Policy adaptation and navigation method

Abstract

The invention relates to the technical field of policy matching, in particular to a policy adaptation and navigation method which comprises the following steps: acquiring policy original text data through a data acquisition interface, and analyzing the policy original text data into structured policy element information by using a natural language processing model; obtaining multi-source enterprise data through a data access and filling channel, and fusing and constructing the multi-source enterprise data into dynamically updated enterprise portrait information; inputting the policy factor information and the enterprise portrait information into a coarse matching filter, performing preliminary screening based on industry and regional conditions, and outputting a policy-enterprise candidate set; and inputting the candidate set into a pre-trained fine matching learning model, receiving policy quantification requirements and enterprise state data by the model, performing processing analysis based on decision-making knowledge obtained by training, and finally dividing each policy-enterprise pair into three enterprise sets, namely, a completely-conforming enterprise set, a nearly-standard enterprise set and a non-conforming enterprise set. According to the invention, the coverage rate and accuracy of policy matching are improved, and upgrading from policy search to intelligent navigation is realized.

Description

Technical Field

[0001] This invention relates to the field of policy matching technology, and in particular to a policy adaptation and navigation method. Background Technology

[0002] In the current technology, governments have introduced a large number of policies covering multiple fields to promote economic development and social progress. However, enterprises and institutions face serious challenges in obtaining relevant policy information. Traditional policy matching methods mainly rely on keyword search or screening systems based on rigid rules. These methods have limitations: they can only identify enterprises that fully comply with the explicit terms, while excluding a large number of enterprises that are close to meeting the standards but have slight deficiencies in some indicators but have overall development potential. Summary of the Invention

[0003] In view of this, the purpose of this invention is to propose a policy adaptation and navigation method to solve the problem of excluding a large number of near-compliant enterprises that are only slightly deficient in some indicators but have overall development potential.

[0004] To achieve the above objectives, the present invention provides a policy adaptation and navigation method, the method comprising the following information processing steps: Policy text data is collected from multiple official government sources via data acquisition interfaces. The original policy data is parsed using a natural language processing model, transforming unstructured policy text information into structured policy element information. The policy elements include at least the policy name, issuing department, applicable industry, funding method, quantitative indicators, and application deadline. Through data access and reporting channels, multi-source enterprise data flows in and is integrated to construct dynamically updated enterprise profile information; The policy element information and enterprise profile information are input into the coarse matching filter, and a preliminary screening is performed based on the core industry and regional conditions of the policy, outputting a policy-enterprise candidate set information.

[0005] The method also includes a fine-sorting matching information processing step: The policy-enterprise candidate set information is input into a pre-trained fine-ranking matching learning model; The information received by the model includes: all quantitative requirements from policy element information and corresponding status data from enterprise profile information; The fine-ranking matching learning model processes and distinguishes the above information based on the decision knowledge it has acquired during training. Finally, it outputs a matching result for each policy-enterprise pair in the candidate set and divides them into sets of enterprises that fully comply, are close to comply, and do not comply.

[0006] Optionally, the fine-ranking matching learning model is trained in the following way: acquiring training data, which includes quantitative requirements of policy element information, corresponding status data of enterprise profile information, and historical enterprise sample data marked as fully compliant, nearly compliant, and completely non-compliant; training the model through a contrastive learning loss function, which is configured to: reduce the distance between fully compliant samples and nearly compliant samples in the feature representation space of the model, while increasing the distance between fully compliant samples and completely non-compliant samples.

[0007] Optionally, the process of constructing the enterprise profile information includes: Real-time enterprise financial data and employee social security data flow in from the authorized interface; Intellectual property status data flowing into enterprises from public databases; The above data will be merged, disambiguated, and aligned with the text description data submitted by the enterprises themselves. Based on the fused data, vectorized enterprise profile information containing enterprise qualification tags, financial capability tags, and innovation capability tags is dynamically generated and updated.

[0008] Optionally, the method further includes an information processing step for quantifying ambiguous policy provisions: Supporting information related to the target policy flows in from internet sources, including official interpretations, press releases, and public disclosures of historical success stories; Key success factor information is extracted from the auxiliary information using natural language processing technology; The key success factor information is quantified into computable weighted index information; The weighting index information is injected into the decision-making process of the fine-matching model as a correction factor for the final classification result of the policy-enterprise adjustment.

[0009] This approach offers a solution for ambiguous clauses in policies that are difficult to quantify directly (such as industry-leading technologies). The system receives supplementary information related to the policy from internet sources (such as departmental websites and news), including official interpretations and publicized success stories. Subsequently, natural language processing technology is used to extract key success factors from these texts and quantify these factors into calculable weighted indicators. Finally, during the fine-tuning and matching process, these weighted indicators are used as correction factors to adjust the final matching results of policy-enterprise pairs.

[0010] This method creatively quantifies and integrates tacit knowledge and review experience in policy implementation, solving the industry problem of handling ambiguous clauses in policy matching. It enables intelligent systems to understand the implied meanings and potential thresholds of policies like human experts, thereby greatly improving the accuracy and reliability of matching.

[0011] Optionally, the information processing sub-steps for extracting key success factors from publicized historical success cases include: Lock in the list of successful companies that correspond to the policies over the years, and compile information on the entities of these successful companies. Based on the successful enterprise entity information, the panoramic profile data of these enterprises at the time of application is retrieved from the enterprise information database. By using clustering and statistical analysis methods, common and frequently occurring qualification and capability characteristics are extracted from the panoramic profile data to form a profile of successful enterprise groups under this policy. By comparing the profiles of successful business groups with the explicit elements of policies, key success factors that are not explicitly stated in the policy documents but have a real impact on the outcome are identified and output.

[0012] Optionally, when faced with the problem of policy matching that lacks or is insufficient in historical success cases, the method further includes the steps of policy intent vectorization and cross-domain mapping: Policy Intent Deep Vectorization: Using a large language model, a deep semantic analysis is performed on the full text of the target policy to generate a policy intent vector; this vector is used to quantitatively represent the policy's strategic orientation, the key issues it aims to address, and the technological and economic characteristics it aims to support. Constructing a cross-domain technology-industry knowledge graph: Building a technology-industry knowledge graph; Potential target identification and mapping: The technical characteristics and business functions of the companies to be matched are queried and semantically matched in the technology-industry knowledge graph to find technical paths or industry links that are semantically similar to the policy intent vector; For companies that do not meet the policy requirements in traditional industry classifications but whose technological paths or industrial links are semantically similar, the system identifies them as companies that are close to meeting the standards.

[0013] Optionally, for near-compliant enterprises identified through cross-domain mapping, the method further includes a potential simulation assessment step based on policy intent: constructing an ideal enterprise model: generating one or more virtual ideal target enterprise profiles based on the policy intent vector; conducting gap and potential analysis: comparing the near-compliant enterprises to be assessed with the ideal target enterprise profiles to generate a strategic gap analysis report; outputting forward-looking recommendations: incorporating the aforementioned enterprises and their strategic gap analysis reports into the final recommendation results and marking them as high-potential strategic candidates. This prompts decision-makers to pay attention to the alignment between the enterprise's future growth potential and policy objectives.

[0014] Optionally, the nodes of the technology-industry knowledge graph include technology concepts, product functions, and industrial chain links, and the edge relationships include technology substitution, functional similarity, and upstream and downstream of the industrial chain.

[0015] Optionally, the search for technological paths or industrial segments that are semantically similar to the policy intent vector is obtained by calculating the semantic similarity between the enterprise's technological business vector and the policy intent vector.

[0016] This method constructs a complete data processing and decision-making pipeline. First, the system automatically collects original policy texts from official sources and uses a natural language processing model to parse them into structured policy elements, transforming unstructured text into machine-readable key data. Simultaneously, by integrating enterprise data from multiple sources, a comprehensive and dynamically updated enterprise profile is constructed. Subsequently, the system performs a two-stage matching process: in the coarse-grained matching stage, core conditions such as industry and region are used for rapid and lenient initial screening, forming a broad candidate set; in the fine-grained matching stage, a pre-trained intelligent model is activated. It receives detailed data from policies and enterprises and uses its training-acquired knowledge for in-depth analysis and decision-making, ultimately outputting a precise matching result for each policy-enterprise pair, categorizing it as fully compliant, nearly compliant, or non-compliant.

[0017] This invention constructs an end-to-end automated policy service system. Its core beneficial effect lies in the fact that through a two-stage matching mechanism, it achieves a dual improvement in policy matching efficiency and coverage. Coarse ranking ensures breadth and avoids missing potential targets, while fine ranking ensures depth and accuracy, especially in identifying those potential near-compliant enterprises, thereby upgrading the traditional policy search tool into an intelligent policy adaptation and navigation platform. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "include" or "comprising" mean that the element or object preceding the term covers the element or object listed after the term and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "up," "down," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] like Figure 1 As shown, a policy adaptation and navigation method includes the following information processing steps: Policy text data is collected from multiple official government sources via data acquisition interfaces. The original policy data is parsed using a natural language processing model, transforming unstructured policy text information into structured policy element information. The policy elements include at least the policy name, issuing department, applicable industry, funding method, quantitative indicators, and application deadline. Through data access and reporting channels, multi-source enterprise data flows in and is integrated to construct dynamically updated enterprise profile information; The policy element information and enterprise profile information are input into the coarse matching filter, and a preliminary screening is performed based on the core industry and regional conditions of the policy, outputting a policy-enterprise candidate set information.

[0023] The method also includes a fine-sorting matching information processing step: The policy-enterprise candidate set information is input into a pre-trained fine-ranking matching learning model; The information received by the model includes: all quantitative requirements from policy element information and corresponding status data from enterprise profile information; The fine-ranking matching learning model processes and distinguishes the above information based on the decision knowledge it has acquired during training. Finally, it outputs a matching result for each policy-enterprise pair in the candidate set and divides them into sets of enterprises that fully comply, are close to comply, and do not comply.

[0024] This method constructs a complete data processing and decision-making pipeline. First, the system automatically collects original policy texts from official sources and uses a natural language processing model to parse them into structured policy elements, transforming unstructured text into machine-readable key data. Simultaneously, by integrating enterprise data from multiple sources, a comprehensive and dynamically updated enterprise profile is constructed. Subsequently, the system performs a two-stage matching process: in the coarse-grained matching stage, core conditions such as industry and region are used for rapid and lenient initial screening, forming a broad candidate set; in the fine-grained matching stage, a pre-trained intelligent model is activated. It receives detailed data from policies and enterprises and uses its training-acquired knowledge for in-depth analysis and decision-making, ultimately outputting a precise matching result for each policy-enterprise pair, categorizing it as fully compliant, nearly compliant, or non-compliant.

[0025] This invention constructs an end-to-end automated policy service system. Its core beneficial effect lies in the fact that through a two-stage matching mechanism, it achieves a dual improvement in policy matching efficiency and coverage. Coarse ranking ensures breadth and avoids missing potential targets, while fine ranking ensures depth and accuracy, especially in identifying those potential near-compliant enterprises, thereby upgrading the traditional policy search tool into an intelligent policy adaptation and navigation platform.

[0026] In some embodiments, the fine-ranking matching learning model is trained by: acquiring training data, which includes quantitative requirements for policy element information, corresponding status data of enterprise profile information, and historical enterprise sample data marked as fully compliant, nearly compliant, and completely non-compliant; and training the model by a contrastive learning loss function, which is configured to: reduce the distance between fully compliant samples and nearly compliant samples in the feature representation space of the model, while increasing the distance between fully compliant samples and completely non-compliant samples.

[0027] The model's training data comprises three main elements: quantitative policy requirements, corresponding enterprise status data, and manually labeled historical enterprise samples (categorized as fully compliant, nearly compliant, and completely non-compliant). The core of the training lies in using a contrastive learning loss function, which performs a specific optimization within the model's feature representation space: it aims to reduce the feature distance between fully compliant and nearly compliant samples, while simultaneously increasing the distance between fully compliant and completely non-compliant samples.

[0028] This specific training mechanism endows the model with a unique ability to understand the flexible boundaries of policy conditions, enabling the system to accurately identify and recommend near-compliant enterprises that are slightly deficient in a single indicator but perform well in other aspects. This significantly improves the coverage and practicality of policy matching and effectively solves the core pain point of potential high-quality enterprises being mistakenly screened out due to rigid screening in existing technologies.

[0029] For example, a policy requires that R&D expenditure account for no less than 8% of total revenue. A company with an R&D expenditure of 7.5% (slightly lower) but possessing a number of core intellectual property rights far exceeding the requirement might be directly rejected in a traditional model. However, the model trained using this method, having learned the boundary knowledge that intellectual property advantages can partially compensate for small gaps in R&D investment, will accurately identify the company as close to meeting the requirement and strongly recommend it.

[0030] In some embodiments, the process of constructing the enterprise profile information includes: Real-time enterprise financial data and employee social security data flow in from the authorized interface; Intellectual property status data flowing into enterprises from public databases; The above data will be merged, disambiguated, and aligned with the text description data submitted by the enterprises themselves. Based on the fused data, vectorized enterprise profile information containing enterprise qualification tags, financial capability tags, and innovation capability tags is dynamically generated and updated.

[0031] This invention draws real-time financial and personnel data from authorized interfaces (such as ERP and social security systems) and intellectual property data from public databases. It then integrates, disambiguates (resolves data conflicts), and aligns (associates with the same corporate entity) these objective data with descriptive text data self-reported by enterprises. Finally, based on this integrated data, it dynamically generates and updates a vectorized profile composed of multiple dimensions such as corporate qualifications, financial capabilities, and innovation.

[0032] Through multi-source data fusion and dynamic updates, the enterprise profile constructed by this method is no longer static and one-sided. It can provide a comprehensive, real-time and computable digital twin of the enterprise status, providing a solid and reliable data foundation for subsequent accurate matching, and making the policy matching results more reflective of the enterprise's true strength and development potential.

[0033] In some embodiments, the method further includes an information processing step for quantifying ambiguous policy provisions: Supporting information related to the target policy flows in from internet sources, including official interpretations, press releases, and public disclosures of historical success stories; Key success factor information is extracted from the auxiliary information using natural language processing technology; The key success factor information is quantified into computable weighted index information; The weighting index information is injected into the decision-making process of the fine-matching model as a correction factor for the final classification result of the policy-enterprise adjustment.

[0034] This approach offers a solution for ambiguous clauses in policies that are difficult to quantify directly (such as industry-leading technologies). The system receives supplementary information related to the policy from internet sources (such as departmental websites and news), including official interpretations and publicized success stories. Subsequently, natural language processing technology is used to extract key success factors from these texts and quantify these factors into calculable weighted indicators. Finally, during the fine-tuning and matching process, these weighted indicators are used as correction factors to adjust the final matching results of policy-enterprise pairs.

[0035] This method creatively quantifies and integrates tacit knowledge and review experience in policy implementation, solving the industry problem of handling ambiguous clauses in policy matching. It enables intelligent systems to understand the implied meanings and potential thresholds of policies like human experts, thereby greatly improving the accuracy and reliability of matching.

[0036] In some embodiments, the information processing sub-step for extracting key success elements from historical success case disclosures includes: Lock in the list of successful companies that correspond to the policies over the years, and compile information on the entities of these successful companies. Based on the successful enterprise entity information, the panoramic profile data of these enterprises at the time of application is retrieved from the enterprise information database. By using clustering and statistical analysis methods, common and frequently occurring qualification and capability characteristics are extracted from the panoramic profile data to form a profile of successful enterprise groups under this policy. By comparing the profiles of successful business groups with the explicit elements of policies, key success factors that are not explicitly stated in the policy documents but have a real impact on the outcome are identified and output.

[0037] First, a list of successful companies in the history of the policy is identified to form an entity list. Then, the comprehensive data of these companies at the time of their application is retrieved from the enterprise information database to form a group portrait of successful companies. Finally, through clustering and statistical analysis, common and frequently occurring characteristics are extracted from this group portrait and compared with the explicit policy text to identify the key success factors that are not explicitly stipulated but actually affect the results.

[0038] By reverse-engineering and quantifying the genes of success from historical success cases, the system is provided with a data-driven capability to automatically discover implicit policy evaluation standards. This makes the quantification process of ambiguous clauses more objective and accurate, and avoids biases caused by subjective settings.

[0039] Example: For a policy requiring significant social benefits, a systematic analysis of 50 successful companies reveals that 90% of the cases mention annual carbon emission reductions exceeding 1,000 tons. The system then quantifies this 1,000 tons as a key implicit success factor and, in subsequent matching, judges companies whose carbon emission reductions far exceed this figure, even if they have slight shortcomings in other aspects, as highly meeting the requirement of significant social benefits.

[0040] In some embodiments, when faced with the problem of policy matching that lacks or is insufficient in historical success cases, the method further includes the steps of policy intent vectorization and cross-domain mapping: Policy Intent Deep Vectorization: Using a large language model, a deep semantic analysis is performed on the full text of the target policy to generate a policy intent vector; this vector is used to quantitatively represent the policy's strategic orientation, the key issues it aims to address, and the technological and economic characteristics it aims to support. Constructing a cross-domain technology-industry knowledge graph: Building a technology-industry knowledge graph; Potential target identification and mapping: The technical characteristics and business functions of the companies to be matched are queried and semantically matched in the technology-industry knowledge graph to find technical paths or industry links that are semantically similar to the policy intent vector; For companies that do not meet the policy requirements in traditional industry classifications but whose technological paths or industrial links are semantically similar, the system identifies them as companies that are close to meeting the standards.

[0041] This method is activated when faced with entirely new policies and a lack of historical case studies. It utilizes a large language model to perform deep semantic analysis of the full policy text, generating a high-dimensional vector representing the policy's strategic intent. Simultaneously, the system constructs a knowledge graph encompassing technologies, products, supply chain nodes, and their relationships (such as substitution, similarity, and upstream / downstream). Finally, the technological characteristics of the companies to be matched are semantically matched against the policy intent vector within the knowledge graph, seeking paths or links that are highly compatible in function or technology, thereby identifying potential cross-industry targets.

[0042] This method breaks through the limitations of traditional policy matching, which relies on industry classification and past data. It enables the system to match policies with zero or few samples, and can proactively discover and recommend high-potential enterprises that are in the cross-domain, have potential value, but cannot be identified by traditional standards. This achieves a value leap from matching to discovery.

[0043] In some embodiments, for near-compliant enterprises identified through cross-domain mapping, the method further includes a potential simulation assessment step based on policy intent: constructing an ideal enterprise model: generating one or more virtual ideal target enterprise profiles based on the policy intent vector; performing gap and potential analysis: comparing the near-compliant enterprises to be assessed with the ideal target enterprise profiles to generate a strategic gap analysis report; outputting forward-looking recommendations: incorporating the aforementioned enterprises and their strategic gap analysis reports into the final recommendation results and marking them as high-potential strategic candidates. This prompts decision-makers to pay attention to the alignment between the enterprise's future growth potential and policy objectives.

[0044] This method further conducts potential assessment. First, based on a policy intent vector, it generates a virtual, ideal target company profile as a benchmark. Then, it conducts a comprehensive comparison between the company to be evaluated and this ideal benchmark, generating a strategic gap analysis report. Finally, these companies and the analysis report are recommended as high-potential strategic candidates.

[0045] This approach upgrades policy matching output from a simple "whether it meets the criteria" to insightful strategic decision support. Its core benefit is that it provides decision-makers (governments or enterprises) with a clear and actionable navigation path, not only indicating who has potential, but also clarifying where that potential lies and how to bridge the gap, thus greatly improving the efficiency of policy resource allocation and the precision of support.

[0046] For example: Policy scenario: A city issued "Several Measures on Precisely Supporting the Innovative Development of Synthetic Biology Manufacturing Industry" (This is a brand-new policy aimed at laying out the future industry, that is, facing a brand-new policy and lacking historical cases).

[0047] Core policy intent (derived through system analysis): Strategic orientation: To break free from dependence on foreign technology platforms and build an independent and controllable design-build-test-learning cycle capability.

[0048] The key issue to be addressed is that local biotechnology companies are mostly focused on traditional fermentation optimization processes and lack cutting-edge life design and automated engineering platforms.

[0049] The technological and economic characteristics to be supported: Support the use of advanced technologies such as gene editing and automated strain construction to produce high-value chemicals and pharmaceutical raw materials.

[0050] Information on Company D (formerly a traditional enzyme preparation company): Traditional business: Production of industrial enzyme preparations for industries such as detergents and feed.

[0051] Core technology: Based on traditional microbial mutagenesis and screening technology, it has a highly efficient 5000-liter large-scale fermentation production line.

[0052] Industry classification: It belongs to enzyme preparation manufacturing under biochemical industry, which does not completely match the industry classification of synthetic biology manufacturing as stated in the policy.

[0053] System processing flow and output: Intent parsing and cross-domain mapping: The system uses a large language model to analyze the full text of policies and generate policy intent vectors, with a core focus on gene design and automated construction.

[0054] Searching for company D's large-scale fermentation technology in the technology-industry knowledge graph reveals that this technology is a key supporting capability in the construction and testing phases of the synthetic biology manufacturing DBTL cycle, and is highly complementary to the policy intent in terms of function.

[0055] Despite the mismatch in industry classification, the presence of entirely new policies, and the lack of historical case studies, the system still identified Company D as a potential strategic fitter that is close to meeting the criteria.

[0056] Potential simulation and report generation: The system generates an ideal target enterprise profile based on policy intent vectors: it has complete capabilities in gene design, strain construction, high-throughput testing, and large-scale scaling.

[0057] By comparing Company D with this ideal model, the following "Strategic Gap Analysis and Navigation Recommendations" report is automatically generated: [Policy Adaptation and Navigation Report] Company Positioning: High-potential strategic candidate. Although your company does not fully match the traditional industry classification, its core capabilities are highly aligned with the strategic intent of this synthetic biology manufacturing policy.

[0058] Strategic Gap Analysis: Key gap: Lack of top-level life system design capabilities limits the expansion of products into higher-value fields such as medicine and new materials.

[0059] Potential risks: Existing strain construction technologies rely on traditional mutagenesis, which is inefficient and may not meet the needs of future rapid iteration in research and development.

[0060] Core strengths identification: Having a mature, stable, and cost-controllable large-scale fermentation and purification platform is a rare industrial infrastructure that many design-focused startups lack.

[0061] Precise navigation suggestions: Path 1 (Neclipse Integration): We recommend actively transforming your company into a pilot-scale and large-scale production center for synthetic biology manufacturing. The application materials should focus on demonstrating the core value of your platform in bridging the last mile of the industrial chain and accelerating the industrialization of laboratory results.

[0062] Path Two (Collaborative Innovation): It is recommended to form an innovation consortium with upstream research institutions or startups possessing strong gene design capabilities for the application process. Under this recommendation, the system can further match you with relevant technology holders already identified within the jurisdiction.

[0063] In some embodiments, the technology-industry knowledge graph nodes include technology concepts, product functions, and industrial chain links, and the edge relationships include technology substitution, functional similarity, and upstream and downstream of the industrial chain.

[0064] The nodes of the graph include technological concepts, product functions, and links in the industrial chain; the edge relationships define the semantic connections between nodes, such as technological substitution (technology A can be replaced by technology B), functional similarity (product A and product B have similar functions), and upstream and downstream of the industrial chain (link A is a supplier of link B).

[0065] By constructing this knowledge network rich in semantic relationships, a structured knowledge foundation is provided for computers to understand the internal logic and technological connections of industries, making cross-domain and cross-industry intelligent mapping and semantic matching possible.

[0066] In some embodiments, the search for technological paths or industrial segments that are semantically similar to the policy intent vector is obtained by calculating the semantic similarity between the enterprise's technological business vector and the policy intent vector.

[0067] This is achieved by calculating the cosine similarity or Euclidean distance between the enterprise's technology and business vectors and the policy intent vectors in a high-dimensional space. Employing vector similarity calculation, a mature and interpretable technical solution, allows for the precise quantification of the correlation between policy intent and enterprise technology, providing an objective and unified evaluation standard for cross-domain matching and ensuring the reproducibility of the system's decision-making process.

[0068] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in the details for the sake of brevity.

[0069] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A policy adaptation and navigation method, characterized by, The method comprises the following information processing steps: Policy original text data is flowed in from multiple government official sources through a data collection interface; Natural language processing models are used to analyze the policy original text data, so that unstructured policy text information is converted into structured policy element information, and the policy element at least includes a policy name, a publishing department, an applicable industry, a funding method, a quantitative indicator, and a reporting deadline; Through a data access and reporting channel, multi-source enterprise data is flowed in and fused to build dynamic updated enterprise portrait information; Policy element information and enterprise portrait information are input into a coarse matching filter, and a policy-enterprise candidate set information is output based on the industry and regional core conditions of the policy; The method further comprises a fine matching information processing step: The policy-enterprise candidate set information is input into a pre-trained fine matching learning model; The information received by the model includes all quantitative requirements from the policy element information and corresponding state data from the enterprise portrait information; The fine matching learning model processes and distinguishes the above information based on the decision knowledge obtained by training, and finally outputs a matching result for each policy-enterprise pair in the candidate set, and divides it into a complete compliance, close to the standard and non-compliant enterprise set.

2. The method of claim 1, wherein, The fine matching learning model is trained by: obtaining training data, which includes quantitative requirements of policy element information, corresponding state data of enterprise portrait information, and historical enterprise sample data marked as complete compliance, close to the standard and complete inconsistency; the model is trained by comparing the learning loss function, which is configured to: in the feature representation space of the model, the distance between the complete compliance sample and the close to the standard sample is reduced, while the distance between the complete compliance sample and the complete inconsistency sample is increased.

3. The method of claim 1, wherein, The process of building the enterprise portrait information comprises: Real-time enterprise financial data and personnel social security data are flowed in from an authorized interface; Enterprise intellectual property status data is flowed in from a public database; The above data and enterprise self-reported text description data are fused, disambiguated and aligned; Based on the fused data, vectorized enterprise portrait information including enterprise qualification labels, financial capability labels and innovation labels is dynamically generated and updated.

4. The method of claim 1, wherein, The method further comprises a policy ambiguous clause quantification information processing step: Auxiliary information related to the target policy is flowed in from internet sources, including official interpretations, news releases and historical success case publications; Key success element information is extracted from the auxiliary information using natural language processing technology; The key success element information is quantified into computable weight indicator information; The weight indicator information is injected into the decision-making process of the fine matching model as a correction factor for adjusting the final division result of the policy-enterprise pair.

5. The method of claim 4, wherein, The information processing sub-step of extracting key success elements from historical success case publications comprises: Locking the list of successful enterprises corresponding to the policy over the years to form successful enterprise entity information; According to the successful enterprise entity information, the panoramic portrait data of these enterprises at the time of reporting is pulled from the enterprise information library in reverse; The common and repeated characteristics of qualifications and abilities are extracted from the panoramic image data through clustering and statistical analysis methods to form the successful enterprise group portrait information of the policy; The successful enterprise group portrait information is compared with the plaintext element information of the policy to identify and output the key success factor information that is not expressed in the plaintext but actually affects the results.

6. The method according to claim 4 or 5, characterized in that, When facing the policy matching problem of the lack of historical successful cases, the method further includes the steps of policy intention vectorization and cross-domain mapping: Policy intention deep vectorization: using a large language model to perform deep semantic analysis on the full text of the target policy to generate a policy intention vector; the vector is used to quantify the strategic orientation, key problems to be solved, and technical and economic characteristics to be supported of the policy; Building a cross-domain technology-industry knowledge graph: building a technology-industry knowledge graph; Potential target identification and mapping: querying and semantically matching the technical features and business functions of the enterprise to be matched in the technology-industry knowledge graph to find technical paths or industrial links that are semantically similar to the policy intention vector; For enterprises that do not meet the policy requirements in traditional industry classification but have semantically similar technical paths or industrial links, the system identifies them as close-to-meet-the-standard enterprises.

7. The method of claim 6, wherein, For close-to-meet-the-standard enterprises identified through cross-domain mapping, the method further includes the potential simulation evaluation step based on the policy intention: building an ideal enterprise model: based on the policy intention vector, generate one or more virtual ideal target enterprise portraits; perform gap and potential analysis: compare the close-to-meet-the-standard enterprise to be evaluated with the ideal target enterprise portrait to generate a strategic gap analysis report; output forward-looking recommendations: include the above enterprise and its strategic gap analysis report in the final recommendation results and mark it as a high-potential strategic alternative.

8. The method of claim 6, wherein, The technology-industry knowledge graph nodes include technical concepts, product functions, and industrial chain links, and the edge relationships include technology substitution, functional similarity, and industrial chain upstream and downstream.

9. The method of claim 6, wherein, The search for technical paths or industrial links that are semantically similar to the policy intention vector is obtained by calculating the semantic similarity between the enterprise's technical business vector and the policy intention vector.

Images