to the procurement needs can be initially screened through category matching, followed by compliance eligibility verification. Both sequences are within the scope of protection of this application.

[0061] Through the combined screening of product category filtering and compliance filtering, suppliers who do not meet the basic access qualifications or whose product categories do not meet the requirements are excluded from the subsequent evaluation process, and the remaining suppliers form a candidate supplier pool. This candidate supplier pool serves as the data basis for subsequent quantitative scoring and ranking processes, ensuring that all suppliers entering the subsequent evaluation stage have met the category compliance requirements and compliance baselines set in the procurement needs.

[0062] 104. For each supplier in the candidate supplier pool, according to the second constraint in the supplier constraint conditions, obtain the multi-dimensional feature data corresponding to each supplier from the supplier business database and perform quantitative scoring.

[0063] Figure 4 This is a schematic diagram of a scene-adaptive multi-dimensional feature index fusion calculation provided in an embodiment of this application, as shown below. Figure 4As shown, for each supplier in the candidate supplier pool obtained after compliance filtering, multi-dimensional feature data acquisition and quantitative scoring operations are performed.

[0064] Specifically, the second constraint is a screening constraint related to matching the procurement requirements, parsed from the procurement needs. This includes, but is not limited to, delivery cycle requirements and supply batch requirements, used to limit the time window and business scenario scope of the data used for this evaluation. The second constraint is used to extract multi-dimensional feature data matching the current procurement requirements for each candidate supplier from each data table in the supplier business database.

[0065] In this embodiment of the application, a feature evaluation system covering multiple evaluation dimensions is constructed based on the field system of the supplier's business database to comprehensively measure the supplier's capability level in different dimensions. For example... Figure 4 As shown, this multi-dimensional feature evaluation system specifically includes the following feature dimensions: basic static features, historical cooperation features, production and processing capability features, qualification and compliance features, R&D technology features, performance service features, and operational risk features.

[0066] Among them, the basic static characteristics are used to evaluate the inherent attributes of the supplier company, including: years of operation, registered capital, company size, company nature, location of operation, product category matching degree, and basic production capacity level.

[0067] Historical cooperation characteristics are used to evaluate the stability of cooperation and the reputation of suppliers, including: frequency of supply of similar products, historical cooperation evaluation, supply stability, batch qualification rate, cooperation level and completion rate of rectification, etc.

[0068] Production and processing capabilities are used to evaluate a supplier's processing precision and mass production capabilities, including: automation level, maximum capacity, precision equipment level, tooling completeness, expedited response capability, and production stability.

[0069] Qualification and compliance features are used to provide rigid access constraints and manage compliance risks, including: three-system certification, special qualifications, technology recognition, credit rating, administrative penalties, and access qualification matching degree, etc.

[0070] R&D technical characteristics, used to adapt to customized R&D procurement needs, including: number of patents, proportion of R&D personnel, laboratory level, self-developed processes, experience in similar supporting technologies, and response speed for technology optimization, etc.

[0071] Fulfillment service characteristics are used to evaluate the supplier's delivery capabilities and after-sales cooperation, including: on-time delivery rate, delivery fluctuation coefficient, after-sales response time, exception handling cycle, payment terms capability, and collaboration score.

[0072] Operational risk characteristics are used for negative deductions and screening of high-risk supplier companies, including: number of administrative penalties, number of lawsuits, abnormal operations, debt-to-asset ratio, risk of supply disruption, and blacklist marking.

[0073] After obtaining the raw business data of each supplier company for the aforementioned dimensions, it is necessary to convert the raw data for each dimension into comparable scores. Optionally, in this embodiment, the scores for all evaluation dimensions are uniformly quantified into a range of 0 to 100 points. Differentiated single-dimensional calculation models are adopted for the data characteristics of different dimensions, as detailed below: Basic static dimension scores The scoring is based on a comprehensive assessment of factors including the company's years of operation, registered capital, company size, geographical matching between its operating location and the location of the procurement need, and the matching between the supplier's main product categories and the product categories required for the procurement. The product category matching degree is obtained by calculating the vector similarity between the supplier's main product categories and the product categories required for the procurement, and is used to measure the supplier's experience coverage in the product categories corresponding to this procurement need. (Basic static dimension score) The calculation formula is: = ·T years+ ·Scale+ • Geographic matching degree+ • Category matching degree.

[0074] Historical cooperation dimension score The scoring system primarily uses a deduction system, supplemented by a bonus system. When a supplier's supply stability is poor or their historical cooperation evaluation is low, the system deducts corresponding points; conversely, when a supplier's historical performance is excellent, the system awards corresponding bonus points. (Score based on historical cooperation dimension) The calculation formula is: =100-( Supply fluctuation coefficient + • Batch non-compliance rate + • Number of abnormal occurrences) + • Historical contract fulfillment rate.

[0075] Production and processing capability dimension score The evaluation is based on four dimensions: production line automation level, maximum capacity, batch stability, and expedited adaptation capability. The production and processing capacity dimension score is as follows: The calculation formula is: = Automation Level+ • Capacity+ Batch stability+ • Expedited adaptation capability.

[0076] Qualification and compliance dimension score The score is calculated using a combination of points awarded for qualifications and points deducted for penalties. For example, positive qualifications such as system certifications, specialized qualifications, and honorary qualifications earn points, while negative records such as administrative penalties deduct points. When a supplier has a major violation record, that dimension's score is directly deemed unqualified, specifically zero. (Qualification Compliance Dimension Score) The calculation formula is: =System certification score + Special qualification score + Honorary qualification score - Penalty deductions.

[0077] R&D technology dimension score The evaluation assesses the supplier's R&D capabilities and technology reserves, including the number of patents, the proportion of R&D personnel, laboratory level, and experience in similar R&D areas. R&D technology dimension score. The calculation formula is: = • Patent score + W52 • R&D personnel percentage + Laboratory Level+ • Similar R&D experience.

[0078] Performance service dimension score The calculation is based on the supplier's historical delivery and after-sales service records, including on-time delivery rate, after-sales response score, and business cooperation score. (Performance service dimension score) The calculation formula is: = On-time delivery rate + • After-sales response score + Business score.

[0079] Operational risk dimension score All scores are based on a negative deduction mechanism. Legal risks, financial risks, and operational irregularities of supplier companies will all incur corresponding deductions. When a supplier is blacklisted or has significant litigation involvement, the score for that dimension will be reduced to zero. (Operational Risk Dimension Score) The calculation formula is: =100 - (legal risk deduction + financial risk deduction + abnormal operation deduction).

[0080] In the above single-dimensional calculation formulas to These are the weights of sub-indicators within each dimension, used to differentiate the importance of different sub-indicators within the same evaluation dimension. Their values ​​can be pre-configured according to actual business needs, and the sum of the weights within each sub-dimension is 1.

[0081] Through the above calculations, each supplier in the candidate supplier pool obtains a quantitative score normalized to a range of 0 to 100 on each evaluation dimension. That is, each supplier corresponds to a set of seven-dimensional scoring vectors, which are used to calculate the final comprehensive matching score of each supplier.

[0082] 105. Obtain the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight value corresponding to each dimension feature. Different scenario types correspond to different dynamic weights.

[0083] In this embodiment, the scenario weight library in the procurement knowledge base pre-stores a set of weight values ​​corresponding one-to-one with each scenario type. Each set of weight values ​​contains the weight values ​​corresponding to each evaluation dimension. The sum of the weight values ​​of all evaluation dimensions within the same set of weight values ​​is always equal to 1. At least one evaluation dimension has a different weight value in the set of weight values ​​corresponding to different scenario types, so as to achieve differentiated configuration of evaluation focus under different procurement scenarios.

[0084] After outputting the target scene type in step 102, the label of the target scene type can be used as a query index to read the set of weight values ​​that match the scene label from the scene weight library as the target dynamic weight.

[0085] Optionally, the set of weight values ​​in the scenario weight library is generated by combining a benchmark weight template with scenario-specific adjustments. Specifically, a set of benchmark weight templates is pre-defined, in which the weight values ​​of each evaluation dimension can be set to uniform distribution or initially configured according to the general preferences of enterprise procurement.

[0086] Based on this, a corresponding weight value adjustment strategy is configured for each scenario type. According to the focus of the procurement task under the scenario type, the weight values ​​of each evaluation dimension in the benchmark weight template are adjusted differently. After adjustment, the sum of the weight values ​​of each evaluation dimension remains constant, for example, 1, thereby generating a set of weight values ​​corresponding to the scenario type.

[0087] As mentioned earlier, procurement scenarios can be categorized into mass production scenarios, expedited procurement scenarios, R&D customization scenarios, routine inventory preparation scenarios, and long-term supply scenarios. In mass production scenarios, the procurement task focuses on the supplier's ability to provide stable, large-volume supplies; therefore, the weighting of production and processing capabilities and historical cooperation is increased, while the weighting of R&D technology is reduced. In expedited procurement scenarios, the procurement task focuses on the supplier's rapid response and emergency support capabilities; therefore, the weighting of fulfillment services is increased, while the weighting of R&D technology is reduced. In R&D customization scenarios, the procurement task focuses on the supplier's technological R&D and process optimization capabilities; therefore, the weighting of R&D technology is increased, while the weighting of fulfillment services is reduced. In routine inventory preparation scenarios, the weighting of each evaluation dimension is evenly distributed to comprehensively assess the supplier's overall strength. In long-term supply scenarios, the procurement task focuses on the supplier's operational stability and long-term cooperation risks; therefore, the weighting of qualification compliance and operational risk dimensions is increased.

[0088] Based on this, in this embodiment, when obtaining the target dynamic weight corresponding to the target scenario type, a preset benchmark weight template is used as the basis. According to the weight value configuration strategy corresponding to the target scenario type, the weight values ​​of each evaluation dimension in the benchmark weight template are adjusted differentially. After adjustment, the sum of the weight values ​​of each evaluation dimension remains constant, thus generating the target dynamic weight. Through the above-mentioned differential adjustment, each scenario type corresponds to a unique set of weight values, enabling the weight configuration to accurately match the focus of the procurement task.

[0089] After the above processing, the target dynamic weight corresponding to the target scenario type can be obtained. The target dynamic weight is represented by a set of seven-dimensional weight vectors, the sum of the weight values ​​of each dimension is 1, and the relative size of the weight values ​​of each dimension reflects the degree of emphasis on different evaluation dimensions in the current procurement scenario.

[0090] 106. The quantitative scores of multiple dimensions for each supplier are weighted and summed with the weight values ​​in the target dynamic weights to obtain the comprehensive matching score for each supplier.

[0091] Specifically, the individual quantitative scores for each dimension calculated in step 104 for each supplier in the candidate supplier pool, i.e., the basic static dimension scores, are used. Historical cooperation dimension score Production and processing capacity dimension score Score in qualification and compliance dimension R&D technology dimension score Score for performance service dimension Score for operational risk dimension The weighted sum of the target dynamic weights for the corresponding dimensions obtained in step 105 is used to obtain the comprehensive matching score for each supplier.

[0092] The formula for calculating the overall matching score is as follows: = · ·+ · + · + · + · + · + · .

[0093] in, To generate dynamic target weights by reading from the scene weight library based on the target scene type. + + + + + + =1.

[0094] It should be noted that this embodiment uses seven feature dimensions as an example for illustration, but it is not limited to this. For example, the type and number of feature dimensions can be customized.

[0095] In this embodiment, by using differentiated weighting, the proportion of scores for key assessment dimensions in the current procurement scenario can be amplified, while the influence of irrelevant dimensions can be weakened, so that the overall score fits the preferences of each procurement business and achieves scenario-adaptive intelligent evaluation.

[0096] Through the weighted summation calculation described above, each supplier in the candidate supplier pool receives a comprehensive matching score. This score reflects the supplier's overall capability level across multiple evaluation dimensions in the current procurement scenario. Since different scenario types have different dynamic weight configurations, the comprehensive matching score of the same supplier will vary depending on the procurement scenario. For example, the same supplier may receive a higher comprehensive matching score in a customized R&D scenario than in an expedited procurement scenario, thus achieving an adaptive adjustment of the evaluation results according to the focus of the procurement task.

[0097] 107. Sort the comprehensive matching scores according to the scores and output the sorted list of qualified suppliers, as well as the recommendation reasons for each qualified supplier. The recommendation reasons include the multi-dimensional feature data and comprehensive matching score of each supplier.

[0098] Figure 5 This is a schematic diagram illustrating a supplier sorting selection method provided in an embodiment of this application, such as... Figure 5 As shown, after calculating the comprehensive matching score of each supplier, operations such as sorting, generating recommendation reasons, and outputting a list of qualified suppliers are performed.

[0099] In the specific implementation process, all suppliers in the candidate supplier pool are sorted in descending order according to their calculated comprehensive matching scores, forming a sorted list of candidate suppliers. A higher comprehensive matching score indicates a better match between the supplier's overall capabilities across all evaluation dimensions and the procurement requirements in the current procurement scenario. Based on the sorting results, a list of qualified suppliers is output in sorted order, displaying each supplier's ranking, comprehensive matching score, and detailed scores for each dimension.

[0100] Optionally, you can select the top N suppliers from the sorted candidate supplier list to generate a qualified supplier list.

[0101] After sorting, recommendation reasons are generated for the top-ranked qualified suppliers. Optionally, recommendation templates and compliance evaluation templates can be retrieved from the historical case library of the procurement expertise base, and combined with the matching scores of each evaluation dimension of the candidate supplier and the original field information of each data table in the supplier's business database to generate structured and traceable recommendation reasons.

[0102] Optionally, the recommendation reasons include criteria for product category matching, qualification and compliance, production capacity, performance service, and risk warnings. Each recommendation description is linked to original field data from the supplier's business database, thus preventing the use of fabricated information in the large language model.

[0103] Optionally, while outputting the list of qualified suppliers, the original text of this procurement requirement, semantic parsing results, target scenario type tags, the first constraint used for compliance filtering, detailed scores of each supplier in each dimension and comprehensive matching scores, final ranking results and recommendation reasons can all be written into the log database to achieve full-process traceability of supplier selection, so as to facilitate subsequent auditing and traceability and ensure that the decision-making process is transparent and controllable.

[0104] In one alternative implementation, after sorting the overall matching scores by score, semantic fine-tuning can also be performed on the sorted candidate supplier list.

[0105] Specifically, when there are two suppliers in the ranked candidate supplier list whose comprehensive matching scores differ by less than a preset threshold, the obtained procurement requirement text described in natural language is input into a preset embedding model and converted into a requirement semantic vector. Simultaneously, unstructured description texts of the first and second suppliers are retrieved from the supplier business database and input into the same embedding model, respectively converted into first supplier semantic vectors and second supplier semantic vectors. The unstructured description text includes at least one of the following: the supplier's company profile, product introduction, and process description.

[0106] Based on this, the semantic similarity between the demand semantic vector and each supplier semantic vector is calculated. Optionally, the semantic similarity is obtained by calculating the vector distance between the demand semantic vector and the supplier semantic vector, such as cosine distance.

[0107] Finally, the relative positions of the first and second suppliers in the sorted candidate supplier list are adjusted according to their semantic similarity: suppliers with higher semantic similarity are ranked first, and suppliers with lower semantic similarity are ranked last.

[0108] After semantic fine-tuning is completed, the overall matching score of each supplier involved in the fine-tuning and all scoring data except for the ranking position remain unchanged. That is, semantic fine-tuning only affects the relative ranking position between suppliers with similar overall matching scores, without changing any scoring results.

[0109] This solution introduces semantic vector-based similarity comparison as an auxiliary judgment criterion, which can compensate for the problem of insufficient identification of implicit semantic needs by pure quantitative scoring when structured scoring cannot distinguish suppliers with extremely small score differences.

[0110] In summary, the supplier evaluation and selection scheme based on large model semantic parsing and scene adaptation provided in this application can achieve at least the following beneficial effects: By constructing a quantitative evaluation index system covering multiple dimensions such as basic static indicators, historical cooperation, production and processing, qualification and compliance, R&D technology, performance services, and operational risks, the system achieves automated and standardized scoring of supplier capabilities, effectively avoiding biases from subjective human judgment. Simultaneously, by pre-setting dynamic weight templates corresponding to each procurement scenario and automatically retrieving and weighting the matching set of weight values ​​based on the target scenario type identified by the large language model, the evaluation focus can adaptively shift with the type of procurement task, ensuring that the evaluation system maintains a high degree of scientific rigor and objectivity across different business scenarios.

[0111] A progressive ranking logic combining rigid compliance filtering, multi-dimensional quantitative scoring, and semantic similarity fine-tuning is adopted. First, non-compliant suppliers with poor credit or abnormal qualifications are excluded from the candidate pool through the first constraint, ensuring the bottom line of procurement compliance. Second, objective capability matching evaluation is achieved through multi-dimensional quantitative scoring and scenario-adaptive weighting. Finally, through the semantic similarity fine-tuning mechanism, the ranking of adjacent suppliers that are difficult to distinguish through quantitative scoring is corrected by comparing the semantic vectors of unstructured descriptive text. This takes into account both explicit procurement requirements and implicit business needs, reduces the risk of incorrect or missed selections, and effectively improves the accuracy of supply and demand matching.

[0112] By storing the original business data of each supplier in a supplier business database, and binding each recommendation description to the original fields in the database when generating recommendation reasons, the recommendation results can not only be quantified and ranked, but also traced back to the original data source of each score, effectively overcoming the problem that traditional black-box models cannot explain the decision-making basis. Simultaneously, the original text of the procurement requirements, semantic parsing results, scene recognition results, screening rules, scoring details, and final recommendation results are all written to a log database, achieving full-process traceability of supplier selection, meeting the management requirements of enterprise procurement auditing and compliance management, and enhancing the transparency and credibility of the decision-making process.

[0113] The following will describe in detail one or more embodiments of the supplier evaluation and selection apparatus based on large model semantic parsing and scene adaptation of this application. Those skilled in the art will understand that these apparatuses can be configured using commercially available hardware components through the steps taught in this solution.

[0114] Figure 6 A schematic diagram of a supplier evaluation and selection device based on large model semantic parsing and scene adaptation provided in this application embodiment is shown below. Figure 6 As shown, the device includes: a construction module 11, a processing module 12, and an output module 13.

[0115] Module 11 is used to build a supplier business database and a procurement professional knowledge base. The supplier business database is used to store business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports supplier selection tasks in procurement scenarios.

[0116] Processing module 12 is used to input the acquired procurement requirements expressed in natural language into a large language model, and use the large language model to perform semantic normalization processing on the procurement requirements through the procurement professional knowledge base. It then parses the supplier constraints from the semantic normalization processing results and identifies the target scenario type to which the procurement requirements belong. Based on the first constraint in the supplier constraints, it performs product category filtering and compliance filtering on all suppliers to obtain a candidate supplier pool. For each supplier in the candidate supplier pool, based on the second constraint in the supplier constraints, it obtains multi-dimensional feature data corresponding to each supplier from the supplier business database and performs quantitative scoring. It obtains the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight values ​​corresponding to each dimension feature, with different dynamic weights corresponding to different scenario types. Finally, it performs a weighted sum of the quantitative scoring results of multiple dimensions corresponding to each supplier and the weight values ​​in the target dynamic weight to obtain a comprehensive matching score for each supplier.

[0117] Output module 13 is used to sort the comprehensive matching scores according to the scores and output the sorted list of qualified suppliers, as well as the recommendation reasons for each qualified supplier. The recommendation reasons include the multi-dimensional feature data and comprehensive matching scores of each supplier.

[0118] Figure 6 The device shown can perform the steps described in the foregoing embodiments. For detailed execution process and technical effects, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0119] In one possible design, the above Figure 6 The structure of the supplier evaluation and selection device based on large model semantic parsing and scene adaptation shown can be implemented as an electronic device, such as... Figure 7 As shown, the electronic device may include: a memory 21, a processor 22, and a communication interface 23. The memory 21 stores a computer program, which, when executed by the processor 22, enables the processor 22 to at least implement the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as provided in the foregoing embodiments.

[0120] The aforementioned memory 21 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0121] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium.

[0122] Accordingly, this application also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is able to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. In addition, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable supplier evaluation and selection device based on large model semantic parsing and scene adaptation, so that the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable supplier evaluation and selection device based on large model semantic parsing and scene adaptation can be implemented as a means to implement the corresponding functions in the above method embodiments.

[0123] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0124] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and software. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. This application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] Finally, it should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0126] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A supplier evaluation and selection method based on large model semantic parsing and scenario adaptation, characterized in that, include: Construct a supplier business database and a procurement professional knowledge base. The supplier business database is used to store business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports supplier selection tasks in procurement scenarios. The procurement requirements expressed in natural language are input into a large language model. The large language model is used to perform semantic normalization on the procurement requirements through the procurement professional knowledge base. The supplier constraints are parsed from the semantic normalization results, and the target scenario type to which the procurement requirements belong is identified. Based on the first constraint in the supplier constraints, product category filtering and compliance filtering are performed on all suppliers to obtain a candidate supplier pool. For each supplier in the candidate supplier pool, according to the second constraint in the supplier constraint conditions, the multi-dimensional feature data corresponding to each supplier is obtained from the supplier business database and quantitatively scored. Obtain the target dynamic weight corresponding to the target scenario type from the procurement professional knowledge base. The target dynamic weight includes the weight value corresponding to each dimension feature. Different scenario types correspond to different dynamic weights. The quantitative scoring results of multiple dimensions corresponding to each supplier are weighted and summed with the weight values ​​in the target dynamic weight to obtain the comprehensive matching score for each supplier. The comprehensive matching scores are sorted according to their values, and a list of qualified suppliers is output, along with a recommendation reason for each qualified supplier. The recommendation reason includes the multi-dimensional feature data and comprehensive matching score of each supplier.

2. The method according to claim 1, characterized in that, The method further includes: When the candidate supplier pool is sorted and a first supplier and a second supplier are found in the sorted candidate supplier list where the difference between their comprehensive matching scores is less than a preset threshold, the procurement requirement is converted into a requirement semantic vector, and the unstructured description texts of the first supplier and the second supplier obtained from the supplier business database are converted into supplier semantic vectors respectively. Calculate the semantic similarity between the demand semantic vector and each supplier semantic vector; Based on the semantic similarity, the relative positions of the first supplier and the second supplier in the sorted candidate supplier list are adjusted.

3. The method according to claim 2, characterized in that, The unstructured description text includes at least one of the supplier's company profile text, product introduction text, and process description text; the semantic similarity is obtained by calculating the vector distance between the demand semantic vector and the supplier semantic vector, and the higher the similarity value, the higher the supplier's ranking position.

4. The method according to claim 1, characterized in that, The procurement expertise base includes: a standard terminology base, an evaluation rule base, a scenario weight base, a procurement common sense base, and a historical case base; When the large language model performs semantic normalization on the procurement requirements, it calls the standard terminology library to perform term mapping and ambiguity elimination on the procurement requirements. When parsing the supplier constraints from the semantic normalization processing result, the evaluation rule base and the procurement common sense base are called to determine the constraints used to evaluate supplier compliance; When obtaining the target dynamic weight corresponding to the target scene type, the set of weight values ​​matching the target scene type is read from the scene weight library as the target dynamic weight; When generating the recommendation reason, the recommendation statement template and compliance evaluation template in the historical case library are called to generate a standardized recommendation reason and compliance evaluation description.

5. The method according to claim 1, characterized in that, The target scene type is identified in the following ways: The large language model extracts scenario determination information from the semantic normalization processing result. The scenario determination information includes at least one of the following: the urgency of delivery, the degree of product customization, and the purchase batch size. The large language model will match the extracted scenario judgment information with multiple scenario judgment conditions preset in the procurement professional knowledge base, and output the successfully matched scenario tags; the scenario tags include mass production scenarios oriented towards mass production stability, expedited procurement scenarios oriented towards delivery timeliness, and R&D customization scenarios oriented towards technical capabilities.

6. The method according to any one of claims 1 to 5, characterized in that, The procurement expertise base contains a set of weight values ​​that correspond one-to-one with each scenario type. Each set of weight values ​​contains the weight values ​​corresponding to each evaluation dimension. The sum of the weight values ​​of all evaluation dimensions in the same set of weight values ​​is always equal to 1. At least one evaluation dimension has a different weight value in the set of weight values ​​corresponding to different scenario types. When obtaining the target dynamic weight corresponding to the target scene type, based on the preset benchmark weight template, the weight values ​​of each evaluation dimension in the benchmark weight template are adjusted differently according to the weight value configuration strategy corresponding to the target scene type. After adjustment, the sum of the weight values ​​of each evaluation dimension remains constant, and the target dynamic weight is generated.

7. A supplier evaluation and selection device based on large model semantic parsing and scenario adaptation, characterized in that, include: The module is used to build a supplier business database and a procurement professional knowledge base. The supplier business database is used to store business capability data of supplier companies, and the procurement professional knowledge base is used to store domain knowledge that supports supplier selection tasks in procurement scenarios. The processing module is used to input the procurement requirements expressed in natural language into a large language model, and use the large language model to perform semantic normalization processing on the procurement requirements through the procurement professional knowledge base. The module then parses the supplier constraints from the semantic normalization processing results and identifies the target scenario type to which the procurement requirements belong. Based on the first constraint in the supplier constraints, the module performs product category filtering and compliance filtering on all suppliers to obtain a candidate supplier pool. For each supplier in the candidate supplier pool, according to the second constraint in the supplier constraint conditions, the multi-dimensional feature data corresponding to each supplier is obtained from the supplier business database and quantitatively scored; the target dynamic weight corresponding to the target scenario type is obtained from the procurement professional knowledge base, and the target dynamic weight includes the weight value corresponding to each dimension feature, with different dynamic weights corresponding to different scenario types. The quantitative scoring results of multiple dimensions corresponding to each supplier are weighted and summed with the weight values ​​in the target dynamic weight to obtain the comprehensive matching score for each supplier. The output module is used to sort the comprehensive matching scores according to the scores and output the sorted list of qualified suppliers, as well as the recommendation reasons for each qualified supplier. The recommendation reasons include the multi-dimensional feature data and comprehensive matching scores of each supplier.

8. An electronic device, characterized in that, include: The system includes a memory, a processor, and a communication interface; wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor of an electronic device, causes the processor to perform the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, include: A computer program or instruction that, when executed by a processor of an electronic device, causes the processor to perform the supplier evaluation and selection method based on large model semantic parsing and scene adaptation as described in any one of claims 1 to 6.