Search

CN-121981796-A - Supplier intelligent recommendation matching method based on large language model

CN121981796ACN 121981796 ACN121981796 ACN 121981796ACN-121981796-A

Abstract

The invention provides a large language model-based intelligent supplier recommendation matching method, which aims to solve the problems of insufficient semantic understanding and rough matching result of the traditional supply and demand matching method. The method comprises the steps of receiving product demand description in a standard natural language form, adopting a large model based on a demand side industry knowledge base and demand characteristics to extract a special algorithm and a rule, extracting multidimensional demand characteristics and converting the multidimensional demand characteristics into a structured demand attribute label set, integrating supplier multidimensional data, extracting semantic information, a potential service mode and a check structured index through the large model based on a supply side industry knowledge base and a capacity index quantization special algorithm and rule, outputting a supplier attribute label set, converting the two models into uniform dimension value vectors, calculating scores through a vector similarity algorithm, dividing matching grades based on a preset threshold value, and outputting layering results with the matching grades and decision suggestions. The invention can realize the accurate matching of the deep attributes of supply and demand.

Inventors

  • ZHANG QI
  • CHEN SHUANG
  • LUO FAN
  • SUN CHUANGCHUANG
  • MIN PENGFEI

Assignees

  • 中国电子科技集团公司第十五研究所

Dates

Publication Date
20260505
Application Date
20251225

Claims (10)

  1. 1. The intelligent provider recommendation matching method based on the large language model is characterized by comprising the following steps of: Receiving a product demand description submitted by a demand issuing unit in a standard natural language, wherein the demand description covers a technical adaptation requirement, a product accessibility target, a performance capability standard and response liveness; a large model constructed based on a special algorithm and a special rule for extracting the demand side industrial knowledge base and the demand characteristics is used as a modeling tool, the multi-dimensional demand characteristics are extracted according to the product demand description, and a structured demand attribute label set is output through reasoning and normalization processing, so that a product demand model is formed; Integrating relevant data of each provider, wherein the relevant data comprises enterprise basic information, production product information, intellectual property information, equipment facility information, historical contract information, social information record and production and detection equipment facility data; extracting semantic information and potential business modes from the related data by adopting a large model constructed based on a supply side industry knowledge base and a capacity index quantization special algorithm and rule, and outputting a supplier attribute tag set to form a supply capacity model corresponding to each supplier; the method comprises the steps of converting a product demand model and a plurality of supply capacity models into m-dimensional value vectors, wherein m is the feature quantity, calculating similarity scores Si of the product demand vectors and the supply capacity vectors by adopting a vector similarity matching algorithm, classifying matching grades based on preset different similarity thresholds, identifying potential supplier sets of corresponding grades, and outputting layered matching results with matching grades and decision suggestions.
  2. 2. The method of claim 1, wherein the large models are incorporated into industry dynamic knowledge updating mechanisms to absorb new technical standards, regulatory guidelines, and compliance requirements of corresponding fields in real time and update synchronously for the demand side industry knowledge base and the supply side industry knowledge base.
  3. 3. The method of claim 1, wherein the product demand description is filled in by a demand issuing entity according to a preset template specification, the template comprising demand core element prompts and uniform format requirements.
  4. 4. The method of claim 1, wherein the structured demand attribute tags include at least a location of a publication entity, a demand response time, a project domain feature, a demand type, a core function requirement, a performance parameter index, a specification model, a special qualification, a budget span, a lead time requirement, a delivery location.
  5. 5. The method of claim 1, wherein the vendor attribute tags include at least social security number, registered capital, business scope, industry qualification certificate, underwriting scope, core product/service performance parameters, core function modules, historical order status, performance indicators, technical specifications, intellectual property.
  6. 6. The method of claim 1, wherein the capability index quantization specific algorithms and rules include a qualification checking algorithm, a capacity threshold decision rule, a historical performance rate quantization rule.
  7. 7. The method of claim 1, wherein the vector similarity matching algorithm is a cosine similarity algorithm.
  8. 8. A large language model based vendor intelligent recommendation matching device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the large language model based vendor intelligent recommendation matching method as claimed in any one of claims 1 to 7.
  9. 9. A computer-readable storage medium, wherein a program for implementing information transfer is stored on the computer-readable storage medium, and the program when executed by a processor implements the steps of the large language model-based vendor intelligent recommendation matching method according to any one of claims 1 to 7.
  10. 10. A computer program product comprising a computer program which when executed by a processor implements the steps of the large language model based vendor intelligent recommendation matching method of any one of claims 1 to 7.

Description

Supplier intelligent recommendation matching method based on large language model Technical Field The invention relates to the technical field of supply chain management and resource allocation, in particular to a large language model-based intelligent supplier recommendation matching method. Background In the modern supply chain management and resource allocation process, accurate and efficient matching between a demander and a supplier is realized, and the method is a key link for improving the market response speed, reducing the operation cost and optimizing the resource utilization efficiency. With the development of global division and digital economy, the number of supply chains is explosive, and the product demand is increasingly characterized by individualization, complexity and dynamics, which puts higher demands on supply and demand matching technology. How to quickly identify the partner most meeting the requirements from mass provider resources is an important problem for enterprises to promote core competitiveness. At present, the supply and demand matching problem is mostly processed by adopting a statistical class method of multi-criterion decision, a semantic class method of keyword matching and the like. The multi-criterion decision method is characterized in that evaluation indexes such as price, delivery period, quality grade, capacity scale and the like are preset through the core dimension focused by a carding demand party, and comprehensive scoring and sorting are carried out on suppliers by combining manually set index weights; the keyword matching formula takes core words in the requirement description as retrieval basis, and potential cooperative objects with information intersection are screened out from a provider database in an accurate or fuzzy matching mode, or candidate providers are screened out through keyword similarity scores. Both types of methods play an important role in supply and demand matching scenes. In the face of increasingly complex demand descriptions and massive and diverse supplier information, the conventional supply and demand matching methods such as multi-criterion decision-making, keyword matching and the like are difficult to meet the current refined and intelligent matching demands: (1) The semantic understanding capability is limited, the method depends on rule matching or keyword similarity of a surface layer, the complex semantics, context logic and deep analysis capability of industry professional terms in unstructured data such as product demand description are lacked, misunderstanding or deviation of real demands are easy to cause, and meanwhile, the matching relationship of supply and demand parties on deep attributes such as technical specifications, service capability and quality standards cannot be accurately captured. (2) The matching result is rough, the output result of the traditional method is often single, the binary result which is 'in line' or 'out of line' is usually given, or the matching score is obtained through calculation, the refined grading of the matching degree of supply and demand is lacking, meanwhile, corresponding differentiated analysis, risk prompt, decision proposal and the like cannot be provided for different matching grades, and the scientific and efficient supplier screening and cooperation strategy is difficult to support for the demander. Disclosure of Invention In order to solve the problems of insufficient semantic understanding and rough matching result of the traditional supply and demand matching method, the invention provides a large language model-based intelligent supplier recommendation matching method, which aims to capture the matching relation of deep and fine-grained demand attributes of supply and demand parties in technical specifications, service capacity and the like, and quickly and accurately find the most suitable supplier for specific product demands. The provider intelligent recommendation matching method based on the large language model provided by the embodiment of the invention comprises the following steps: Receiving a product demand description submitted by a demand issuing unit in a standard natural language, wherein the demand description covers a technical adaptation requirement, a product accessibility target, a performance capability standard and response liveness; a large model constructed based on a special algorithm and a special rule for extracting the demand side industrial knowledge base and the demand characteristics is used as a modeling tool, the multi-dimensional demand characteristics are extracted according to the product demand description, and a structured demand attribute label set is output through reasoning and normalization processing, so that a product demand model is formed; Integrating relevant data of each provider, wherein the relevant data comprises enterprise basic information, production product information, intellectual property information, equipment f