A method for matching science and technology policy with characteristics of government and enterprise

Abstract

This invention relates to the field of science and technology policy services and intelligent information matching technology, specifically to a government-enterprise characteristic-coupled science and technology policy matching method. This method includes policy data structuring and dynamic policy graph construction, enterprise multi-dimensional characteristic system construction, deep coupling matching, dynamic updating and feedback optimization, and matching result analysis and application suggestion generation. By synchronizing policy changes in real time via API, optimizing enterprise characteristic weights using gradient descent, and integrating transparent judgment of application conditions and rights requirements with targeted application guidance, it addresses issues such as lagging adaptation of static policy databases, one-sided matching based on single dimensions, lack of logical transparency, lack of guidance, and decreased model adaptability. This achieves deep and accurate matching between policies and enterprises, dynamic adaptation across the entire chain, and transparent and traceable logic, improving matching accuracy and scenario adaptability, lowering application thresholds, helping enterprises obtain effective policies, and improving implementation efficiency.

Description

Technical Field

[0001] This invention relates to the field of science and technology policy services and intelligent information matching technology, specifically to a science and technology policy matching method that couples government and enterprise characteristics. Background Technology

[0002] The following technical solutions are currently mainly adopted in the field of matching science and technology policies with enterprises: 1. Keyword-based search solution: This solution relies on literal matching between the core terminology of the policy text and the enterprise registration information, and determines policy suitability based on the degree of string overlap; 2. Single-dimensional policy comparison approach: This approach focuses only on superficial characteristics such as the applicable industry sectors and funding amounts of policies, ignoring key matching factors such as application conditions, implementation period, and supporting requirements. 3. Static policy database matching scheme: The model is built using a fixed policy text database. There is no dynamic policy update mechanism, which cannot capture changes such as policy clause revisions, new policy releases, and expired policies in a timely manner. It also does not adjust the matching strategy according to the dynamic changes in the company's operating status and R&D capabilities.

[0003] The existing technology has significant core problems: 1. Low matching accuracy: It only captures literal or surface-level feature overlap, making it difficult to identify effective matching combinations such as "semantic compatibility of policy clauses but differences in terminology" and "company dynamic characteristics meeting the implicit requirements of policies". This can easily lead to companies missing out on applicable policies or applying for policies that do not meet the requirements. 2. Policy adaptation lag: The static policy database cannot synchronize policy updates in real time, making it difficult for enterprises to obtain the latest applicable policies; at the same time, it does not track changes in enterprise characteristics and cannot push new applicable policies in a timely manner. 3. The matching logic is not transparent: It only outputs a binary result of "suitable" or "unsuitable", which cannot locate the specific matching basis or explain the reason for the unsuitability, making it difficult for enterprises to adjust their application strategies accordingly; 4. Weak scenario adaptability: The policy needs of enterprises of different sizes and development stages are not differentiated, and a uniform matching standard is used, which leads to misjudgments such as small and micro enterprises being matched with policies exclusive to large enterprises, and start-ups being matched with policies suitable for mature enterprises. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention proposes a science and technology policy matching method that couples government and enterprise characteristics, thereby achieving precise matching between policies and enterprises, dynamic updating of policy and enterprise characteristics, transparent presentation of matching logic, adaptation and optimization for differentiated scenarios, and targeted guidance for application strategies.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for matching science and technology policies that couples government and enterprise characteristics includes the following steps: S1. Policy Data Structuring and Dynamic Policy Graph Construction: Structured analysis of multi-source science and technology policy data to construct a dynamic policy graph that supports real-time updates; S2. Construction of a multi-dimensional feature system for enterprises: Extract dynamic features from multi-source enterprise data and construct a multi-dimensional feature system covering basic, operational, R&D, and qualification aspects; S3. Deep Coupling Matching: A two-level matching strategy of "policy map coarse screening - element feature fine matching - fusion judgment" is adopted to achieve deep coupling matching between policies and enterprises; S4. Dynamic Updates and Feedback Optimization: Based on policy changes, changes in enterprise characteristics, and feedback on matching results, a closed-loop optimization mechanism is constructed; S5. Matching Result Analysis and Application Suggestion Generation: Generates a matching result analysis report that includes the basis for matching, score breakdown, and application suggestions.

[0006] Further, step S1 includes the following steps: S11. Policy Data Collection and Preprocessing: Collect policy texts from multiple sources, clean up redundant content, and standardize them into text fragments; S12. Structural extraction of policy elements: The Policy-BERT model is used to extract three types of elements: basic elements, application condition elements, and rights and interests constraint elements. S13. Dynamic Policy Graph Construction: Modeling with elements as nodes and relationships as edges, connecting to official interfaces to synchronize policy additions / revisions / expirations in real time, and updating the graph; S14. Assigning weights to policy elements: Determine the initial weights using the analytic hierarchy process (AHP).

[0007] Further, step S2 includes the following steps: S21. Enterprise Feature Data Collection: Collect multi-source data including business registration, finance, R&D, qualifications, and historical applications; S22. Enterprise Feature Structuring Processing: Standardized conversion, basic feature mapping to category / regional codes, operational feature classification into levels, R&D feature statistics on patent quantity, qualification feature status marking; S23. Dynamic updates of enterprise characteristics: Automatic updates are triggered by a mechanism, and official data is synchronized monthly to ensure timeliness; S24. Assigning weights to enterprise features: Based on historical matching contribution and enterprise development stage, the weights are optimized using gradient descent.

[0008] Further, step S3 includes the following steps: S31. Preliminary screening of policy map: Screening a preliminary set of suitable policies based on three criteria: regional suitability, type suitability, and timeliness suitability; S32. Fine-grained matching of feature elements: Application criteria fit: Calculate the scores of numerical and categorical elements, and sum them by weight to obtain the result; Rights and interests matching: mapping enterprise needs with policy rights and interests, scoring the matching degree by type / scale; S33. Fusion determination: Weighted fusion of two types of fit, classified according to threshold.

[0009] Further, step S4 includes the following steps: S41. Dynamic updates to the policy graph: Connects to the API of the science and technology department to synchronize policy changes in real time and automatically update graph nodes / relationships; S42. Dynamic Update of Enterprise Characteristics: The trigger mechanism automatically updates characteristics and regularly synchronizes and corrects deviations monthly. S43. Matching Model Feedback Optimization: Adjust the element weights based on enterprise feedback and retrain the model every quarter.

[0010] Further, step S5 includes the following steps: S51. Matching Criteria: Link policy graph nodes with enterprise characteristics to pinpoint specific reasons for suitability / missuitability; S52. Multi-dimensional score breakdown: Displays application requirements, rights and interests needs, and final matching score, explaining the contribution of each dimension; S53. Generation of application suggestions: Highly adaptable: Outputs timelines, bills of materials, and process guidelines; Conditions to be improved: The output needs to include a feature list and improvement paths; Incompatibility: Cause of core output obstacle.

[0011] Furthermore, in step S12, during the construction of the dynamic policy map, the policy map is dynamically updated by connecting to the API of the science and technology management department. New policies automatically trigger element extraction and are processed through formulas. When establishing element relationships, corresponding element attributes are updated and weights are adjusted during policy revisions. Element extraction employs a combination of sequence labeling and relationship classification. The formula for determining element relationships is as follows:

[0012] in, , To extract policy elements, , The BERT hidden layer output for the elements, , For trainable parameters, The weight matrix for the relation classification task. For the bias term in relation classification tasks, This refers to the type of relationship between elements.

[0013] Furthermore, in step S32, the enterprise characteristic update trigger mechanism includes automatic updates upon new qualification certifications, patent applications, or the release of financial statements, using a formula. Calculate the fit between the new feature and the policy threshold, and regularly synchronize official data monthly to ensure the timeliness of the feature. The formula is:

[0014] in, These are enterprise characteristic values. As a threshold for policy elements, The fit is for numerical features, ranging from [0, 1].

[0015] Furthermore, in step S32, the gradient descent method is used to optimize the enterprise feature weights during the matching model feedback optimization, combined with the formula... The contribution of each element in the application criteria is adjusted, and the model parameters are retrained quarterly based on feedback data. The formula for calculating the fit of the application criteria is as follows:

[0016] in: For the first The weighting of the elements in the application requirements for each policy item. For the first The suitability of each element.

[0017] Furthermore, in step S33, the matching result parsing is performed using the formula... Decomposition of application conditions and their suitability Alignment with rights and interests To enhance the transparency of the matching logic, the contribution of each dimension to the final score is clearly shown. The formula is as follows:

[0018] in: To incorporate weights, the values ​​range from [0, 1]. When science and technology management departments conduct policy dissemination... When enterprises submit their own plans Set matching threshold ,like If it is deemed a high-quality policy fit, the company is recommended to apply; if If the policy conditions need improvement, the required enterprise characteristics will be output; if If so, it is determined to be an incompatible policy, and the core reasons for the incompatibility are output.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly Improved Matching Accuracy: This invention achieves deep coupling between "policy semantics and enterprise characteristics" by constructing a policy element map and an enterprise dynamic feature system. The enterprise feature system integrates multi-dimensional real-time data, and the coupling of the two can identify implicit policy requirements and dynamic adaptation points of enterprises. This solves the problem of missed or false judgments caused by existing technologies that rely solely on literal or surface-level matching, and adapts to the differentiated requirements of policies at different levels in different fields.

[0020] 2. Outstanding Dynamic Adaptation Capabilities: This invention achieves end-to-end dynamic adaptation through real-time synchronization of the policy map, dynamic updates of enterprise characteristics, and a model feedback optimization mechanism. The policy map is updated in real-time via API to reflect policy changes, enterprise characteristics are corrected for deviations through triggering mechanisms and periodic synchronization, and the model optimizes weights based on application feedback. This solves the problems of lagging adaptation of static policy databases and missed policies due to failure to track enterprise changes, ensuring that enterprises obtain the latest and most effective policies and facilitating accurate policy delivery.

[0021] 3. Transparent Matching Logic: This invention makes the matching logic clear and traceable by using matching criteria and multi-dimensional score decomposition. It clearly shows the core reasons for policy fit and the key obstacles to misfit. Enterprises can intuitively understand the reasons and obstacles for fit and adjust their application strategies accordingly. Policy management departments can intuitively evaluate the effectiveness of policy promotion.

[0022] 4. Strong scenario adaptability: This invention dynamically adjusts feature weights and fusion weights to address the differentiated needs of enterprises of different sizes and development stages. The fusion weights prioritize application conditions during the initial push notification phase, while balancing conditions and requirements during self-application. This solves the misjudgment problem caused by the "one-size-fits-all" matching of existing technologies and meets the personalized policy adaptation needs of various enterprises.

[0023] 5. Excellent Application Guidance: This invention generates a matching analysis report containing full-process guidance. For policies that are highly compatible, it provides application timelines, materials, and procedures; for policies with incomplete conditions, it provides a supplementary feature list and improvement paths; and for policies that are not compatible, it explains the core obstacles. This solves the problems of lack of application guidance and high thresholds in existing results, reducing the difficulty for enterprises to apply and improving the efficiency of policy implementation. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a technology policy matching method that couples government and enterprise characteristics. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] like Figure 1 The present invention is illustrated in detail with reference to a specific embodiment, but does not limit the scope of the claims of the present invention in any way.

[0027] A method for matching science and technology policies that couples government and enterprise characteristics includes the following steps: S1. Policy data structuring and dynamic policy graph construction: Structured analysis of multi-source science and technology policy data to construct a dynamic policy graph that supports real-time updates.

[0028] Further, step S1 includes the following steps: S11. Policy data collection and preprocessing: Collect policy texts issued by the Ministry of Science and Technology, provincial science and technology departments, and municipal science and technology bureaus, including policy texts, application guidelines, revision notices, expiration announcements, etc. Simultaneously collect policy interpretation materials, official application Q&A and other auxiliary information. Remove redundant content through text cleaning and generate standardized policy text fragments by sentence segmentation and word segmentation. S12. Structural extraction of policy elements: The Policy-BERT model is used to extract three types of elements, which consist of policy foundation elements, application condition elements, and rights and interests constraint elements. Policy foundation elements include policy name, issuing unit, policy type, applicable region, validity period, and application window period. Application condition elements include enterprise type requirements (e.g., high-tech enterprise, technology-based SME), R&D investment requirements (e.g., R&D investment as a percentage of operating revenue), intellectual property requirements (e.g., number of invention patents, number of software copyrights), and project field requirements (e.g., applicable technical field). Rights and interests constraint elements include funding amount, funding method, project acceptance standards, and subsequent supervision requirements. During the construction of the dynamic policy map, the policy map is dynamically updated by connecting to the API of the science and technology management department. New policies automatically trigger element extraction and are processed through formulas. When establishing element relationships, corresponding element attributes are updated and weights are adjusted during policy revisions. Element extraction employs a combination of sequence labeling and relationship classification. The formula for determining element relationships is as follows:

[0029] , To extract policy elements, , The BERT hidden layer output for the elements, , For trainable parameters, The weight matrix for the relation classification task. For the bias term in relation classification tasks, The type of relationship between elements; S13. Dynamic Policy Graph Construction: A policy graph is constructed using extracted policy elements as nodes and relationships between elements as edges. Node types include four categories: "Policy Subject," "Policy Element," "Constraints," and "Rights / Benefits." Edge types include "Issuance-Policy," "Requirement-Element," "Adaptation-Field," and "Constraint-Rights." A dynamic policy graph update mechanism is established. By connecting to the policy release interface of the science and technology management department, it captures policy additions, revisions, and expiration events in real time. For new policies, elements are automatically parsed and added to the graph; for revised policies, corresponding element nodes and relationships are updated; for expiration policies, the "expiration" status is marked and matching is disabled. S14. Assigning weights to policy elements: Initial weights are determined using the Analytic Hierarchy Process (AHP). Based on policy implementation priority, enterprise application attention, and policy constraint strictness, the initial weights of each policy element are determined using the AHP, with a weight range of [0, 1]. Specifically, step S1 utilizes a pre-trained semantic model to parse policy semantics, constructs a dynamic knowledge graph to capture the relationships between elements, maintains the timeliness of the graph by synchronizing policy changes in real time through API, and quantifies the importance of elements using the analytic hierarchy process. This addresses the problems of existing static policy databases failing to update in real time, leading to enterprises missing new policies or submitting expired applications, as well as the issues of lagging adaptation and the difficulty in identifying implicit semantic adaptation by only capturing surface features. It achieves structured modeling of multiple policy elements, tracks policy changes in real time, and avoids adaptation lag; it accurately identifies implicit policy requirements, ensuring that enterprises obtain the latest and most effective policies.

[0030] S2. Construction of a multi-dimensional feature system for enterprises: Extract dynamic features from multi-source enterprise data and construct a multi-dimensional feature system covering basic, operational, R&D, and qualification aspects; Further, step S2 includes the following steps: S21. Enterprise characteristic data collection, collecting multi-source data such as business registration (type, address), finance (R&D investment ratio), R&D (number of patents), qualifications (high-tech enterprise certification), and historical applications (past records); S22. Enterprise characteristics are structured and processed. The basic characteristics are that the enterprise type is mapped to a standard category and the address is associated with the third-level administrative region code; the operational characteristics are that the R&D investment ratio is standardized as a continuous value and divided into "high / medium / low" levels according to industry level; the R&D characteristics are the number of patents granted in the past 3 years and the proportion of R&D personnel; the qualification characteristics are marked with "valid / expired / not obtained" status and validity period. S23. Enterprise characteristics are dynamically updated, and the trigger mechanism automatically updates the characteristics when the enterprise adds new patents, qualifications or releases financial reports. Official data is synchronized monthly to correct deviations and ensure timeliness. S24. Assigning weights to enterprise features: Based on historical matching contribution and enterprise development stage, the gradient descent method is used to optimize the weights, and the sum of the weights of similar features is 1.

[0031] Specifically, step S2 integrates multi-source dynamic features of enterprises, optimizes weights to adapt to the needs of different scales / development stages, and maintains the timeliness of features through triggering mechanisms and regular synchronization. This solves the problem that existing single-dimensional matching is one-sided and cannot track the dynamic changes of enterprises, resulting in matching results that are out of touch with reality. It builds a comprehensive dynamic feature system that adapts to the needs of enterprises of different scales / development stages, ensures that features are consistent with the actual state of enterprises, and avoids "one-size-fits-all" matching.

[0032] S3. Deep Coupling Matching: A two-level matching strategy of "policy map coarse screening - element feature fine matching - fusion judgment" is adopted to achieve deep coupling matching between policies and enterprises; Further, step S3 includes the following steps: S31. Policy Mapping Preliminary Screening: This process uses three criteria—regional fit, type fit, and timeliness fit—to initially select a set of applicable policies. Regional fit screening determines whether the administrative region where the company's registered address is located is within the policy's applicable area. Type fit screening matches the corresponding policy type based on the company's application needs. Timeliness fit screening determines whether the current time is within the policy application window, excluding policies that have expired or for which application has not yet been initiated. S32. Fine-grained matching of elements and characteristics: For the policies after the initial screening, calculate the fit between policy elements and enterprise characteristics, including two categories: fit of application conditions and fit of rights and interests. The application condition fit is calculated by matching each element of the policy application conditions with the corresponding characteristics of the enterprise. For numerical elements, the similarity between the normalized feature value and the policy threshold is calculated using the following formula:

[0033] in, These are enterprise characteristic values. As a threshold for policy elements, For numerical element fit, the range is [0, 1]; if the enterprise characteristic category matches the policy requirement category, then... If the enterprise characteristic category belongs to a subset of the policy requirement category, then ;otherwise ; When optimizing the matching model feedback, the gradient descent method is used to optimize the enterprise feature weights. The suitability of each application condition element is weighted and summed according to the element weights to obtain the application condition suitability. The formula is:

[0034] in: For the first The weighting of the elements in the application requirements for each policy item. For the first The suitability of each element; The suitability of policy benefits is determined by matching the enterprise's policy needs with the benefits offered by the policy. A mapping table of "Enterprise Need Type - Policy Benefit Type" is established. If the type of policy benefit matches the type of enterprise need, and the benefit size meets the enterprise's expectations, then... If the type matches but the size is not fully satisfied, then If the types do not match, then ; S33. Integration Judgment: Weighted integration of two types of fit, categorized by threshold, and the fit of application conditions and the fit of rights and interests are integrated according to weight to obtain the final matching score between policy and enterprise. The formula is:

[0035] in: To incorporate weights, the values ​​range from [0, 1]. When science and technology management departments conduct policy dissemination... When enterprises submit their own plans Set matching threshold ,like If it is deemed a high-quality policy fit, the company is recommended to apply; if If the policy conditions need improvement, the required enterprise characteristics will be output; if If so, it is determined to be an incompatible policy, and the core reasons for the incompatibility are output; Specifically, the secondary matching strategy first narrows down the scope, and then achieves accurate judgment by integrating multi-dimensional scores. It makes the matching logic transparent, solves the problems of low matching accuracy and opaque logic in the existing matching system, which easily leads to misjudgment and missed judgment, improves matching accuracy, and makes the logic transparent to avoid misjudgment and missed judgment.

[0036] S4. Dynamic Updates and Feedback Optimization: Based on policy changes, changes in enterprise characteristics, and feedback on matching results, a closed-loop optimization mechanism is constructed; Further, step S4 includes the following steps: S41. The policy map is dynamically updated, and the API of the science and technology department is connected to obtain policy changes in real time. New policies are automatically extracted into the map, the attributes of policy update nodes are revised, and expired policies are marked and blocked. S42. Enterprise characteristics are dynamically updated. When an enterprise adds a new patent / qualification / financial report, the characteristics are automatically updated, and official data is synchronized monthly to correct any discrepancies. S43. Matching model feedback optimization: Adjust element weights based on enterprise feedback, and retrain the model every quarter; increase the corresponding element weights for highly compatible policies that have been successfully applied for; adjust the corresponding element weights for policies that require improvement after feature addition; correct the matching logic for misjudged unsuitable policies; and retrain the model parameters every quarter based on feedback data.

[0037] S5. Matching Result Analysis and Application Suggestion Generation: Generates a matching result analysis report that includes the basis for matching, score breakdown, and application suggestions.

[0038] Further, step S5 includes the following steps: S51. Matching Criteria: Link policy graph nodes with enterprise characteristics to pinpoint specific reasons for suitability / missuitability; S52, Multi-dimensional score decomposition: Display , , Explain the contribution in each dimension; S53. Generation of application suggestions: Highly adaptable: Outputs timelines, bills of materials, and process guidelines; Conditions to be improved: The output needs to include a feature list and improvement paths; Incompatibility: Cause of core output obstacle; Specifically, step S5 uses graph association and score decomposition to transparently match logic, generating targeted application guidance to solve the problems of opaque matching results, lack of application guidance, and difficulty for enterprises to adjust their strategies. It provides clear matching basis and full-process application guidance, lowers the application threshold for enterprises, and improves the efficiency of policy implementation.

[0039] Working principle: First, a dynamic policy graph is constructed to achieve the structuring and real-time updating of policy elements: texts and supplementary interpretation materials of science and technology policies at the national, provincial, and municipal levels are collected, and after cleaning and preprocessing, the Policy-BERT model is used to extract the basic policy elements, application condition elements, and rights and interests constraint elements; the graph is constructed with elements as nodes and relationships as edges, and the API of the science and technology department is connected to synchronize policy addition, revision, and expiration information in real time. Initial weights are assigned to each element through the analytic hierarchy process to solve the problem of adaptation lag in the static policy database. Secondly, a multi-dimensional dynamic feature system for enterprises is constructed: This involves collecting multi-source data on enterprise registration, finance, R&D, qualifications, and historical applications; mapping basic features to standard categories; classifying operational features into levels; statistically analyzing the number of patents and the proportion of personnel in R&D; and marking the status of qualification features. Triggering and periodic synchronization mechanisms are established, and based on historical matching contributions and the enterprise's development stage, gradient descent is used to optimize feature weights, addressing the problem of one-sided matching in single-dimensional contexts. Secondly, a deep coupling matching strategy is implemented: through a two-level mechanism of "policy map rough screening - element fine matching - integration judgment", firstly, preliminary matching policies are screened based on applicable regions, policy types, and timeliness; then, the matching degree of application conditions and rights and interests needs are calculated; finally, the final score is obtained by using integration weight, and the matching level is determined according to the threshold, thereby improving the matching accuracy and logical transparency. Next, a closed-loop optimization mechanism is constructed to achieve dynamic updates: real-time synchronization of policy maps; collection of enterprise feedback to adjust factor weights; retraining of model parameters every quarter; continuous optimization of matching logic; and resolution of the problem of declining adaptability of static models. Finally, a matching analysis report and application suggestions are generated: the specific basis for matching is identified, the application conditions, rights and interests requirements and the final matching score are broken down; for highly compatible policies, the application time nodes, material lists and process guidelines are provided; for policies with conditions that need to be improved, the feature lists and improvement paths need to be supplemented; for uncompatible policies, the core obstacles are provided, and the problems of opaque matching logic and lack of application guidance are resolved.

[0040] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for matching science and technology policies based on the coupling of government and enterprise characteristics, characterized in that, Includes the following steps: S1. Policy Data Structuring and Dynamic Policy Graph Construction: Structured analysis of multi-source science and technology policy data to construct a dynamic policy graph that supports real-time updates; S2. Construction of a multi-dimensional feature system for enterprises: Extract dynamic features from multi-source enterprise data and construct a multi-dimensional feature system covering basic, operational, R&D, and qualification aspects; S3. Deep Coupling Matching: A two-level matching strategy of "policy map coarse screening - element feature fine matching - fusion judgment" is adopted to achieve deep coupling matching between policies and enterprises; S4. Dynamic Updates and Feedback Optimization: Based on policy changes, changes in enterprise characteristics, and feedback on matching results, a closed-loop optimization mechanism is constructed; S5. Matching Result Analysis and Application Suggestion Generation: Generates a matching result analysis report that includes the basis for matching, score breakdown, and application suggestions.

2. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, Step S1 includes the following steps: S11. Policy Data Collection and Preprocessing: Collect policy texts from multiple sources, clean up redundant content, and standardize them into text fragments; S12. Structural extraction of policy elements: The Policy-BERT model is used to extract three types of elements: basic elements, application condition elements, and rights and interests constraint elements. S13. Dynamic Policy Graph Construction: Modeling with elements as nodes and relationships as edges, connecting to official interfaces to synchronize policy additions / revisions / expirations in real time, and updating the graph; S14. Assigning weights to policy elements: Determine the initial weights using the analytic hierarchy process (AHP).

3. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, Step S2 includes the following steps: S21. Enterprise Feature Data Collection: Collect multi-source data including business registration, finance, R&D, qualifications, and historical applications; S22. Enterprise Feature Structuring Processing: Standardized conversion, basic feature mapping to category / regional codes, operational feature classification into levels, R&D feature statistics on patent quantity, qualification feature status marking; S23. Dynamic updates of enterprise characteristics: Automatic updates are triggered by a mechanism, and official data is synchronized monthly to ensure timeliness; S24. Assigning weights to enterprise features: Based on historical matching contribution and enterprise development stage, the weights are optimized using gradient descent.

4. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, Step S3 includes the following steps: S31. Preliminary screening of policy map: Screening a preliminary set of suitable policies based on three criteria: regional suitability, type suitability, and timeliness suitability; S32. Fine-grained matching of feature elements: Application criteria fit: Calculate the scores of numerical and categorical elements, and sum them by weight to obtain the result; Rights and interests matching: mapping enterprise needs with policy rights and interests, scoring the matching degree by type / scale; S33. Fusion determination: Weighted fusion of two types of fit, classified according to threshold.

5. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, Step S4 includes the following steps: S41. Dynamic updates to the policy graph: Connects to the API of the science and technology department to synchronize policy changes in real time and automatically update graph nodes / relationships; S42. Dynamic Update of Enterprise Characteristics: The trigger mechanism automatically updates characteristics and regularly synchronizes and corrects deviations monthly. S43. Matching Model Feedback Optimization: Adjust the element weights based on enterprise feedback and retrain the model every quarter.

6. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, Step S5 includes the following steps: S51. Matching Criteria: Link policy graph nodes with enterprise characteristics to pinpoint specific reasons for suitability / missuitability; S52. Multi-dimensional score breakdown: Displays application requirements, rights and interests needs, and final matching score, explaining the contribution of each dimension; S53. Generation of application suggestions: Highly adaptable: Outputs timelines, bills of materials, and process guidelines; Conditions to be improved: The output needs to include a feature list and improvement paths; Incompatibility: Cause of core output obstacle.

7. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 2, characterized in that, In step S12, during the construction of the dynamic policy map, the API of the science and technology management department is connected when the policy map is dynamically updated. New policies automatically trigger element extraction and are processed through formulas. When establishing element relationships, corresponding element attributes are updated and weights are adjusted during policy revisions. Element extraction employs a combination of sequence labeling and relationship classification. The formula for determining element relationships is as follows: ； in, , To extract policy elements, , The BERT hidden layer output for the elements, , For trainable parameters, The weight matrix for the relation classification task. For the bias term in relation classification tasks, This refers to the type of relationship between elements.

8. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 1, characterized in that, In step S32, the enterprise characteristic update trigger mechanism includes automatic updates upon new qualification certifications, patent applications, or the release of financial statements, using a formula. Calculate the fit between the new feature and the policy threshold, and regularly synchronize official data monthly to ensure the timeliness of the feature. The formula is: ； in, These are enterprise characteristic values. As a threshold for policy elements, The fit is for numerical features, ranging from [0, 1].

9. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 4, characterized in that, In step S32, the gradient descent method is used to optimize the enterprise feature weights during the matching model feedback optimization, combined with the formula... The contribution of each element in the application criteria is adjusted, and the model parameters are retrained quarterly based on feedback data. The formula for calculating the fit of the application criteria is as follows: ； in: For the first The weighting of the elements in the application requirements for each policy item. For the first The suitability of each element.

10. The method for matching science and technology policies based on government-enterprise characteristic coupling according to claim 9, characterized in that, In step S33, the matching result is parsed using the formula... Decomposition of application conditions and their suitability Alignment with rights and interests To enhance the transparency of the matching logic, the contribution of each dimension to the final score is clearly shown. The formula is as follows: ； in: To incorporate weights, the values ​​range from [0, 1]. When science and technology management departments conduct policy dissemination... When enterprises submit their own plans Set matching threshold ,like If it is deemed a high-quality policy fit, the company is recommended to apply; if If the policy conditions need improvement, the required enterprise characteristics will be output; if If so, it is determined to be an incompatible policy, and the core reasons for the incompatibility are output.

Images