An industry chain precise investment promotion matchmaking method based on large-middle-small model cooperation

Abstract

The application discloses an industry chain precise investment promotion matching method based on large-middle-small model cooperation, comprising: acquiring multi-source industry chain related data and performing cleaning and feature extraction; judging the task complexity through an intermediate scheduling model, and distributing the task to a large model or a small model for cooperative execution, wherein the large model is used for industry chain structure analysis, gap identification or risk assessment, and the small model is used for information query, preliminary matching calculation or on-site assistance; constructing industry chain relationship data based on the model cooperative execution result, and performing periodic or triggered updating; performing dynamic weight matching calculation based on dimensions such as industry chain correlation strength, policy adaptation degree, resource matching degree, development stage adaptation degree and risk controllability, generating an investment promotion matching result and outputting the same. The method can ensure real-time response, realize dynamic analysis and precise matching of the industry chain, and improve the accuracy and adaptability of investment promotion matching.

Description

Technical Field

[0001] This invention relates to the fields of intelligent investment promotion, industrial chain analysis, and data processing technology of multi-model collaboration in artificial intelligence. Specifically, it relates to a method for analyzing, matching, and dynamically updating multi-source data related to the industrial chain through the collaborative work of a large model, an intermediate scheduling model, and a small model, thereby achieving precise investment promotion and matching within the industrial chain. This method is applicable to application scenarios such as investment promotion in local industrial parks, cross-regional industrial cooperation zones, and industrial platforms. Background Technology

[0002] With the continuous segmentation of industrial structures and the intensification of inter-regional industrial competition, targeted investment attraction based on the industrial chain has gradually become one of the core methods of investment promotion. In the actual investment attraction process, it is often necessary to comprehensively consider multi-dimensional information such as the upstream and downstream structure of the industrial chain, key technological links, park resources, policy environment, and the company's own development stage and risk status.

[0003] In existing technologies, one type of solution typically relies on a single type of model or processing module to analyze and match industry chain data. For example, it may use only a general-purpose large model for unified processing, or rely solely on a rule engine or lightweight model for matching calculations. When faced with scenarios involving large-scale data, multiple dimensions, and high analytical depth requirements, this type of solution often suffers from high computational resource consumption, high response latency, or superficial analysis results.

[0004] Another approach involves centrally integrating park resource data, enterprise data, and policy data, and then making matching decisions based on manually preset rules or static weights. This type of approach struggles to dynamically adjust to changes in the industrial chain structure, enterprise needs, and investment attraction objectives, resulting in insufficient real-time performance and adaptability of the matching results.

[0005] In addition, some industrial chain analysis tools assist investment promotion decisions by constructing static industrial chain structures, but their data update cycle is long and it is difficult to reflect dynamic changes such as technological iteration and supply chain relationship adjustments in a timely manner, which limits their application effect in the actual investment promotion process.

[0006] Therefore, there is an urgent need for a business matching method that can ensure processing efficiency while also taking into account complex analysis capabilities, and supporting dynamic updates of industry chain relationships and multi-dimensional matchmaking evaluation. Summary of the Invention

[0007] The purpose of this invention is to provide a precise investment promotion and matching method for the industrial chain based on the collaboration of large, medium and small models. By constructing a processing architecture in which large models, intermediate scheduling models and small models work together, the invention performs hierarchical analysis and processing of multi-source data related to the industrial chain, realizes the dynamic construction and updating of industrial chain relationships, and generates investment promotion and matching results based on a matching algorithm with multi-dimensional dynamic weights, thereby improving the accuracy, real-time performance and adaptability of investment promotion and matching.

[0008] To achieve the above objectives, this invention, based on existing investment promotion and matchmaking information processing technologies, introduces a multi-model collaborative processing mechanism and, combined with the needs of dynamic analysis of the industrial chain, proposes the following technical solution: A method for precise investment attraction and matchmaking based on the collaboration of large, medium, and small models in the industrial chain includes: acquiring industrial chain-related data from multiple data sources, and cleaning, de-identifying, and extracting features from the data; allocating different tasks to large or small models for collaborative execution through an intermediate scheduling model based on task complexity and processing time requirements; constructing dynamic industrial chain relationship data based on the collaborative execution results of the models; and performing dynamic weight matching calculations on target objects based on multiple dimensions such as industrial chain correlation strength, policy adaptability, resource matching degree, development stage adaptability, and risk controllability to generate investment attraction and matchmaking results.

[0009] In the above technical solution, the small model is used to undertake investment promotion assistance or information query tasks with high response time requirements. The response delay of the processing results provided to the outside world is lower than a preset threshold to meet the real-time interaction requirements in the investment promotion process. The preset threshold can be set according to the actual deployment environment, preferably 300ms.

[0010] In the above technical solution, the large model is used to undertake complex analysis tasks such as industrial chain structure analysis, industrial chain gap identification, or risk assessment. Its parameter scale is greater than a preset threshold to improve the accuracy of industrial chain relationship modeling and complex correlation analysis. The preset threshold can be adjusted according to computing resource conditions, preferably greater than 5 billion parameters.

[0011] Compared with the prior art, the present invention has at least the following beneficial effects: Through a collaborative processing architecture of large, medium, and small models, in-depth analysis of the industrial chain is achieved while ensuring real-time response capabilities; by dynamically constructing and updating industrial chain relationship data, the responsiveness of the matching results to changes in the industrial chain is improved; and through a multi-dimensional dynamic weighting matching mechanism, the accuracy and adaptability of investment matching results are enhanced. Detailed Implementation

[0012] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0013] I. Collection and Preprocessing of Multi-Source Industrial Chain Data In this embodiment, industry chain-related data from multiple data sources are first acquired, and the industry chain-related data includes at least the following categories: 1) Supply chain data, including information on upstream raw material suppliers, midstream core manufacturing enterprises, and downstream application enterprises, as well as data reflecting the technological evolution of the supply chain, such as the distribution of patent types, the number of new patents each year, and changes in technology routes; 2) Park data, including the supply of land and factory buildings in the park (such as area and plot ratio), the composition of existing enterprises in the park, the list of supporting enterprises, infrastructure conditions, investment promotion policy documents (such as tax incentive period and subsidy amount), and talent reserve. 3) Enterprise data, including enterprise registration information, business scope, registered capital, financial indicators, core technology direction, capacity expansion needs and investment preferences; 4) Dynamic data, including industry public opinion information, policy updates, corporate cooperation announcements, and supply chain relationship changes.

[0014] The data was preprocessed, including removing duplicate, outlier, and missing data; and desensitizing sensitive information such as core business data and patent parameters. Based on this, feature extraction was performed on the cleaned data, converting textual data into semantic vectors and normalizing structured data to create feature data in a unified format for subsequent model processing and analysis.

[0015] II. Collaborative Architecture and Division of Labor for Large, Medium, and Small Models In this implementation, a three-level model collaborative architecture consisting of a large model, an intermediate scheduling model, and a small model is constructed.

[0016] 1) The large-scale model, serving as the core analytical model, is deployed in the cloud to perform in-depth industry chain analysis tasks. This large-scale model can perform detailed analysis of the industry chain structure, identify key links and technological barriers in the industry chain, and infer potential enterprise needs and industry chain risks based on enterprise technical information and dynamic data.

[0017] 2) The small model, as a real-time execution model, is deployed on edge or front-end devices to perform tasks with high timeliness requirements, such as policy information query, preliminary matching score calculation, and on-site data collection and processing. The small model is obtained by knowledge distillation of the output results of the large model and has the characteristics of small parameter size and fast response speed.

[0018] 3) The intermediate scheduling model is used for task scheduling and data collaboration between the large and small models. Based on the data size, analytical depth requirements, and response time requirements of the tasks to be processed, this model assesses the complexity of the tasks and allocates them to either the large or small model for execution. Simultaneously, the intermediate scheduling model compresses the analytical results output by the large model into structured data for use by the small model, and feeds back real-time data collected by the small model during execution to the large model, thus forming a data loop.

[0019] III. Task Scheduling and Collaborative Processing Mechanism In the specific implementation process, when a demand for investment promotion or an enterprise's investment intention is received, the intermediate scheduling model first analyzes the task and judges the complexity of the task based on the data scale involved, the number of analytical dimensions required, and the response time requirements.

[0020] For complex tasks involving supply chain structure analysis, supply chain gap identification, and risk assessment, large models perform the tasks; for tasks involving information retrieval, result display, or on-site assistance, small models perform the tasks. An intermediate scheduling model handles data transfer and result integration between different models, thereby enabling collaborative model processing.

[0021] IV. Construction and Updating of Dynamic Supply Chain Relationship Data Based on the analysis results obtained from the collaborative execution of the model, industrial chain relationship data is constructed to characterize the industrial chain structure. The industrial chain relationship data includes at least industrial chain link nodes, enterprise nodes, park nodes, and policy nodes, and records the supporting relationships, supply relationships, and policy adaptation relationships between each node.

[0022] The industry chain relationship data is maintained using a combination of periodic and triggered updates. Periodic updates are used to synchronously add new data within a preset time period; triggered updates are used to update the relevant industry chain relationship data when changes in industry chain-related data, changes in enterprise cooperation relationships, or changes in matching results exceed a preset threshold, so as to reflect the dynamic changes in the industry chain structure.

[0023] V. Matching Calculation Based on Multi-Dimensional Dynamic Weights In this implementation, multiple matching dimensions are defined, including the strength of industrial chain linkage, policy compatibility, resource matching, development stage compatibility, and risk controllability. The intermediate scheduling model dynamically adjusts the weights of these matching dimensions based on the investment attraction targets and the life cycle stage of the industrial chain.

[0024] By weighting and calculating each matching dimension, a matching score is generated between the target enterprise and the industrial park, and a list of investment matching results is generated based on the matching score. The dynamic weighting mechanism can be adjusted according to the development stage of the industrial chain and changes in investment targets, thereby improving the adaptability of the matching results.

[0025] VI. Matching Result Output and Feedback Iteration The generated investment promotion results are output in report form, including matching scores, core matching factors, and investment promotion recommendations. During the investment promotion process, a small model collects feedback information from enterprises gathered on-site by investment promotion personnel, and this feedback information is transmitted to the intermediate scheduling model and the large model to update the industrial chain relationship data and optimize subsequent analysis processes, thus forming a continuously iterative closed-loop mechanism.

[0026] In a preferred embodiment, the small model adopts a model structure with relatively small parameter size and short inference path to reduce the amount of computation required for a single inference, thereby controlling its response latency within 300ms in the actual operating environment, ensuring that investment promotion personnel can obtain auxiliary decision-making information in a timely manner in on-site or online interactive scenarios.

[0027] In a preferred embodiment, the large model employs a high-parameter deep learning model structure for joint modeling and deep analysis of multi-source industry chain data. By increasing the model's parameter scale, the model can express more complex industry chain relationships, thereby improving the accuracy and stability of industry chain matching results. The aforementioned parameter scale is not a limitation of the present invention, and those skilled in the art can make reasonable adjustments according to actual application scenarios.

Claims

1. A method for precise investment attraction and matchmaking in the industrial chain based on the synergy of large, medium, and small models, characterized in that, Includes the following steps: S1. Obtain industry chain-related data from multiple data sources, including at least enterprise data, park resource data, policy data, and dynamic data reflecting changes in the industry chain; S2. Perform data cleaning and feature extraction processing on the industry chain-related data to transform data from different sources and with different structures into feature data in a unified format. S3. The complexity of the task to be processed is judged by the intermediate scheduling model, and the task is assigned to the large model or the small model for collaborative execution according to the judgment result. The large model is used to perform complex analysis tasks such as industrial chain structure analysis, industrial chain gap identification or risk assessment; the small model is used to perform information query, preliminary matching calculation or on-site auxiliary tasks with high response time requirements. S4. Based on the collaborative execution results of the large model and the small model, construct industrial chain relationship data to characterize the relationship between industrial chain links, enterprises, parks and policies, and update the industrial chain relationship data periodically or triggeredly. S5. Based on multiple preset matching dimensions, perform dynamic weight matching calculations on the target objects, generate investment matching results, and output them.

2. The method according to claim 1, characterized in that: The multiple data sources include at least one external data source and at least one internal data source. The external data source includes information on upstream and downstream enterprises in the industrial chain, industry public opinion information, or enterprise cooperation announcements. The internal data source includes park resource allocation data or investment promotion policy data.

3. The method according to claim 1, characterized in that: The data cleaning includes at least the data processing operations of removing duplicate data, abnormal data and missing fields, and the desensitization processing of sensitive information involving enterprise business data or technical data.

4. The method according to claim 1, characterized in that: The complexity determination is based on at least one or more of the following factors: the data size of the data to be processed, the number of data dimensions, and the timeliness requirements of the task response.

5. The method according to claim 1, characterized in that: The intermediate scheduling model is used to transfer data between the large model and the small model, compress the analysis results output by the large model into structured data for the small model to call, and feed back the real-time data collected by the small model during execution to the large model to form a data closed loop.

6. The method according to claim 1, characterized in that: The industrial chain relationship data includes at least the nodes of the industrial chain links, enterprise nodes, park nodes, and policy nodes, and records the supporting relationships, supply relationships, or policy adaptation relationships between the nodes.

7. The method according to claim 6, characterized in that: The trigger-based update is triggered when any of the following conditions are detected: changes in industry chain-related data, changes in enterprise cooperation relationships, or changes in matching results exceeding a preset threshold.

8. The method according to claim 1, characterized in that: The small model is a lightweight model used to perform real-time response tasks. Its response latency for providing query or auxiliary results is lower than a preset threshold, which is preferably 300ms.

9. The method according to claim 1, characterized in that: The large model is a high-parameter model used to perform supply chain structure analysis or risk assessment. Its parameter scale is greater than a preset threshold, which is preferably 5 billion parameters.

10. The method according to claim 1, characterized in that: The investment promotion results are output in the form of a report, which includes the matching score of the target object, the core matching factors, and investment promotion suggestions.