Intelligent management system for enterprise policy matching and planning

Abstract

The application belongs to the technical field of intelligent management of enterprise policy, and discloses an intelligent management system for enterprise policy matching and planning, wherein a policy coding acquisition unit is used to analyze implicit conditions by means of NLP semantic analysis and industry rule engine, and a double-dimensional model of industry risk and policy effect is used to optimize thresholds in combination with historical execution data, so as to generate a unique policy code containing information such as region and valid period, and to form a traceability chain by connecting a government affairs block chain; both quantitative constraints of implicit conditions and coding authenticity are ensured, the situation of report rejection is greatly reduced, and the efficiency and reliability of automatic verification of policy matching are improved; a multi-dimensional matching unit driven by the code can monitor dynamic changes of the code, automatically recalculate the matching degree, and load tax preferential and R&D subsidy priority codes for small and micro enterprises and high-tech enterprises respectively, so as to strengthen the corresponding implicit condition verification.

Description

Technical Field

[0001] This invention belongs to the field of intelligent management technology for enterprise policies, specifically an intelligent management system for matching and planning enterprise policies. Background Technology

[0002] In current enterprise-government collaboration scenarios, the number of policies benefiting enterprises (tax reduction and exemption, R&D subsidies, talent support, etc.) has surged. The professionalism of policy terms and the complexity of constraints have increased simultaneously. Enterprises urgently need practical solutions for accurate policy matching, efficient implementation, and dynamic tracking of effects. However, existing policy matching and planning technologies have the following technical problems, resulting in low policy implementation efficiency, poor benefits for enterprises, and failure to meet actual application needs.

[0003] Traditional systems rely on keyword matching technology, which cannot transform implicit constraints such as no major administrative penalties in the past three years and R&D expenses accounting for ≥5% into quantifiable and automatically verifiable codes. On the one hand, this results in the rejection of enterprise application materials due to failure to meet implicit conditions, wasting human and time costs. On the other hand, due to the lack of standardized codes, it is impossible to connect to government data verification interfaces and complete condition verification through automated processes, resulting in low matching reliability and efficiency.

[0004] Dynamic information such as policy validity period adjustments, clause revisions, enterprise qualification renewals, and changes in business scope are not included in the policy and enterprise coding logic; the system requires manual data updates and reconstruction of associations periodically, resulting in matching delays. For example, if the system does not synchronize after an enterprise's high-tech enterprise qualification is renewed, the application window may be missed; the system may still recommend policies after they expire, increasing the risk of invalid applications.

[0005] The existing system only covers the matching and solution generation stage, and does not link the execution progress and implementation effect with policy / enterprise codes. It cannot trace execution problems through codes, such as the reasons for overdue steps or failed material review, nor can it feed back effect data such as subsidy payment amount to the algorithm and coding rule optimization. The system is trapped in a cycle of rigid logic and stagnant effect, and cannot adapt to policy adjustments and enterprise development needs. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent management system for enterprise policy matching and planning, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent management system for enterprise policy matching and planning, the system comprising:

[0008] Policy coding collection unit: crawls policy data through multi-source crawlers, decomposes implicit conditions, automatically optimizes implicit condition thresholds, establishes a mapping relationship between constraint codes and real data values, and generates unique policy codes;

[0009] Enterprise coding integration unit: The dynamic perception module captures key events and triggers update signals; the coding is adjusted based on the update signals, and the data cleaning module removes invalid signals;

[0010] Coding-driven policy matching unit: Based on policy and enterprise coding, it decomposes three types of feature vectors: explicit, implicit, and dynamic, establishes a dynamic weight self-iterative algorithm to adjust weights in real time, establishes a two-dimensional scenario-based matching matrix, and automatically recalculates the matching degree when coding changes;

[0011] Solution generation and coding tracking unit: Generates implementation plans and assigns progress codes and effect codes, performs risk rating for the current steps, and monitors progress codes and effect data through a triple coding system linking policy, enterprise, and progress.

[0012] Policy Inquiry and Feedback Unit: Information is obtained through regular QR code scanning devices and visualized charts are generated; users can submit opinions through feedback codes.

[0013] Security and Access Control Unit: Encrypts encoded and associated data; assigns different permissions to different roles; and records operation logs.

[0014] Update and Management Unit: Assign version numbers to policies and enterprise codes, archive and store old versions; compare enterprise performance codes with baseline codes, automatically adjust matching algorithm parameters, and establish a mapping table between new and old codes.

[0015] Preferably, the policy coding acquisition unit is specifically as follows:

[0016] By regularly crawling policy texts and explicit metadata from government platforms at all levels through a multi-source crawler module, implicit conditions are decomposed using NLP semantic parsing and an industry rule engine, and a dual-dimensional training model is set up to assess industry risk and policy effectiveness. The implicit condition thresholds are automatically optimized based on historical policy execution data, and effect codes are bound to transform the implicit conditions into codes that can be quantified and constrained.

[0017] By connecting with third-party data sources, a mapping relationship between constraint codes and real data values ​​is established. Unique policy codes are generated based on regional codes, policy type codes, release timestamps, constraint code sets, and validity period codes. The policy codes are then written into the blockchain by connecting with local government blockchain platforms, forming a cycle traceability chain of code generation, modification, and repeal.

[0018] Latent conditional dynamic threshold optimization formula:

[0019] ;

[0020] This represents the optimized implicit condition threshold.

[0021] This represents the initial threshold before optimization;

[0022] This represents the industry risk coefficient weight, with a value ranging from 0.1 to 0.5.

[0023] This represents the industry risk index, with a value ranging from 0 to 1.

[0024] This represents the weight of the policy effectiveness coefficient, with a value ranging from 0.2 to 0.6.

[0025] This represents the historical policy effectiveness index, with a value range of 0 to 1.

[0026] This represents the matching accuracy feedback coefficient, with a value ranging from 0.8 to 1.2.

[0027] Preferably, the enterprise coding integration unit is specifically as follows:

[0028] After completing the policy data coding and collection, the system collects qualification, business and financial data by connecting to the enterprise system via API. The dynamic data perception module monitors the data change log in real time, captures key events including the renewal of high-tech enterprise qualifications and triggers update signals.

[0029] A unique master code is assigned to each enterprise, and a sub-code is assigned to each data type, forming a master code and sub-code system. The code is dynamically adjusted based on the update signal, and the data cleaning module removes invalid data according to the dynamic code verification rules.

[0030] Preferably, the coding-driven policy matching unit is specifically as follows:

[0031] Based on policy codes and enterprise codes, the policy and enterprise codes are read by conventional scanning devices, and three types of feature vectors are decomposed: explicit, implicit, and dynamic. A dynamic weight self-iterative algorithm is established by using a weighted calculation model of cosine similarity and constraint satisfaction, and the weights are adjusted in real time based on the matching effect feedback data.

[0032] Establish a scenario-based matching matrix with two dimensions: enterprise type and policy area. For micro and small enterprises, load tax incentive priority codes for priority verification, including implicit conditions such as VAT reduction and exemption. For high-tech enterprises, load R&D subsidy priority codes to strengthen the verification weight of codes including R&D expense ratio and number of patents. Real-time trigger module monitors dynamic changes in codes. When enterprise codes or policy codes change, the matching degree is automatically recalculated. The final output includes a matching list, matching degree, and key conditions.

[0033] Formula for calculating multi-dimensional policy matching degree:

[0034] ;

[0035] This indicates the final match between the policy and the enterprise, with a value ranging from 0 to 1.

[0036] This represents the weight of the dominant feature, with a value ranging from 0 to 1.

[0037] This represents the weight of the latent features, with a value ranging from 0 to 1.

[0038] This represents the dynamic feature weights, with a value range of 0 to 1.

[0039] This represents the similarity of explicit feature vectors, with a value ranging from 0 to 1.

[0040] This represents the similarity of latent feature vectors, with a value ranging from 0 to 1.

[0041] This represents the dynamic feature vector similarity, with a value ranging from 0 to 1.

[0042] This represents the constraint satisfaction correction coefficient, with a value ranging from 0 to 1.

[0043] Preferably, the scheme generation and encoding tracking unit is specifically as follows:

[0044] Based on the output of the policy matching unit driven by the coding, a coding process for policy, implementation, and effect is established. An implementation plan is automatically generated according to the matching results. A progress code including step requirements and deadlines is assigned to each step, and an effect indicator code is bound to it. An LSTM time series prediction model is set up to perform risk rating for the current step based on historical execution coding data and divide it into three levels: low, medium, and high. If the predicted risk of online application exceeding the deadline is >70%, an expedited code is automatically pushed to the enterprise, and a priority processing signal is triggered at the government approval end.

[0045] Meanwhile, a new cross-departmental coding collaboration gateway has been added to establish a mapping between the progress codes on the enterprise side and the process codes on the approval side. When regular barcode scanning devices scan the enterprise progress codes, they can obtain the current review node and estimated completion time on the approval side in real time. Through the triple coding association of policy, enterprise, and progress, the completion status and effect data of each step are bound and stored. The progress tracking module supports regular barcode scanning devices to scan the progress codes to view the real-time status and effect data, and automatically pushes early warning information related to the coding when the deadline is exceeded.

[0046] Formula for calculating the probability of risk in execution steps:

[0047] ;

[0048] In the formula: This represents the probability of the execution step being delayed, with a value ranging from 0 to 1.

[0049] This represents the overdue rate of similar steps in the past, with a value ranging from 0 to 1.

[0050] This represents the step similarity coefficient, with a value ranging from 0 to 1.

[0051] This indicates the weight of the remaining time, with a value ranging from 0.3 to 0.8.

[0052] Indicates the remaining time for the step;

[0053] and If the types are the same, it indicates the total time spent on each step;

[0054] This represents the expedited signal correction factor, with a value range of 1.0 to 1.5.

[0055] Preferably, the policy query and feedback unit is specifically as follows:

[0056] Policy codes, enterprise codes, or progress codes can be scanned using conventional scanning devices, including mobile phones and barcode scanners, to obtain associated information, including the full text of the policy, matching results, progress, and effects; and generate multi-dimensional charts, including policy value heatmaps and solution progress Gantt charts.

[0057] Simultaneously, a feedback code is generated. Users can submit their opinions by scanning the feedback code. The feedback data is then bound and stored with the policy code or enterprise code.

[0058] Preferably, the security protection and access control unit is specifically as follows:

[0059] The AES-256 algorithm is used to encrypt the transmission and storage of policy codes, enterprise codes and related data. The key is double-bound to the enterprise master code and the code version number, and only authorized users can decrypt it.

[0060] Establish a coding version and role-based permission system. Administrators have the right to read all versions of the coding system across the entire enterprise, while department users can only read the current version of the coding system within their own department. Users need to scan the authorized code to verify their identity. All coding operations are recorded, and the system has the function of tracing the track by version. Unauthorized operations trigger real-time alarms.

[0061] Preferably, the update and management unit is specifically as follows:

[0062] Regularly connect to the government policy platform coding database to synchronize policy codes and policy effectiveness benchmark codes, assign version numbers to policy and enterprise codes, archive and store old versions, and have the function of tracing historical data; compare enterprise effectiveness codes with benchmark codes, automatically adjust matching algorithm parameters, and if a policy effectiveness indicator code is lower than 80% of the policy effectiveness benchmark code for three consecutive months, it will trigger policy code adjustment and planning template update.

[0063] A mapping table between old and new codes is established. When the system is upgraded or the coding rules are adjusted, the AI ​​mapping model automatically identifies the core fields of the old codes, matches them with the new coding rules, and generates a comparison table.

[0064] The beneficial effects of this invention are as follows:

[0065] 1. This invention uses a policy coding collection unit to decompose implicit conditions through NLP semantic parsing and an industry rule engine. It combines historical execution data with a two-dimensional model of industry risk and policy effect to optimize thresholds and generate unique policy codes containing information such as region and validity period. It also connects to the government blockchain to form a traceability chain. This not only achieves quantitative constraints on implicit conditions but also ensures the authenticity of the codes, significantly reducing the number of application rejections and improving the efficiency and reliability of automated verification of policy matching.

[0066] 2. The enterprise coding integration unit of this invention captures key events such as the renewal of high-tech enterprise qualifications in real time through the dynamic perception module and triggers coding updates; the coding-driven multi-dimensional matching unit can monitor dynamic changes in coding, automatically recalculate the matching degree, and also load tax incentive and R&D subsidy priority codes for micro and small enterprises and high-tech enterprises respectively, strengthening the verification of corresponding implicit conditions; realizing real-time synchronization of dynamic information and coding, avoiding invalid applications and missed window periods, and improving the timeliness and accuracy of policy matching.

[0067] 3. The scheme generation and coding tracking unit of this invention assigns progress codes and binds effect codes to the execution steps, uses an LSTM model to rate risks, and automatically pushes expedited codes when the risk of exceeding the deadline is >70%. Cross-departmental gateways enable coding collaboration between enterprises and the approval end. The system adaptive update unit compares the enterprise's effect with the benchmark code. If the standard is not met for three consecutive months, the policy code and planning template are adjusted. This forms a matching-execution-feedback-optimization closed loop, which not only facilitates progress tracking and risk warning, but also promotes the system to continuously adapt to policy adjustments and enterprise development needs, thereby improving the efficiency of policy implementation. Attached Figure Description

[0068] Figure 1 This is a flowchart of the intelligent management system for enterprise policy matching and planning in this invention. Detailed Implementation

[0069] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] like Figure 1 As shown in the figure, this embodiment of the invention provides an intelligent management system for enterprise policy matching and planning, the system comprising:

[0071] Policy coding collection unit: crawls policy data through multi-source crawlers, decomposes implicit conditions, automatically optimizes implicit condition thresholds, establishes a mapping relationship between constraint codes and real data values, and generates unique policy codes;

[0072] Enterprise coding integration unit: The dynamic perception module captures key events and triggers update signals; the coding is adjusted based on the update signals, and the data cleaning module removes invalid signals;

[0073] Coding-driven policy matching unit: Based on policy and enterprise coding, it decomposes three types of feature vectors: explicit, implicit, and dynamic, establishes a dynamic weight self-iterative algorithm to adjust weights in real time, establishes a two-dimensional scenario-based matching matrix, and automatically recalculates the matching degree when coding changes;

[0074] Solution generation and coding tracking unit: Generates implementation plans and assigns progress codes and effect codes, performs risk rating for the current steps, and monitors progress codes and effect data through a triple coding system linking policy, enterprise, and progress.

[0075] Policy Inquiry and Feedback Unit: Information is obtained through regular QR code scanning devices and visualized charts are generated; users can submit opinions through feedback codes.

[0076] Security and Access Control Unit: Encrypts encoded and associated data; assigns different permissions to different roles; and records operation logs.

[0077] Update and Management Unit: Assign version numbers to policies and enterprise codes, archive and store old versions; compare enterprise performance codes with baseline codes, automatically adjust matching algorithm parameters, and establish a mapping table between new and old codes.

[0078] Taking the 2024 R&D subsidy policy for high-tech enterprises in a certain province and XX Technology Co., Ltd. as an example, the system implementation will be carried out as follows:

[0079] 1. Policy coding collection

[0080] A multi-source web crawler retrieves policy texts from the Science and Technology Department of a certain province every week (explicit metadata: region = a certain province, validity period 2024.6.1-12.31) and historical data from 2023 (500 applications submitted, success rate 68%).

[0081] NLP deconstructs the implicit conditions of no penalties in the past 3 years and R&D ratio ≥ 5%, and combines industry risk (α=0.3, R=0.2) and policy effect (β=0.5, E=0.82) to optimize it according to the threshold formula to no penalties in the past 2 years and R&D ratio ≥ 3.6%; connect with the government credit platform to establish a constraint code mapping, generate a unique policy code and write it into the government blockchain to achieve traceability.

[0082] Traditional systems take an average of 4.5 hours to manually break down the implicit conditions of a single policy (such as no major administrative penalties in the past 3 years, or the threshold for R&D expenses as a percentage of revenue). This system, through automatic NLP parsing and historical data-assisted threshold optimization, takes only 0.8 hours on average, improving efficiency by 82.2% and significantly reducing labor costs.

[0083] Traditional systems lack coding traceability capabilities, and the risk rate of policy codes being tampered with or forged is approximately 3.2%, which may lead to enterprises submitting applications based on false policies. This system writes policy codes into the government blockchain, forming a full-cycle traceability chain from generation to modification to repeal, reducing the risk rate of code tampering to 0%, and the authenticity can be verified simply by scanning the code, without the need for manual verification.

[0084] Traditional systems use fixed thresholds, such as the original policy requirement that R&D expenses account for ≥5%, without dynamically adjusting based on industry risks and policy effects. This results in a misjudgment rate of 18.5% for implicit conditions, such as using low thresholds for high-risk industries, leading to non-compliant applications. This system optimizes the thresholds through a dual-dimensional model of industry risk (α) + policy effect (β), reducing the misjudgment rate to 4.2% and effectively reducing invalid applications caused by unreasonable thresholds.

[0085] 2. Enterprise coding integration

[0086] The API connects to XX Technology's ERP and qualification system to collect data such as the 2023 R&D ratio of 4% and the renewal of the high-tech enterprise qualification on March 15, 2024. The dynamic perception module captures qualification renewal events, triggers code updates, and generates the main code ENT-8823 and sub-codes "ZZ-001-V2.0 (high-tech enterprise)" and "CW-001 (R&D ratio)". The data cleaning module removes invalid estimated data.

[0087] Traditional systems require manual updates of key data such as enterprise high-tech enterprise qualification renewal and R&D expense adjustments, taking an average of 2 days per update and potentially missing policy application windows. This system's dynamic sensing module captures data change logs in real time; for example, for enterprise qualification renewal events, it triggers code updates within 0.5 hours, improving efficiency by 95.8%.

[0088] Traditional systems lack data cleaning mechanisms, resulting in approximately 12% of enterprise data being invalid, such as estimated data and duplicate data, which affects matching accuracy. This system's data cleaning module removes invalid data based on dynamic coding verification rules (such as sub-codes needing to be associated with valid timestamps), reducing the invalid data rate to 2.1% and ensuring that enterprise codes correspond one-to-one with real data.

[0089] 3. Encoding-driven matching

[0090] Scan the code to read the encoding, and break down the explicit characteristics (region / industry matching, cos). θ1 =1), Implicit (R&D ratio meets the standard, cosplay) θ2 =0.95), Dynamic (Qualification valid, cosplay) θ3 =1) Feature; based on the weights (W1=0.2, W2=0.55, W3=0.25) and the matching degree formula, the matching degree is calculated to be 95.3%, which is determined to be the preferred recommendation.

[0091] Traditional systems, due to the lack of quantification of implicit conditions (such as the inability to automatically verify whether the R&D expense ratio meets the standards), result in a 35.8% rejection rate for enterprise applications, wasting manpower and time. This system, through coded verification of implicit conditions and dynamic characteristics, reduces the application rejection rate to 8.3%, a reduction of 76.8%.

[0092] Traditional systems manually verify enterprise and policy conditions, with an average time of 2.3 hours per enterprise and policy. This system automatically decomposes feature vectors and calculates matching degrees, taking only 0.1 hours, improving efficiency by 95.7%.

[0093] Traditional systems achieve only 52.1% effective recommendations (for companies that ultimately successfully apply) through keyword matching, with a large number of policies being ineffective. This system, through a two-dimensional scenario-based matching matrix (such as prioritizing tax incentive codes for micro and small enterprises) and dynamic weight iteration, increases the effective recommendation rate to 91.4%, helping companies quickly identify suitable policies.

[0094] 4. Solution Generation and Tracking

[0095] The process includes: material preparation (EXEC-001), online application (EXEC-002), and approval follow-up (EXEC-003); the LSTM model predicts that EXEC-002 will have 1 day remaining on August 24, 2024, with a 100% risk of exceeding the deadline, and will push an expedited code URG-001; the cross-departmental gateway links the enterprise with the approval end code, and scanning the code can check the preliminary review node and estimated completion time of the Provincial Department of Science and Technology.

[0096] Traditional systems lack overdue warnings, resulting in an overdue rate of 28.7% for key policy application steps (such as online application and material submission). Our system's LSTM time-series prediction model assesses risk based on historical data; when the overdue risk exceeds 70%, it automatically pushes an expedited code, reducing the overdue rate to 7.5%, a reduction of 73.9%.

[0097] Traditionally, businesses need to manually contact government departments to check approval status, which takes an average of 1.5 days. This system, however, uses a cross-departmental coding collaboration gateway to link the enterprise's progress code with the approval process code. By scanning the code, the current review status and estimated completion time can be obtained in just 0.02 hours, improving efficiency by 99.4%.

[0098] Traditional systems typically take an average of 3 months to provide feedback on policy effect data (such as subsidy payments), making it difficult to optimize matching rules in a timely manner. This system, by binding effect coding with progress coding, shortens the feedback cycle to 7 days, providing real-time data support for adjusting algorithm parameters.

[0099] The policy coding collection unit is as follows:

[0100] By regularly crawling policy texts and explicit metadata from government platforms at all levels through a multi-source crawler module, implicit conditions are decomposed using NLP semantic parsing and an industry rule engine. A dual-dimensional training model is set up to assess industry risk and policy effectiveness. Implicit condition thresholds are automatically optimized based on historical policy execution data, and effect codes are bound to them. If the matching accuracy of a policy is less than 60% due to an unreasonable implicit threshold, the threshold is automatically recalculated, and implicit conditions are converted into codes that can be quantified and constrained.

[0101] By connecting with third-party data sources such as government credit platforms, a mapping relationship between constraint codes and real data values ​​is established. Unique policy codes are generated based on regional codes, policy type codes, release timestamps, constraint code sets, and validity period codes. Supporting RFID and QR Code formats, these codes can be read and parsed by conventional barcode scanning devices. Furthermore, by connecting with local government blockchain platforms, policy codes (including constraint code sets and validity period codes) are written into the blockchain, forming a cycle of code generation, modification, and repeal traceability chain. This ensures that the authenticity of the code can be verified simultaneously when scanned by conventional barcode scanning devices, avoiding the risk of tampering.

[0102] Latent conditional dynamic threshold optimization formula:

[0103] ;

[0104] This represents the optimized implicit condition threshold, such as X for no major administrative penalties in the past X years, and Y for R&D expenses accounting for ≥ Y%. The value depends on the type of implicit condition: time-related thresholds > 0, and proportion-related thresholds 0~100%.

[0105] This represents the initial threshold before optimization, such as the threshold explicitly stated in the policy text as the past 3 years and R&D ratio ≥ 5%, with values ​​equal to... The types are consistent and determined by the original policy text or industry defaults;

[0106] This represents the industry risk coefficient weight, used to adjust the degree of influence of industry risk on the threshold. The value ranges from 0.1 to 0.5, with high-risk industries such as chemical and construction using 0.4 to 0.5, and low-risk industries such as education and consulting using 0.1 to 0.2.

[0107] This represents the industry risk index, calculated based on the rate of corporate violations in historical policy implementation data. The higher the violation rate, the higher the risk index. The larger the value, the range is 0~1, and the industry's average annual violation rate is <1%. >5% ;

[0108] This represents the policy effectiveness coefficient weight, used to adjust the degree of influence of policy implementation effect on the threshold. The value range is 0.2~0.6, with core policies such as high-tech enterprise subsidies taking 0.5~0.6 and general policies such as training subsidies taking 0.2~0.3.

[0109] This represents the historical policy effectiveness index, calculated based on the subsidy disbursement rate / application success rate, with a value ranging from 0 to 1. Excellent effectiveness is indicated by a disbursement rate > 90%. Poor performance: When the payment receipt rate is less than 50%. ;

[0110] This represents the matching accuracy feedback coefficient. If the pre-optimization threshold results in a matching accuracy <60%, then... Adjust the threshold, with a value ranging from 0.8 to 1.2, when the accuracy is <60%. Force a reduction in the threshold; when accuracy > 80% The threshold should be increased appropriately.

[0111] The specific details of the enterprise coding integration unit are as follows:

[0112] After completing the policy data coding and collection, the system connects to enterprise ERP, CRM and other systems via API to collect qualification, business and financial data. The dynamic data perception module monitors the data change log in real time, captures key events such as the renewal of high-tech enterprise qualifications and triggers update signals.

[0113] A unique master code, such as ENT-001, is assigned to each enterprise, and a sub-code is assigned to each data type, such as ZZ=qualification, forming a master code and sub-code system. The code is dynamically adjusted based on the update signal. For example, after the qualification is renewed, it is upgraded to ZZ-001-V2.0 and a timestamp is embedded. The data cleaning module removes invalid data according to the dynamic code verification rules to ensure that the enterprise data and the code correspond one-to-one, so as to provide standardized enterprise code data for the future.

[0114] The coding-driven policy matching unit is as follows:

[0115] Based on policy codes and enterprise codes, the system reads these codes using conventional scanning devices and extracts three types of feature vectors: explicit (region, industry), implicit (credit rating, R&D ratio), and dynamic (qualification version, business status). A dynamic weighted self-iterative algorithm is established using a weighted calculation model based on cosine similarity and constraint satisfaction. The weights are adjusted in real time based on the matching effect feedback data (application success rate, subsidy receipt rate).

[0116] If a tax policy results in a failure rate of over 30% due to a missed judgment of dynamic features (qualification renewal status), the weight of the dynamic features will be automatically increased from 0.3 to 0.5. If a talent policy results in matching redundancy due to misjudgment of explicit features (region), the weight of the explicit features will be reduced from 0.2 to 0.1.

[0117] A dual-dimensional scenario-based matching matrix is ​​established, encompassing both enterprise type and policy area. For micro and small enterprises, a tax incentive priority code (TA-001) is added to prioritize the verification of implicit conditions such as VAT reduction and exemption. For high-tech enterprises, a research and development subsidy priority code (RD-002) is added to strengthen the verification weight of codes such as the proportion of research and development expenses and the number of patents. A real-time trigger module monitors dynamic changes in codes, and automatically recalculates the matching degree when enterprise codes or policy codes change. The final output includes a matching list, matching degree, and key conditions.

[0118] Formula for calculating multi-dimensional policy matching degree:

[0119] ;

[0120] This indicates the final match between the policy and the enterprise, used to determine the matching priority. The value ranges from 0 to 1, where 0 represents no match and 1 represents a perfect match. The default value is... For a valid match result;

[0121] This represents the weight of explicit features, corresponding to explicit coding features such as region and industry. It is adjusted in real time by a dynamic weight self-iteration algorithm, with a value range of 0 to 1. The default initial value is 0.2 because explicit weights are the lowest. If misjudgment of region leads to matching redundancy, it can be reduced to 0.1 or below.

[0122] This represents the weight of implicit features, corresponding to implicit constraint coding features such as credit rating and R&D ratio. The value range is 0 to 1, with a default initial value of 0.5 because the implicit weight is the highest. If the application fails due to the omission of implicit conditions, it can be increased to 0.6 to 0.7.

[0123] This represents the dynamic feature weight, corresponding to dynamic coding features such as qualification version and business status. It is updated in real time with the enterprise coding iteration, and the value range is 0~1. The default initial value is 0.3, which is the second highest due to the dynamic weight. If the qualification renewal fails due to a missed judgment, it can be increased to 0.5 or above.

[0124] This represents the similarity of explicit feature vectors, calculated using explicit policy codes (such as regional codes) and explicit enterprise codes (such as registration location codes). The value ranges from 0 to 1, indicating that the vectors are completely identical. When completely inconsistent ;

[0125] Representing the similarity of implicit feature vectors, it is calculated by encoding implicit policy constraints (such as CREAA) and implicit enterprise data (such as credit rating), with a value ranging from 0 to 1, and is calculated when all implicit conditions are met. When the core conditions are not met ;

[0126] The dynamic feature vector similarity is calculated using dynamic policy coding (e.g., validity period) and enterprise coding (e.g., qualification renewal timestamp), with a value ranging from 0 to 1. A perfect dynamic match is achieved when all features are matched. When the qualification expires ;

[0127] This represents the constraint satisfaction correction coefficient, used to verify the completeness of the match between the policy constraint code set and the actual enterprise data. Its value ranges from 0 to 1, and it is used when all constraint codes are satisfied. Each missing non-core constraint Decrease by 0.1, one core constraint is missing. .

[0128] The scheme generation and coding tracking unit is specifically as follows:

[0129] Based on the output of the coding-driven policy matching unit, a coding process for policy, implementation, and effects is established. Implementation plans are automatically generated based on the matching results, such as the process for "High-tech Enterprise R&D Subsidy"—material preparation → online application → approval follow-up. Each step is assigned a progress code including step requirements and deadlines, such as EXEC-001 = Material Preparation, and bound to an effect indicator code, such as EVA-003 = Subsidy Amount Received. An LSTM time-series prediction model is set up, and based on historical execution coding data, such as the overdue rate of material preparation (EXEC-001) and the average time spent on approval follow-up (EXEC-003), a risk rating is assigned to the current step, categorized into low, medium, and high levels. If the predicted overdue risk of online application is >70%, an expedited code (URG-001) is automatically pushed to the enterprise, triggering a priority processing signal from the government approval end.

[0130] Meanwhile, a new cross-departmental coding collaboration gateway has been added to establish a mapping between the progress codes on the enterprise side and the process codes on the approval side. When regular barcode scanning devices scan the enterprise progress codes, they can obtain the current review node and estimated completion time on the approval side in real time. Through the triple coding association of policy, enterprise, and progress, the completion status and effect data of each step are bound and stored. The progress tracking module supports regular barcode scanning devices to scan the progress codes to view the real-time status and effect data, and automatically pushes early warning information related to the coding when the deadline is exceeded.

[0131] Formula for calculating the probability of risk in execution steps:

[0132] ;

[0133] In the formula: This indicates the probability of exceeding the time limit for the execution step, such as the probability of exceeding the time limit for online reporting. The value ranges from 0 to 1, and the system default is... Urgent signals and warnings are triggered in a timely manner;

[0134] This represents the historical overdue rate for similar steps, calculated based on historical execution code data (e.g., the historical overdue rate of EXEC-001), with a value ranging from 0 to 1. For example, if the historical overdue rate for the material preparation step is 30%, then... ;

[0135] The step similarity coefficient measures the degree of matching between the complexity / time consumption features of the current step and historical steps. It is output by the LSTM model and ranges from 0 to 1. A value is considered to be less than 10% when the time consumption deviation between the current step and historical steps is less than 10%. When the deviation is greater than 50% ;

[0136] This indicates the weight of remaining time, used to adjust the degree of impact of time urgency on risk. The value ranges from 0.3 to 0.8, with core steps such as approval follow-up taking 0.7 to 0.8 and auxiliary steps taking 0.3 to 0.4.

[0137] This represents the remaining time for the step, the difference between the current time and the deadline, which is positive before the deadline and negative after the deadline;

[0138] and The same type indicates the total time spent on each step. For example, the total time allotted for online application is 5 days, preset by the progress code EXEC-XX, such as EXEC-002 for the total time spent. sky;

[0139] This indicates the expedited signal correction coefficient. If expedited coding has been triggered, then... Increase the risk weight, with a value ranging from 1.0 to 1.5, if expedited processing is not triggered. Expedited processing has been triggered. Strengthen risk warning.

[0140] The policy inquiry and feedback unit is as follows:

[0141] Policy codes, enterprise codes, or progress codes can be scanned using conventional scanning devices, including mobile phones and barcode scanners, to obtain associated information. This associated information includes the full text of the policy, matching results, progress, and effects, without the need for keyword searches. It generates multi-dimensional charts, including a policy value heatmap (nodes correspond to policy codes, and colors represent the EVA-003 average) and a Gantt chart of the solution progress (task bars are associated with progress codes, and clicking them allows you to view effect data), to present the data intuitively.

[0142] Simultaneously, a feedback code is generated, which users can scan to submit their opinions. If the subsidy payment period is too long, the feedback data is bound and stored with the policy code or enterprise code.

[0143] The security protection and access control unit is as follows:

[0144] The AES-256 algorithm is used to encrypt the transmission and storage of policy codes, enterprise codes and related data. The key is double-bound to the enterprise master code and the code version number, and only authorized users can decrypt it.

[0145] Establish a coding version and role-based access control system. Administrators have the authority to read all versions of coding across the entire enterprise, such as ENT-001-ZZ-001-V1.0 / V2.0. Department users can only read the current version of coding within their department. Users need to scan authorized codes (such as employee ID RFID codes) to verify their identity. All coding operations are recorded, such as ENT-ADMIN updating ENT-001-ZZ-001 to V2.0. The system has the function of tracing the trajectory by version. Unauthorized operations trigger real-time alarms to ensure the security of coding data throughout its entire lifecycle.

[0146] This system establishes a four-level role system with a three-tiered separation of powers. The specific roles and permissions are shown in the table below:

[0147]

[0148] The update and management unit is detailed below:

[0149] Regularly connect to the government policy platform's coding database to synchronize policy codes (including new / repealed / updated clauses) and policy effectiveness benchmark codes (e.g., STD-005 = subsidy benchmark payment cycle). Assign version numbers to policy and enterprise codes, archive and store old versions, and have the function of tracing historical data. Compare enterprise effectiveness codes with benchmark codes and automatically adjust matching algorithm parameters, such as increasing the matching priority of "policies whose payment cycle meets the benchmark requirements". If a policy effectiveness indicator code is lower than 80% of the policy effectiveness benchmark code for three consecutive months, it will trigger policy code adjustment (correcting the "R&D ratio" constraint threshold) and planning template update (adding "bank statement proof" requirement to the material preparation steps).

[0150] A mapping table between old and new codes is established. When the system is upgraded or the coding rules are adjusted, the AI ​​mapping model automatically identifies the core fields of the old codes, matches them with the new coding rules, and generates a comparison table to ensure that historical data can be read normally.

[0151] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0152] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art 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 appended claims and their equivalents.

Claims

1. An intelligent management system for enterprise policy matching and planning, characterized in that: The system includes: Policy coding collection unit: This unit regularly crawls policy texts and explicit metadata from government platforms at all levels using a multi-source crawler module. It then utilizes NLP semantic parsing and an industry rule engine to deconstruct implicit constraints within policies. A dual-dimensional training model is established, considering both industry risk and policy effectiveness. Implicit condition thresholds are automatically optimized based on historical policy implementation data. Furthermore, it connects with third-party data sources, including government credit platforms, to establish a mapping relationship between constraint codes and the actual qualifications, financial data, and credit data of enterprises. Unique policy codes are generated based on region code, policy type code, publication timestamp, constraint code set, and validity period code. Enterprise coding integration unit: The dynamic perception module monitors the enterprise data change log in real time, captures key events for the renewal of high-tech enterprise qualification and triggers update signals; it assigns a unique main code to the enterprise and sub-codes to data types to form a main and auxiliary coding system; it adjusts the coding based on the update signals; and the data cleaning module removes invalid signals according to the dynamic coding verification rules. The coding-driven policy matching unit: Based on policy codes and enterprise codes, three types of feature vectors are decomposed: explicit, implicit, and dynamic. Explicit feature vectors include region and industry; implicit feature vectors include credit rating and R&D ratio; and dynamic feature vectors include qualification version and business status. A dynamic weight self-iterative algorithm is established using a weighted calculation model of cosine similarity and constraint satisfaction. The weights are adjusted in real time based on the matching effect feedback data of application success rate and subsidy receipt rate. A two-dimensional scenario-based matching matrix of enterprise type and policy field is established, and the matching degree is automatically recalculated when the code changes. Solution generation and coding tracking unit: Generates implementation plans and assigns progress codes and effect codes, performs risk rating for the current steps, and monitors progress codes and effect data through a triple coding system linking policy, enterprise, and progress. Policy Inquiry and Feedback Unit: Information is obtained through regular QR code scanning devices and visualized charts are generated; users can submit opinions through feedback codes. Security and Access Control Unit: Encrypts encoded and associated data; assigns different permissions to different roles; and records operation logs. Update and Management Unit: Assign version numbers to policies and enterprise codes, archive and store old versions; compare enterprise performance codes with baseline codes, automatically adjust matching algorithm parameters, and establish a mapping table between new and old codes.

2. The intelligent management system for enterprise policy matching and planning according to claim 1, characterized in that: The specific policy coding collection unit is as follows: By regularly crawling policy texts and explicit metadata from government platforms at all levels through a multi-source crawler module, implicit conditions are decomposed using NLP semantic parsing and an industry rule engine, and a dual-dimensional training model is set up to assess industry risk and policy effectiveness. The implicit condition thresholds are automatically optimized based on historical policy execution data, and effect codes are bound to transform the implicit conditions into codes that can be quantified and constrained. By connecting with third-party data sources, a mapping relationship between constraint codes and real data values ​​is established. Unique policy codes are generated based on regional codes, policy type codes, release timestamps, constraint code sets, and validity period codes. The policy codes are then written into the blockchain by connecting with local government blockchain platforms, forming a cycle traceability chain of code generation, modification, and repeal.

3. The intelligent management system for enterprise policy matching and planning according to claim 2, characterized in that: The specific details of the enterprise coding integration unit are as follows: After completing the policy data coding and collection, the system collects qualification, business and financial data by connecting to the enterprise system via API. The dynamic data perception module monitors the data change log in real time, captures key events including the renewal of high-tech enterprise qualifications and triggers update signals. A unique master code is assigned to each enterprise, and a sub-code is assigned to each data type, forming a master code and sub-code system. The code is dynamically adjusted based on the update signal, and the data cleaning module removes invalid data according to the dynamic code verification rules.

4. The intelligent management system for enterprise policy matching and planning according to claim 3, characterized in that: The specific details of the coding-driven policy matching unit are as follows: Based on policy codes and enterprise codes, the policy and enterprise codes are read by conventional scanning devices, and three types of feature vectors are decomposed: explicit, implicit, and dynamic. A dynamic weight self-iterative algorithm is established by using a weighted calculation model of cosine similarity and constraint satisfaction, and the weights are adjusted in real time based on the matching effect feedback data. Establish a scenario-based matching matrix with two dimensions: enterprise type and policy area. For micro and small enterprises, load tax incentive priority codes and prioritize verification of implicit conditions including value-added tax reduction and exemption. For high-tech enterprises, load R&D subsidy priority codes and strengthen the coding verification weight of R&D expense ratio and number of patents. The real-time trigger module monitors dynamic changes in coding. When enterprise coding or policy coding changes, it automatically recalculates the matching degree and finally outputs a result that includes a matching list, matching degree, and key conditions.

5. The intelligent management system for enterprise policy matching and planning according to claim 4, characterized in that: The scheme generation and encoding tracking unit is specifically as follows: Based on the output of the policy matching unit driven by the coding, a coding process for policy, implementation, and effect is established. An implementation plan is automatically generated according to the matching results. A progress code including step requirements and deadlines is assigned to each step, and an effect indicator code is bound to it. An LSTM time series prediction model is set up to perform risk rating for the current step based on historical execution coding data and divide it into three levels: low, medium, and high. If the predicted risk of online application being overdue is greater than 70%, an expedited code is automatically pushed to the enterprise, and a priority processing signal is triggered at the government approval end. At the same time, a new cross-departmental coding collaboration gateway was added to establish a mapping between the progress codes on the enterprise side and the process codes on the approval side. When regular scanning devices scan the enterprise progress codes, they can obtain the current review node and the estimated completion time on the approval side in real time. By linking policy, enterprise, and progress through a triple coding system, the completion status and effect data of each step are bound and stored together. The progress tracking module supports scanning the progress code with regular barcode scanning devices to view real-time status and effect data, and automatically pushes warning information associated with the coding when the deadline is exceeded.

6. The intelligent management system for enterprise policy matching and planning according to claim 5, characterized in that: The policy inquiry and feedback unit is as follows: Policy codes, enterprise codes, or progress codes can be scanned using conventional scanning devices, including mobile phones and barcode scanners, to obtain associated information, including the full text of the policy, matching results, progress, and effects; and generate multi-dimensional charts, including policy value heatmaps and solution progress Gantt charts. Simultaneously, a feedback code is generated. Users can submit their opinions by scanning the feedback code. The feedback data is then bound and stored with the policy code or enterprise code.

7. The intelligent management system for enterprise policy matching and planning according to claim 6, characterized in that: The security protection and access control unit is specifically as follows: The AES-256 algorithm is used to encrypt the transmission and storage of policy codes, enterprise codes and related data. The key is double-bound to the enterprise master code and the code version number, and only authorized users can decrypt it. Establish a coding version and role-based permission system. Administrators have the right to read all versions of the coding system across the entire enterprise, while department users can only read the current version of the coding system within their own department. Users need to scan the authorized code to verify their identity. All coding operations are recorded, and the system has the function of tracing the track by version. Unauthorized operations trigger real-time alarms.

8. The intelligent management system for enterprise policy matching and planning according to claim 7, characterized in that: The update and management unit is as follows: Regularly connect to the government policy platform coding database to synchronize policy codes and policy effectiveness benchmark codes, assign version numbers to policy and enterprise codes, archive and store old versions, and have the function of tracing historical data; compare enterprise effectiveness codes with benchmark codes, automatically adjust matching algorithm parameters, and if a policy effectiveness indicator code is lower than 80% of the policy effectiveness benchmark code for three consecutive months, it will trigger policy code adjustment and planning template update. A mapping table between old and new codes is established. When the system is upgraded or the coding rules are adjusted, the AI ​​mapping model automatically identifies the core fields of the old codes, matches them with the new coding rules, and generates a comparison table.

Images