A stamp duty intelligent management system and method based on artificial intelligence

Abstract

The application provides a stamp duty intelligent management system and method based on artificial intelligence, and relates to the technical field of tax informatization. The system comprises a data acquisition module, an AI analysis module, a policy rule engine module, a tax amount calculation module, a result output module and a feedback optimization module, forming a closed-loop management architecture. The method comprises contract data acquisition preprocessing, AI intelligent analysis, policy rule verification, tax amount calculation, result output and feedback optimization. The application realizes contract classification, accurate extraction of tax basis and intelligent matching of preferential policies through a customized NLP model in the tax field, realizes mandatory verification and policy updating through a dynamic policy rule engine, realizes self-optimization of the model through a feedback loop, can improve the contract classification accuracy to more than 95%, reduce manual operation by 80%, and improve the accuracy of preferential matching to 98%, greatly improving the management efficiency, accuracy and compliance of stamp duty, and being suitable for intelligent management of stamp duty of various enterprises.

Description

Technical Field

[0001] This invention relates to the fields of tax informatization, artificial intelligence and natural language processing, and more specifically, to an intelligent management system and method for stamp duty based on artificial intelligence. Background Technology

[0002] Stamp duty is a tax levied on the creation and receipt of legally binding documents in economic activities and transactions. Its scope is broad, covering 13 tax items including purchase and sale contracts, processing and contracting contracts, construction engineering survey and design contracts, construction and installation engineering contracts, property leasing contracts, freight transportation contracts, warehousing and storage contracts, loan contracts, property insurance contracts, technology contracts, property transfer documents, business ledgers, and licenses and permits. As an important minor tax in my country, stamp duty is characterized by its wide coverage, low tax burden, and the fact that taxpayers fulfill their tax obligations by purchasing and affixing stamp duty stamps themselves.

[0003] With the rapid development of my country's market economy, business activities are becoming increasingly frequent, the number of contracts is growing explosively, and contract types are becoming increasingly complex. New contract forms such as mixed business contracts, special industry contracts, cross-border contracts, framework contracts, and supplementary agreements are constantly emerging, bringing unprecedented challenges to the management of stamp duty for enterprises. According to statistics from the State Taxation Administration, the error rate in stamp duty declarations by Chinese enterprises has remained consistently high. Cases of tax audits and administrative penalties caused by problems such as incorrect contract classification, deviations in tax base extraction, omissions in enjoying preferential policies, and errors in tax calculation are increasing year by year, causing huge economic losses and reputational risks to enterprises.

[0004] Under the existing stamp duty management model, enterprises mainly rely on manual labor to complete contract processing and tax calculation, which has the following drawbacks:

[0005] 1. Low efficiency and poor accuracy in contract classification: Stamp duty taxable documents cover 13 tax categories, requiring manual judgment based on multiple dimensions such as contract name, business description, type of subject matter, and transaction content to determine the tax category. If the contract is a mixed business type (such as a "equipment procurement + installation and commissioning + technical service" contract or a "house rental + property management" contract), manual processing is prone to confusion regarding tax category, resulting in a classification error rate as high as 40%. Furthermore, the average processing time for a single contract exceeds 15 minutes. For large group enterprises with thousands of contracts per month, manual processing efficiency is completely unable to meet business needs.

[0006] 2. Inaccurate Tax Base Calculation: Stamp duty tax base calculation is subject to strict policy regulations, requiring precise differentiation between "tax-inclusive amount / tax-exclusive amount," "core transaction amount / miscellaneous fee amount," and "taxable amount / tax-exempt amount." For example, transportation contracts only use "freight" as the tax base, excluding miscellaneous fees such as loading and unloading fees, insurance fees, and storage fees. Purchase and sale contracts need to distinguish whether VAT is listed separately; if not, the full amount must be used as the tax base. Manual calculation easily overlooks key information such as "VAT separate listing clauses," "itemized amount breakdown," and "tax-exempt clauses," leading to a tax base deviation rate exceeding 35%, and consequently, errors in tax calculation.

[0007] 3. Difficulty in matching preferential policies and easy omissions: There are many types of stamp duty preferential policies, including statutory exemptions, policy-based reductions, phased reductions, and 50% reductions for small-scale taxpayers, all of which have strict scenario-based restrictions. For example, the exemption of loan contracts between financial institutions and micro and small enterprises requires simultaneous fulfillment of four conditions: "financial institution qualifications," "micro and small enterprise identification," "contract signing time," and "loan purpose." The exemption of technology development contracts requires three elements: "registration certificate from the science and technology department," "separate listing of R&D expenses," and "contract content conforming to the scope of technology development." It is difficult for humans to fully remember and accurately match all policy details, resulting in a policy omission rate of over 30%, which not only increases the unnecessary tax burden on enterprises but may also trigger tax audit risks due to misappropriation of preferential policies.

[0008] 4. Tax calculation is prone to errors and lacks verification: In the traditional model, tax calculation relies entirely on financial personnel manually applying tax rates (0.05‰ for loan contracts, 1‰ for property lease contracts, 0.3‰ for purchase and sale contracts, etc.) and preferential coefficients (0.5‰ for small-scale taxpayers, 0‰ for tax exemption), which easily leads to basic errors such as "incorrect tax rate selection", "missed multiplication of preferential coefficients", "decimal point misplacement", and "confusion of monetary units". Moreover, there is no verification mechanism, and errors can only be discovered during the tax audit stage. Enterprises have to bear multiple losses such as back taxes, late payment fees, and fines, resulting in extremely high compliance costs.

[0009] 5. Weak Compliance Traceability: Under the manual management model, contract processing records (classification basis, source of amount extraction, reference documents for preferential policies, and tax calculation process) are not retained in a structured or standardized manner, and are mostly paper records or scattered electronic spreadsheets. When tax authorities conduct audits, they need to review a large number of original contracts and manual records, resulting in extremely low traceability efficiency and making it impossible to clarify the attribution of responsibility. Once a problem occurs, it is difficult to provide evidence, leaving the company in a passive position.

[0010] 6. Poor policy adaptability: Stamp duty policies are constantly being updated and optimized, with frequent changes in tax rates, the expiration of preferential policies, and changes in tax collection and administration. Traditional manual processing methods cannot keep up with the latest policies in a timely manner, which can easily lead to errors in calculating tax amounts based on outdated policies; existing customized program systems require manual code modification to adapt to policy changes, resulting in long development cycles, high costs, and extremely poor flexibility.

[0011] Existing solutions mainly fall into two categories: the first is a purely manual processing model, relying on the professional knowledge of financial personnel to complete contract processing, which is inefficient, error-prone, and costly; the second is a simplified tax management system, which only provides tax declaration and entry functions and does not realize the full-process automation of contract parsing, policy matching, and tax calculation. Furthermore, existing systems are mostly general-purpose financial software, not customized for stamp duty business scenarios, and cannot handle complex scenarios such as mixed contracts, special industry contracts, and cross-border contracts. Their policies and rules are fixed and cannot be dynamically updated, failing to meet enterprises' needs for efficient, accurate, and compliant stamp duty management. Summary of the Invention

[0012] Therefore, the purpose of this invention is to develop an intelligent management system and method based on artificial intelligence technology, capable of realizing full-process management of stamp duty, with dynamic policy updates, model self-optimization, multiple cross-validation, and full-process traceability. Through the deep integration of artificial intelligence technology, natural language processing technology, big data technology and stamp duty policy rules, a closed-loop intelligent management system is constructed, encompassing data collection, AI analysis, policy verification, tax calculation, result output, and feedback optimization. This system solves the difficult problems in traditional stamp duty management, such as low efficiency in contract classification, inaccurate extraction of tax base, omission of preferential policy matching, errors in tax calculation, difficulty in compliance traceability, and poor policy adaptability. The goal is to achieve full-process, intelligent, precise, compliant, and traceable stamp duty management.

[0013] This invention provides an AI-based intelligent management system for stamp duty, comprising: a data acquisition module, an AI analysis module, a policy rule engine module, a tax calculation module, a result output module, and a feedback optimization module; each module is connected through a standardized data interface to collaboratively form a closed-loop management architecture of data acquisition, AI analysis, policy verification, tax calculation, result output, and feedback optimization;

[0014] The data acquisition module supports multiple input formats such as PDF, Word, scanned documents, Excel, and ERP exported data, comprehensively acquiring enterprise stamp duty taxable contract data, and performing standardized preprocessing such as text cleaning, multi-sub-contract splitting, and key paragraph marking on PDF, Word, scanned documents, and Excel contract ledgers;

[0015] The AI ​​analysis module is based on a finely tuned natural language processing (NLP) model in the tax field, which enables contract tax item classification, accurate extraction of tax base, intelligent identification of preferential conditions, and outputs analysis results and confidence scores.

[0016] The policy rule engine module stores a dynamically updated stamp duty policy rule library, performs mandatory compliance verification on the AI ​​analysis results, and updates policies and pushes change notifications.

[0017] Specifically, a three-layer cross-validation system is constructed to achieve mandatory compliance verification of the analysis results and dynamic policy updates;

[0018] The tax calculation module calculates the tax payable according to the legal formula based on the verified standardized data such as tax items, tax base, and preferential coefficients, and handles special scenarios such as unlisted amounts and mixed contracts that have not been split.

[0019] The output module generates a standardized stamp duty management report, which supports manual modification, recalculation, report export and ledger generation, and outputs archiveable PDF and Excel ledgers.

[0020] The feedback optimization module collects manually corrected data, forms a labeled dataset, and fine-tunes the AI ​​model periodically. It retains an immutable operation log throughout the entire process, enabling compliance traceability and model iterative optimization. This achieves a closed-loop iterative optimization process involving manual review, data labeling, model fine-tuning, and accuracy improvement, while retaining an immutable compliance traceability log throughout the entire process.

[0021] Furthermore, the AI ​​analysis module includes a tax item identification submodule, a tax base splitting submodule, and a preferential condition matching submodule. The AI ​​analysis module adopts the entity-relationship-attribute triple annotation rule, and the annotation objects cover entities such as contract business type, taxable amount item, taxpayer qualification, and preferential conditions. It determines the correspondence between contract type and tax item, amount item and taxable attribute, and taxpayer qualification and preferential eligibility, and performs customized fine-tuning in the tax field based on a general large model.

[0022] Preferably, the AI ​​analysis module is equipped with a confidence threshold mechanism. If the confidence level of the analysis result is ≥90%, it directly enters the policy verification stage, and if the confidence level is <90%, it triggers the manual review process.

[0023] Furthermore, the policy rule engine module includes a policy RPA update submodule, a semantic matching intelligent parsing submodule, and an IF-THEN rule verification submodule, constructing a three-layer cross-verification system of AI parsing and policy rule engine verification, tax calculation and industry benchmark comparison verification, and data collection and result output closed-loop verification.

[0024] Specifically, the AI ​​parsing and policy rule engine verification in the three-layer cross-verification system verifies the compliance of tax items, tax base, and preferential conditions; the tax amount calculation and industry benchmark value comparison verification sets floating thresholds according to the industry and issues warnings when the threshold is exceeded; and the data collection and result output closed-loop verification verifies the consistency between the original information and the declared data.

[0025] Preferably, the policy rule engine module connects to the State Taxation Administration's official website 24 / 7 via RPA technology to identify stamp duty policy changes and update the rule base, and push policy change notifications; it also mandates that financial lease contracts be verified as loan contracts, that transportation contracts be removed from loading and unloading fees, insurance fees and miscellaneous charges, and that technology development contracts be verified to have their registration number and R&D expenses listed separately.

[0026] Furthermore, the feedback optimization module includes a manual correction data collection submodule, a labeled dataset storage submodule, and an AI model fine-tuning trigger submodule, realizing the transformation of manually intervened data into model optimization capabilities.

[0027] Furthermore, the feedback optimization module also includes an operation log retention sub-module, which uses blockchain notarization technology to store the entire process operation record, forming an immutable compliance traceability chain.

[0028] Preferably, the AI-based intelligent stamp duty management system of the present invention supports integration with ERP and financial systems via HTTP / HTTPS, RESTful API, WebService, SAP RFC, and SOAP protocols, and supports three data synchronization methods: real-time incremental, scheduled batch, and manual triggering.

[0029] Specifically, the system supports seamless integration with Kingdee, Yonyou, SAP, Oracle, and enterprise-developed ERP systems. It adopts a dual mode of standardized interface and customized adaptation, and data synchronization supports three methods: real-time incremental, scheduled batch, and manual triggering. It also provides specialized parsing capabilities for special scenarios such as hybrid contracts, cross-border contracts, framework contracts, and supplementary agreements.

[0030] This invention also provides an artificial intelligence-based intelligent management method for stamp duty, applied to the artificial intelligence-based intelligent management system for stamp duty as described above, comprising the following steps:

[0031] S1. Data Acquisition and Standardized Preprocessing: Obtain taxable contract data from multiple channels, remove redundant information, intelligently split multiple sub-contracts, mark key paragraphs, and convert them into standardized structured data;

[0032] S2, AI Intelligent Analysis and Confidence Assessment: Based on a tax-specific NLP model, it completes the analysis and identification of tax item classification, tax base extraction, and preferential conditions, and outputs a confidence score; if the confidence score is ≥90%, it enters the verification stage, and if the confidence score is <90%, it triggers manual review.

[0033] S3. Policy rule engine mandatory verification: The parsing results are cross-validated in three layers through a dynamic stamp duty policy rule library. If the verification fails, it is returned for correction. If the verification passes, it proceeds to step S4 for calculation.

[0034] S4. Tax Calculation and Special Scenario Handling: The tax payable shall be calculated according to the formula "Tax Payable = Tax Base × Tax Rate × Preferential Coefficient". Contracts with unspecified amounts shall be required to submit supplementary information. For mixed contracts with unsplit amounts, the higher tax rate shall apply.

[0035] S5. Results Output and Manual Interaction: Generate a management report containing contract information, classification basis, tax calculation details, tax amount, and risk warnings, which can be manually modified and recalculated.

[0036] S6. Feedback Optimization and Compliance Traceability: Collect manually corrected data to form a high-quality labeled dataset, regularly fine-tune the AI ​​model to improve the parsing accuracy, and retain operation logs throughout the process to achieve compliance traceability.

[0037] The system of this invention continuously iterates by manually correcting data, and the accuracy rate of tax item classification remains stable at over 95%.

[0038] Furthermore, the parsing and identification in step S2 accurately matches tax items for different business scenarios, including: "procurement / supply" matches sales contracts, "leasing" matches property lease contracts, "financial leaseback" matches loan contracts, and "technology development" matches technology contracts; the preferential coefficients in step S4 are: tax exemption = 0, tax reduction of half for small-scale taxpayers = 0.5, and normal taxation = 1.

[0039] Furthermore, in step S3, the policy rule engine removes non-taxable bases such as loading and unloading fees and insurance premiums from transportation contracts, and verifies the registration number, subject qualifications, and separate amount required for preferential policies.

[0040] The policy rules engine performs special verification for special tax items: financial lease contracts are forcibly verified as loan contracts (0.05‰) to prevent them from being mistakenly classified as property lease contracts (1‰); transportation contracts are forcibly exempted from miscellaneous fees such as loading and unloading fees and insurance fees, retaining only freight as the tax base; technology development contracts are forcibly verified to have the science and technology department registration number and R&D expenses listed separately.

[0041] Furthermore, the process of collecting manually corrected data to form a high-quality labeled dataset in step S6 includes: using tax-specific OCR and fuzzy content repair on scanned contracts to correct the image, remove noise, fade seals, improve clarity, complete fuzzy content, and trigger manual confirmation if unclear identification occurs.

[0042] Furthermore, the data acquisition and standardization preprocessing operations in step S1 include:

[0043] Redundant information cleaning: Remove headers and footers, irrelevant attachments, duplicate clauses, blank paragraphs, watermarks, and other non-critical information;

[0044] Intelligent splitting of multiple sub-contracts: Identifies and accurately splits multiple independent taxable contracts within a single document;

[0045] Key paragraph markers: Locate key paragraphs in the subject matter description, monetary terms, signatory information, and preferential terms;

[0046] Data format standardization conversion: Converting heterogeneous contract data into structured data with a unified standard.

[0047] Furthermore, the management report generated in step S5 includes: contract number, signing parties, signing date, subject matter description, tax classification results and basis, tax base details (including excluded items), applicable tax rate, preferential policies and basis, tax payable, risk warning, and processing status.

[0048] The stamp duty management report clarifies the basis for each step of the process, realizes blockchain storage of the entire process operation log, and ensures clear and verifiable responsibilities; it allows for one-click traceability during tax audits, strengthens compliance traceability capabilities, and improves traceability efficiency by 90%.

[0049] This invention utilizes AI to streamline the entire process, reducing the processing time for a single contract from the traditional 15 minutes to less than 30 seconds, and ≤10 seconds for plain text contracts. This reduces manual operations by more than 80%, enabling large enterprises to complete the processing of thousands of contracts per month within 2 hours, significantly improving management efficiency.

[0050] Preferably, the system architecture of the present invention adopts a modular design, requiring only updates to the policy rule base and AI model training data, and can quickly be compatible with and expand the intelligent management of other minor taxes such as property tax, urban land use tax, and deed tax.

[0051] This invention is adaptable to contracts from various industries, including finance, construction, and technology, and supports intelligent parsing of special contracts such as cross-border contracts, framework agreements, and supplementary agreements. For loan / financial leasing contracts in the financial industry, the system accurately categorizes them into tax categories based on keywords such as "financial leaseback," "financial institutions," and "micro and small enterprises," verifying tax exemption eligibility and eliminating irrelevant fees. For mixed equipment procurement and installation contracts in the construction industry, the system can separate equipment costs and installation fees, matching them separately to purchase and sale and construction engineering tax categories; if not separated, the higher tax rate applies. For technology development / transfer contracts in the technology industry, the system identifies clauses such as "registration with science and technology departments" and "separate listing of R&D expenses," matching tax exemption benefits and verifying the required documents. For cross-border contracts, the system extracts the domestic and international subject matter, settlement currency, and taxable scope, determining taxability according to policy. Framework agreements initially estimate tax according to agreed rules, with subsequent automatic adjustments based on actual settlement amounts. Supplementary agreements are automatically linked to the main contract, synchronously updating tax categories, amounts, and preferential policies to ensure consistency in tax calculation between the main and supplementary contracts.

[0052] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the artificial intelligence-based intelligent management method for stamp duty as described above.

[0053] The present invention also provides a computer device, the computer device including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the artificial intelligence-based intelligent management method for stamp duty as described above.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] The AI-based intelligent stamp duty management system and method provided by this invention achieves the transformation from contract text to compliant data through triple protection of AI intelligent analysis, mandatory policy rule verification, and three-layer cross-verification. This effectively solves the problems of slow classification, low efficiency, high error rate, high compliance risk, and difficulty in traceability inherent in traditional single-entry and manual calculation methods. The accuracy rate of tax item classification has increased from the traditional 60% to over 95%, the error rate of tax base extraction has decreased to below 5%, the accuracy rate of preferential policy matching reaches 98%, and tax amount calculation is error-free, effectively ensuring calculation accuracy. The policy rule engine is dynamically updated 24 / 7, prompting for preferential correction conditions, policy change risks, and compliance loopholes, avoiding missed or incorrect preferential treatment and tax amount calculation errors, reducing corporate tax risks by over 90%. Policy rules are updated synchronously without manual code modification; the AI ​​model continuously self-optimizes, adapting to policy changes, new contracts, and special business scenarios. It can be widely adapted to characteristic contracts across all industries, including finance, construction, technology, manufacturing, logistics, and commerce, supporting cross-border, framework, supplementary agreement, and hybrid contracts. It is applicable to enterprises of different sizes and has broad prospects for promotion and application. Attached Figure Description

[0056] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0057] In the attached diagram:

[0058] Figure 1 A flowchart illustrating an artificial intelligence-based intelligent management method for stamp duty according to an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of the configuration of a computer device according to an embodiment of the present invention. Detailed Implementation

[0060] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of systems and products consistent with some aspects of this disclosure as detailed in the appended claims.

[0061] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0062] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0063] The embodiments of the present invention will be described in further detail below.

[0064] Example

[0065] This invention provides an AI-based intelligent management system for stamp duty, comprising a data acquisition module, an AI analysis module, a policy rule engine module, a tax calculation module, a result output module, and a feedback optimization module. Each module is seamlessly connected through a standardized data interface, forming a closed-loop management architecture that integrates data acquisition, AI analysis, policy verification, tax calculation, result output, and feedback optimization, thereby achieving a complete transformation from the original contract text to compliant declaration data.

[0066] The data acquisition module serves as the system's input point, comprehensively acquiring enterprise stamp duty taxable contract data and performing standardized preprocessing. It fully supports various input formats, including PDF, Word, Excel, scanned images, photographs, contract ledgers, and data exported from ERP systems. Preprocessing operations include four core functions: redundant information cleaning, intelligent splitting of multiple sub-contracts, key paragraph marking, and standardized data format conversion, ensuring that the input data meets AI parsing requirements.

[0067] The preprocessing operations of the data acquisition module specifically include:

[0068] Redundant information cleaning: Text cleaning algorithms are used to remove non-critical business information such as headers and footers, irrelevant attachments, duplicate clauses, blank paragraphs, watermarks and seals, and advertising information from the contract text, while retaining the core transaction terms.

[0069] Intelligent splitting of multiple sub-contracts: Based on business keywords and clause structure, it identifies multiple independent taxable contracts contained in a single document and accurately splits them into independent sub-contract texts, avoiding misclassification of mixed contracts;

[0070] Key paragraph marking: Using semantic recognition technology, key paragraphs in the contract are located and marked, such as the description of the subject matter, transaction content, monetary terms, signatory information, signing time, preferential policy related clauses, and special agreement clauses.

[0071] Data format standardization conversion: Convert contract data of different formats and structures into structured data with unified standards, unified fields, unified units, and unified coding, providing a high-quality data foundation for subsequent AI analysis.

[0072] The AI ​​analysis module is based on a pre-trained end-to-end natural language processing (NLP) model specifically for the tax field (a finely tuned version of the Doubao large model), which realizes three intelligent analysis functions: contract tax item classification, accurate extraction of tax base, and intelligent identification of preferential conditions. The analysis results output includes classification results, confidence scores, and explanations of the basis. A confidence threshold mechanism is set, and high-confidence results with a confidence score of ≥90% directly enter the policy verification stage, while low-confidence results with a confidence score of <90% trigger the manual review process to ensure the accuracy of the analysis.

[0073] The policy and rules engine module stores a dynamically updated stamp duty policy and rules database, which includes comprehensive policy information such as national legal provisions, announcements from the Ministry of Finance and the State Taxation Administration, local tax authorities' collection and management guidelines, and industry-specific regulations. It achieves two core functions: mandatory compliance verification of parsing results and dynamic policy updates. By using hard rules to block AI parsing errors and by ensuring the timeliness of policies through synchronization, it eliminates compliance risks from the source.

[0074] The tax calculation module uses standardized data verified by the policy and rules engine to accurately calculate the tax payable according to the statutory calculation formula. It has built-in special scenario processing logic to perfectly adapt to complex business scenarios such as contracts with unspecified amounts, mixed contracts with unsplit amounts, cross-border contracts, framework contracts, and supplementary agreements, ensuring tax compliance in all scenarios.

[0075] The output module serves as the system's output outlet, generating standardized, structured, and visualized stamp duty management reports. These reports include comprehensive contract information, processing logs, compliance documentation, detailed tax calculations, and risk warnings. The module supports manual intervention and modification, triggering a full-process recalculation after modification to ensure data consistency. It generates downloadable, printable, and archiveable PDF reports and Excel ledgers, perfectly meeting the needs of internal enterprise management, tax declaration, and external audits.

[0076] The feedback optimization module implements a closed-loop iterative optimization process of manual review, data annotation, model fine-tuning, and accuracy improvement; it collects manually corrected data to form a high-quality labeled dataset, and regularly fine-tunes the AI ​​parsing model to improve recognition accuracy; it stores operation logs throughout the entire process to form an immutable compliance traceability chain, meets the verification requirements of tax authorities, and enables the system to continuously self-optimize.

[0077] The AI ​​analysis module employs a specialized large-scale model based on in-depth fine-tuning in the tax field. Unlike general NLP models, it is specifically optimized for stamp duty business scenarios, and its functions include:

[0078] Contract tax category classification: A keyword and feature database dedicated to 13 stamp duty categories has been constructed. Based on business keywords, subject matter type, transaction method, and service content in the contract text, the corresponding tax category is accurately matched; for example, "purchase", "supply", and "sales" match purchase and sale contracts; "leasing" and "equipment / house / vehicle" match property lease contracts; "finance leaseback" and "finance lease" match loan contracts; and "technology development", "technology transfer", and "technology services" match technology contracts.

[0079] Accurate extraction of tax base: Using entity recognition technology, the total amount, value-added tax (distinguishing between separately listed and not separately listed), itemized amount, miscellaneous fee amount, and tax-exempt amount in the contract are accurately extracted. Non-tax base amounts are excluded in accordance with policy regulations, and the amount unit (ten thousand yuan / yuan), currency, and settlement method are marked.

[0080] Intelligent identification of preferential conditions: Through semantic understanding and condition matching technology, key information related to stamp duty preferential treatment in contracts is extracted, including taxpayer qualifications (micro and small enterprises, small-scale taxpayers, financial institutions), contract type (technology development, loans, agriculture-related), registration certificate number, signing time, tax exemption clauses, etc., to form a standardized list of preferential conditions.

[0081] The policy rules engine module includes a policy RPA update submodule, a semantic matching intelligent parsing submodule, and an IF-THEN rule verification submodule, constructing a cross-module, three-layer cross-verification system to comprehensively prevent error propagation. Specifically, it includes:

[0082] The first layer of verification: AI analysis and policy rule engine verification, to check the compliance of tax item classification, the accuracy of tax base extraction, and the completeness of preferential conditions matching, to prevent classification errors, amount errors, and mismatch of preferential treatment;

[0083] The second layer of verification: the tax amount calculation is compared with the industry benchmark value. The tax benchmark value is established according to industry and contract type, and a reasonable floating threshold is set. Warnings are issued when the threshold is exceeded to prevent abnormal calculation errors.

[0084] The third layer of verification: closed-loop verification of data collection and result output, checking the consistency between the original contract information and the final declared data, ensuring that the data is tamper-free, complete, and without deviation throughout the entire process.

[0085] The policy rule engine module's dynamic policy update mechanism is as follows: It connects to the State Taxation Administration's official website and provincial tax authorities' official websites 24 / 7 using RPA technology. Through text crawling and AI semantic matching technology, it identifies changes in stamp duty policies (tax rate adjustments, expiration / addition of preferential policies, changes in tax collection and management standards), parses policy clauses, updates the rule base, generates policy change notifications, and pushes them to the system administrator to ensure that the system's policies are always synchronized with the official policies.

[0086] The feedback optimization module includes a manual correction data collection submodule, a labeled dataset storage submodule, an AI model fine-tuning trigger submodule, and an operation log full retention submodule, realizing closed-loop optimization of manual review and model fine-tuning;

[0087] The manual correction data collection submodule fully records all correction data for low-confidence analysis results and rule validation failure results, including the result before correction, the result after correction, the basis for correction, the person making the correction, and the correction time.

[0088] The labeled dataset storage submodule standardizes and labels the corrected data according to tax item type, business scenario, and error type, forming a high-quality training dataset specifically for the tax field, which is then stored in the system database.

[0089] The AI ​​model fine-tuning trigger submodule periodically (monthly / quarterly) uses labeled datasets to fine-tune the NLP model of the AI ​​parsing module, optimizing the accuracy of keyword recognition, semantic understanding, and scene adaptation.

[0090] The operation log retention sub-module uses blockchain evidence storage technology to store the entire process operation records of all modules, forming an immutable and fully traceable compliant log chain that meets tax audit requirements.

[0091] This system supports seamless integration with existing enterprise information systems, employing a dual-mode approach of standardized interfaces and customized adaptations. Interface protocols cover HTTP / HTTPS, RESTful API, WebService, SAP RFC, and SOAP. Data synchronization supports three methods: real-time incremental synchronization, scheduled batch synchronization, and manually triggered synchronization. It is compatible with mainstream financial systems such as Kingdee, Yonyou, SAP, Oracle, and enterprise-developed ERP systems, enabling contract data synchronization and application data feedback to ensure business continuity.

[0092] The AI-based intelligent stamp duty management system of this invention provides specialized analysis capabilities for special contract scenarios, specifically including:

[0093] Mixed business contracts: The amounts of different business types are split and matched with the corresponding tax items. The amounts that are not split are taxed at the higher tax rate.

[0094] Cross-border contracts: Identify the subject matter of domestic and foreign transactions, settlement currency, taxable scope, and determine whether taxation is applicable according to policy;

[0095] Framework contract: Taxes are initially estimated according to the agreed rules, and adjustments are made later based on the actual settlement amount;

[0096] Supplementary Agreement: Link to the main contract and update tax items, amounts, and preferential policies synchronously to ensure consistency in tax calculation.

[0097] A preferred embodiment of the present invention integrates the AI-based intelligent stamp duty management system with the enterprise ERP system, as follows:

[0098] Interface pattern: Adopts RESTful API standardized interface;

[0099] Synchronization method: Contract data is synchronized in real time and incrementally, and the results are sent back after the application is completed;

[0100] Compatible systems: Kingdee Cloud Starry Sky, Yonyou U9, SAPS / 4HANA, Oracle, and enterprise-developed systems;

[0101] Data security: Data is transmitted using de-identified and encrypted methods to ensure data security;

[0102] Business continuity: Seamlessly integrates with existing business processes without requiring changes to existing operating habits.

[0103] Through system integration, contract data acquisition and application data feedback were achieved, further improving management efficiency.

[0104] This invention also provides an intelligent management method for stamp duty based on artificial intelligence, such as... Figure 1 As shown, it includes the following steps:

[0105] S1. Data Acquisition and Standardized Preprocessing: Data on stamp duty taxable contracts of enterprises are acquired through multiple channels, redundant information is removed using text cleaning algorithms, multiple sub-contracts are intelligently split, key paragraphs are marked, and the data is converted into standardized structured data.

[0106] S2, AI Intelligent Analysis and Confidence Assessment: Based on a tax-specific NLP model, it completes contract tax item classification, accurate extraction of tax base, and analysis and identification of preferential conditions, outputting analysis results and confidence scores; confidence scores ≥90% directly enter the policy verification stage, while confidence scores <90% trigger the manual review process;

[0107] Specifically, the AI ​​analysis and recognition adopts the entity-relationship-attribute triplet annotation rule. The annotation objects cover entities such as contract business type, taxable amount item, taxpayer qualification, preferential conditions, etc., and clarify the correspondence between contract type and tax item, amount item and taxable attribute, and taxpayer qualification and preferential eligibility.

[0108] S3. Policy and rule engine mandatory compliance verification: Through the dynamically updated stamp duty policy and rule engine, the AI ​​parsing results are cross-verified in three layers. Data that fails the verification is returned for correction, and data that passes the verification enters the tax calculation stage.

[0109] The policy rules engine performs special verification for special tax items: financial lease contracts are forcibly verified as loan contracts (0.05‰) to prevent them from being mistakenly classified as property lease contracts (1‰); transportation contracts are forcibly exempted from miscellaneous fees such as loading and unloading fees and insurance fees, retaining only freight as the tax base; technology development contracts are forcibly verified to have the science and technology department registration number and R&D expenses listed separately.

[0110] S4. Tax Calculation and Special Scenario Handling: Calculate the tax payable according to the legal formula based on the verified tax item, tax rate, tax base, and preferential coefficient; handle special scenarios such as contracts with unspecified amounts, mixed contracts without split amounts, and cross-border contracts in accordance with policy regulations.

[0111] The formula for calculating tax is: Tax payable = Tax base × Applicable tax rate × Preferential coefficient;

[0112] The preferential coefficients are as follows: tax exemption = 0, tax reduction of half for small-scale taxpayers = 0.5, and normal taxation = 1.

[0113] S5. Results Output and Manual Interaction: Generates a stamp duty management report containing comprehensive information, displaying the processing flow, compliance basis, tax amount details, and risk warnings; supports manual modification of parsing results, triggering a full-process re-verification and calculation after modification;

[0114] The generated report includes: contract number, signing parties, signing date, description of the subject matter, tax classification results and basis, details of tax base (including excluded items), applicable tax rate, preferential policies and basis, tax payable, risk warning, and processing status.

[0115] S6. Feedback Optimization and Compliance Traceability: Collect manually corrected data to form a high-quality labeled dataset, and regularly fine-tune the AI ​​model to improve the parsing accuracy; retain operation logs throughout the entire process to form an immutable compliance traceability chain.

[0116] For scanned contracts, a tax-specific OCR + intelligent repair technology for blurry content is used: first, the scanned document is corrected, stains and noise are removed, seal watermarks are faded, and text clarity is improved. Then, the key content is identified using the recognition rules that conform to the stamp duty business, blurry content is filled in, and unclear parts are triggered for manual confirmation.

[0117] Application examples

[0118] In a real-world application scenario of a construction company, the overall system architecture of this invention includes a data acquisition module, an AI analysis module, a policy and rule engine module, a tax calculation module, a result output module, and a feedback optimization module. These modules are interconnected through standardized data interfaces to form a closed-loop management architecture.

[0119] The data acquisition module supports input in all formats, including PDF, Word, scanned documents, and Excel. Preprocessing functions include: cleaning headers and footers, splitting mixed procurement and transportation contracts, marking monetary clauses, and converting data to standardized formats. This module solves the problems of messy contract data formats and cumbersome preprocessing in traditional contracts, providing high-quality input data for AI parsing.

[0120] The AI ​​analysis module, based on a finely tuned Doubao model, analyzes the equipment procurement and installation contracts of the construction company: it identifies the procurement portion as a sales contract and the installation portion as a construction project contract, extracts the equipment cost and installation fee as the tax base, eliminates miscellaneous fees, identifies the company as a small-scale taxpayer eligible for a 50% tax reduction, and with a 96% confidence level, proceeds directly to the verification stage.

[0121] The AI ​​parsing module differs from general NLP models and is custom-built for the tax field. In this application example, the AI ​​parsing module integrates three dedicated sub-modules:

[0122] Stamp Duty Item Recognition Submodule: It has built-in keyword library, feature library, and sample library for 13 tax items, and adopts entity-relation-attribute triple annotation rules to optimize recognition accuracy for mixed contracts and special contracts;

[0123] Tax base breakdown sub-module: accurately identifies tax-inclusive / tax-exclusive, core / miscellaneous fees, taxable / tax-exempt amounts, and supports processing in multiple currencies, multiple units, and multiple settlement methods;

[0124] The preferential conditions matching submodule covers all stamp duty preferential policies, matching conditions such as the entity's qualifications, contract type, time, and number to form a preferential determination conclusion.

[0125] Through specialized fine-tuning in the tax field, the AI ​​analysis module has improved the recognition accuracy by more than 40% compared to general NLP models.

[0126] The policy rule engine module stores the latest stamp duty policy database and verifies the parsed results: confirming that the tax rate for purchase and sale contracts is 0.3‰, the tax rate for construction contracts is 0.3‰, the tax reduction coefficient for small-scale taxpayers is 0.5, and the amount after deducting miscellaneous fees is compliant; the verification passes. The policy rule engine module eliminates AI parsing errors through hard rules, ensuring compliance.

[0127] The policy rule engine employs a three-layer cross-validation system that includes dynamic updates and mandatory verification. The policy RPA update submodule monitors the State Taxation Administration's website 24 / 7. When a certain stamp duty preferential policy expired in 2025, the system identified and updated the rule base, pushing notifications to administrators. The semantic matching intelligent parsing submodule parses the new policy clauses and converts them into system-executable rules. The IF-THEN rule verification submodule sets hard verification rules for special scenarios such as finance lease contracts, transportation contracts, and technology development contracts to prevent errors. This three-layer cross-validation system comprehensively blocks the propagation of errors, ensuring 100% tax compliance.

[0128] The tax calculation module calculates the tax amount using the formula: Tax payable = (Equipment cost × 0.3‰ + Installation fee × 0.3‰) × 0.5, providing an accurate tax amount and handling scenarios where mixed contracts are split for tax calculation.

[0129] The output module generates standardized management reports, which include contract information, classification basis, tax calculation details, tax amount, and risk warnings. It supports exporting to PDF and Excel.

[0130] The feedback optimization module records the entire parsing process log, ensuring stable model accuracy without manual correction. The manual correction data collection submodule, for a specific contract with an 85% confidence level in AI parsing, manually reviews and corrects the tax category classification, recording the correction data. The labeled dataset storage submodule standardizes and labels the corrected data, adding it to the training dataset. The AI ​​model fine-tuning trigger submodule fine-tunes the model using monthly dataset usage, optimizing the recognition accuracy for this type of contract. The operation log retention submodule utilizes blockchain to store all operation records, allowing for easy traceability. Through closed-loop iteration, the system accuracy continuously improves, eventually stabilizing above 99%.

[0131] This application example demonstrates the intelligent management process of stamp duty based on artificial intelligence, which includes the following main steps:

[0132] S1. Data Acquisition and Preprocessing: Obtain a scanned copy of a technology development contract from a technology company, convert it into text using OCR, remove redundant information, and mark key clauses.

[0133] S2, AI Intelligent Analysis: Identifies it as a technology contract, extracts the contract amount, science and technology department registration number, and R&D funding information, and confirms that it meets the tax exemption conditions with a confidence level of 92%.

[0134] This application example analyzes several special contracts, including:

[0135] Mixed contracts: equipment procurement + installation + technical services, with the amount split into three parts, each matched with the purchase and sale, construction engineering, and technical tax categories respectively;

[0136] Cross-border contracts: Identify the portion of the transaction that is taxed domestically and the portion that is tax-exempt overseas;

[0137] Framework contracts: Tax is calculated based on the estimated amount initially, and supplementary tax is reported after the actual settlement.

[0138] Supplementary Agreement: Related to the main contract, with synchronized preferential policies.

[0139] This invention system is adapted to special scenarios and solves the problem that traditional management cannot handle complex contracts.

[0140] S3. Policy and rule verification: Verify the tax exemption conditions of the technology development contract, confirm the validity of the registration number and the separate listing of R&D expenses, and the verification is passed;

[0141] S4. Tax Calculation: With a tax exemption coefficient of 0, the tax payable is 0.

[0142] S5. Output: Generate a tax exemption management report, indicating that no stamp duty needs to be paid;

[0143] S6. Feedback optimization: No manual correction, log retention, model optimization.

[0144] The system performance parameters and actual application effects of this invention in an application example are verified as follows:

[0145] The specific system performance parameters are as follows:

[0146] Maximum contract processing capacity: ≥1 million contracts per server per year; cluster deployment can be expanded indefinitely.

[0147] Average processing time for a single contract: ≤30 seconds (plain text contracts ≤10 seconds, scanned copies including OCR ≤30 seconds);

[0148] Concurrency processing capability: Supports ≥100 concurrent connections, with a peak concurrent connection of ≥500 connections;

[0149] System response: Page operation response time ≤ 1 second, data query response time ≤ 2 seconds;

[0150] Data storage: Supports data storage of millions of contracts, with query response times in seconds.

[0151] The application results for small and medium-sized enterprises (average 50-300 contracts per month) are as follows:

[0152] Processing efficiency increased by 85%, while manual workload decreased by 80%;

[0153] The tax calculation error rate has been reduced from 25% to below 3%.

[0154] The rate of missed offers has been reduced from 30% to less than 2%.

[0155] Labor costs can be reduced by 60%-70%.

[0156] The application results for large groups / multiple subsidiaries (average monthly contracts of 1000-5000) are as follows:

[0157] Cross-entity data collection efficiency improved by 90%;

[0158] 50% of finance and human resources will be freed up.

[0159] Annual savings of 300,000 to 800,000 yuan in stamp duty management personnel costs;

[0160] The application period has been reduced from 3 days to 2 hours;

[0161] The group's compliance risks have been reduced by 90%.

[0162] Application effects of financial agency:

[0163] The number of clients that a single organization can manage has increased threefold;

[0164] Service error rate reduced by 95%.

[0165] The application examples of this invention have realized the intelligent transformation of the entire process of stamp duty management, significantly reducing management costs. Small and medium-sized enterprises can save 60%-70% of their labor costs, while large groups can save 300,000-800,000 yuan in stamp duty management labor costs annually. The declaration cycle has been reduced from the traditional 3 days to 2 hours, and the overall management cost has been reduced by more than 50%.

[0166] This invention also provides a computer device. Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention; see the accompanying drawings. Figure 2As shown, the computer device includes: an input device 23, an output device 24, a memory 22, and a processor 21; the memory 22 is used to store one or more programs; when the one or more programs are executed by the one or more processors 21, the one or more processors 21 implement the artificial intelligence-based intelligent stamp duty management method provided in the above embodiments; wherein the input device 23, the output device 24, the memory 22, and the processor 21 can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0167] The memory 22, as a read / write storage medium for a computing device, can be used to store software programs and computer-executable programs, such as the program instructions corresponding to the AI-based intelligent stamp duty management method described in this embodiment of the invention. The memory 22 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the device. Furthermore, the memory 22 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 22 may further include memory remotely located relative to the processor 21, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0168] Input device 23 can be used to receive input digital or character information, and generate key signal inputs related to user settings and function control of the device; output device 24 may include display devices such as a display screen.

[0169] The processor 21 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 22, thereby realizing the above-mentioned artificial intelligence-based intelligent management method for stamp duty.

[0170] The computer equipment provided above can be used to execute the AI-based intelligent stamp duty management method provided in the above embodiments, and has corresponding functions and beneficial effects.

[0171] This invention also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the AI-based intelligent stamp duty management method provided in the above embodiments. The storage medium can be any type of memory device or storage device, including: mounting media such as CD-ROM, floppy disk, or magnetic tape; computer system memory or random access memory such as DRAM, DDRRAM, SRAM, EDORAM, Rambus RAM, etc.; non-volatile memory such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements; the storage medium may also include other types of memory or combinations thereof; furthermore, the storage medium may reside in a first computer system in which the program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet); the second computer system can provide program instructions to the first computer for execution. The storage medium includes two or more storage media that can reside in different locations (e.g., in different computer systems connected via a network). The storage medium can store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0172] Of course, the computer-executable instructions provided in the embodiments of the present invention are not limited to the artificial intelligence-based intelligent management method for stamp duty as described in the above embodiments, but can also execute related operations in the artificial intelligence-based intelligent management method for stamp duty provided in any embodiment of the present invention.

[0173] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0174] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent management system for stamp duty based on artificial intelligence, characterized in that, include: Data acquisition module, AI analysis module, policy and rule engine module, tax calculation module, result output module, and feedback optimization module; Each module connects via a standardized data interface; The data acquisition module is used to acquire enterprise stamp duty taxable contract data, and to perform text cleaning, multi-sub-contract splitting, and key paragraph marking preprocessing on PDF, Word, scanned and Excel contract ledgers. The AI ​​analysis module is based on a finely tuned Natural Language Processing (NLP) model in the tax field, enabling contract tax item classification, accurate extraction of tax base, and intelligent identification of preferential conditions. The policy rule engine module stores a dynamically updated stamp duty policy rule library, performs mandatory compliance verification on the AI ​​analysis results, and updates policies and pushes change notifications. The tax calculation module calculates the tax payable based on the verified tax item, tax base, and preferential coefficient, and handles special scenarios such as unspecified amounts and mixed contracts that have not been split. The output module generates a standardized stamp duty management report, which supports manual modification, recalculation, report export, and ledger generation. The feedback optimization module collects manually corrected data, forms a labeled dataset, and fine-tunes the AI ​​model periodically. It also retains an unalterable operation log throughout the entire process, enabling compliance traceability and model iteration optimization.

2. The artificial intelligence-based intelligent management system for stamp duty according to claim 1, characterized in that, The AI ​​analysis module includes a tax item identification submodule, a tax base splitting submodule, and a preferential condition matching submodule. The AI ​​analysis module adopts the entity-relationship-attribute triple annotation rule. The annotation objects cover contract business type, taxable amount item, taxpayer qualification, and preferential conditions. It determines the correspondence between contract type and tax item, amount item and taxable attribute, and taxpayer qualification and preferential eligibility. It uses a general large model as a base for customized fine-tuning in the tax field.

3. The artificial intelligence-based intelligent management system for stamp duty according to claim 1, characterized in that, The policy rule engine module includes a policy RPA update submodule, a semantic matching intelligent parsing submodule, and an IF-THEN rule verification submodule, constructing a three-layer cross-verification system: AI parsing and policy rule engine verification, tax calculation and industry benchmark comparison verification, and data collection and result output closed-loop verification.

4. The artificial intelligence-based intelligent management system for stamp duty according to claim 1, characterized in that, The feedback optimization module includes a manual correction data collection submodule, a labeled dataset storage submodule, and an AI model fine-tuning trigger submodule, realizing the transformation of manually intervened data into model optimization capabilities.

5. The artificial intelligence-based intelligent management system for stamp duty according to claim 4, characterized in that, The feedback optimization module also includes a sub-module for full retention of operation logs; it uses blockchain evidence storage technology to store the entire process operation records, forming an immutable and compliant traceability chain.

6. An artificial intelligence-based intelligent management method for stamp duty, applied to the artificial intelligence-based intelligent management system for stamp duty as described in any one of claims 1-5, characterized in that, Includes the following steps: S1. Obtain taxable contract data through multiple channels, remove redundant information, intelligently split multiple sub-contracts, mark key paragraphs, and convert them into standardized structured data; S2. Based on a tax-specific NLP model, complete the classification of tax items, extraction of tax base, and analysis and identification of preferential conditions, and output confidence scores; If the confidence level is ≥90%, the application proceeds to verification; if the confidence level is <90%, manual review is triggered. S3. Perform three-level cross-validation on the parsing results through a dynamic stamp duty policy rule base. If the validation fails, return to the correction; if the validation passes, proceed to step S4 for calculation. S4. Calculate the tax payable according to the formula: Tax payable = Tax base × Tax rate × Preferential coefficient. If the amount is not listed in the contract, please submit a supplementary report. For mixed contracts, the amount that is not split shall be taxed at the higher tax rate. S5. Generate a management report containing contract information, classification basis, tax calculation details, tax amount, and risk warnings. Manually modify and recalculate. S6. Collect manually corrected data to form a high-quality labeled dataset, fine-tune the AI ​​model regularly, and retain operation logs throughout the process to achieve compliance traceability.

7. The intelligent management method for stamp duty based on artificial intelligence according to claim 6, characterized in that, The parsing and identification in step S2 matches tax items for different business scenarios, including: matching purchase / supply with sales contracts, leasing with property lease contracts, finance leaseback with loan contracts, and technology development with technology contracts. The preferential coefficients in step S4 are: tax exemption = 0, tax reduction of half for small-scale taxpayers = 0.5, and normal taxation = 1.

8. The artificial intelligence-based intelligent management method for stamp duty according to claim 6, characterized in that, Step S6 involves collecting manually corrected data to form a high-quality labeled dataset, which includes: using tax-specific OCR and fuzzy content repair on scanned contracts to correct the image, remove noise, fade seals, improve clarity, complete fuzzy content, and trigger manual confirmation if unclear recognition is not possible.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the artificial intelligence-based intelligent management method for stamp duty as described in any one of claims 6-8.

10. A computer device, the computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the artificial intelligence-based intelligent management method for stamp duty as described in any one of claims 6-8.

Images