A tax accounting and declaration method based on a finance and tax ontology

Abstract

This invention provides a tax accounting and declaration method based on a tax ontology, including: acquiring intelligent accounting data; mapping it to an ontology instance based on a pre-constructed tax ontology and extracting core accounting data sources; performing automatic accounting for multiple taxes by combining accounting rule constraints of the tax ontology and reasoning rules of the tax knowledge graph, while simultaneously performing semantic verification; converting the accounting results into declaration data and generating declaration forms, and performing multi-dimensional verification on the declaration forms; automatically submitting the declaration forms electronically; receiving the parsed results and triggering corresponding processing flows; and simultaneously feeding back the entire process data to related modules to drive rule iteration updates. This invention uses a tax ontology as the core semantic constraint, integrates knowledge graph reasoning and retrieval enhancement generation technologies, and achieves fully automated processing from data adaptation, intelligent accounting, automatic declaration to feedback optimization.

Description

Technical Field

[0001] This invention relates to the field of financial and tax data processing technology, specifically to a tax accounting and declaration method based on financial and tax ontology. Background Technology

[0002] In the field of intelligent financial and tax technology, existing tax accounting and declaration processing still generally rely on manual assistance or simple template logic, failing to achieve intelligent collaboration throughout the entire process of bookkeeping, accounting, and declaration.

[0003] First, there is a disconnect between the accounting and bookkeeping processes at the data level. The data source for accounting needs to be manually extracted from the accounting vouchers and re-entered, which makes it impossible to achieve integrated automatic processing of the two processes. This results in low data processing efficiency and is prone to human error.

[0004] Secondly, the accounting rules lack the ability to understand the deeper semantics of the financial and tax field. Existing technologies mostly use fixed templates or keyword matching to configure rules, which cannot make accurate accounting judgments based on business semantics, accounting systems, and tax attributes. For example, it cannot distinguish whether a procurement expenditure should be classified as a period expense or a fixed asset, and the professional accuracy of the accounting highly depends on manual intervention.

[0005] Secondly, the accounting and reporting processes are technically disconnected. After accounting is completed, data verification and reporting forms must be filled out manually, which is cumbersome and prone to inconsistencies between the reported and accounting data. Furthermore, the verification mechanisms for reported data are mostly limited to format and simple balance checks, lacking the ability to conduct in-depth compliance reviews that combine tax regulations and business semantics.

[0006] In addition, the existing accounting and reporting rules are mostly static and general configurations, which do not have the ability to be adapted to different tax types, industries or enterprise sizes. They also cannot automatically synchronize and adjust the rules according to external changes such as updates to tax regulations and iterations of domain knowledge. As a result, the system has high maintenance costs and poor timeliness.

[0007] From a technical architecture perspective, tax accounting and declaration modules typically operate in isolation from other intelligent tax modules such as ontology construction, multimodal data fusion, knowledge graphs, and voucher generation, failing to form a collaborative closed loop for data exchange and knowledge feedback. Furthermore, the lack of automated analysis and utilization mechanisms for external feedback information such as declaration rejections and audits hinders the continuous improvement of accounting and declaration processing quality through technological iterations. When errors occur in accounting, existing technologies cannot automatically identify the error type and generate corrective solutions, resulting in insufficient system fault tolerance. Summary of the Invention

[0008] This invention is made to solve the above-mentioned problems, and its purpose is to provide a tax accounting and declaration method based on the subject of finance and taxation.

[0009] This invention provides a tax accounting and declaration method based on the subject of finance and taxation, characterized by the following steps: S1. Obtain intelligent accounting data. Based on the pre-built financial and tax domain ontology, map the intelligent accounting data to ontology instances and extract the core accounting data source. S2. Based on the core accounting data source and preset accounting rules, combined with the accounting rule constraints in the tax and finance domain ontology and the reasoning rules in the tax and finance knowledge graph, perform automatic accounting for multiple tax types and perform semantic verification simultaneously during the accounting process. S3. Convert the verified accounting results into declaration data, generate declaration forms, and perform multi-dimensional verification on the declaration forms; S4. Automatically submit verified tax return forms to the tax return system electronically, receive and parse the returned return results, trigger corresponding processing flows based on the results, and feed back the entire process data to related modules to drive iterative updates of accounting and return rules.

[0010] Furthermore, in steps S1 and S2, the tax domain ontology predefines the concepts, attributes, relationships between concepts, and rule constraints based on the concept and attribute definitions in the tax domain. The tax knowledge graph stores the association relationships of entity instances and business experience rules. The tax domain ontology and the tax knowledge graph work together to provide rigid rule verification and flexible logical reasoning in the accounting and verification process.

[0011] Furthermore, in step S2, semantic verification is performed synchronously during the accounting process, including checking whether the accounting logic, data associations, and deduction standards used in the accounting process comply with the rule constraints of the tax and finance domain ontology and the reasoning results of the tax and finance knowledge graph.

[0012] Furthermore, after performing automatic accounting for multiple tax types in step S2, the method also includes automatically identifying the type of accounting error when the accounting result fails the preliminary verification. The system calls upon a large language model enhanced by retrieval and generation techniques, combines it with the tax and finance domain ontology, tax and finance knowledge graph, and historical amendment example library, to generate an automatic correction scheme and execute the correction until the verification is passed.

[0013] Furthermore, in step S3, multi-dimensional verification is performed on the declaration form, including data consistency verification, which checks the consistency between the declared data and the accounting data; Compliance verification verifies whether the submitted data complies with tax regulations and the rules governing the tax field; and Semantic logic verification, combined with a financial and tax knowledge graph, verifies the logical consistency between the declared data and the business semantics and tax attributes.

[0014] Furthermore, in step S4, the corresponding processing flow is triggered based on the application result, including automatically parsing the rejection reason code if the application result is rejection, adjusting the application form according to the error correction rules, and resubmitting it. If the reason for rejection cannot be automatically corrected, an exception report will be generated and pushed to the human terminal.

[0015] Furthermore, step S4 drives the iterative updates of accounting and reporting rules, including monitoring the triggering conditions for updates to financial and tax regulations, iterations of the financial and tax domain ontology version, and updates to the financial and tax knowledge graph rule base. When structured updates are detected, the updated content is automatically parsed, a rule update plan is generated and applied automatically, and an impact assessment report is generated and manual rollback is supported. When unstructured updates are detected, update suggestions and impact analysis reports are generated and pushed to a human terminal for confirmation before the update is executed.

[0016] Furthermore, in step S4, the data from the entire process is fed back to the associated modules to drive the iterative updates of the accounting and reporting rules, including feedback from the collection of reporting results, feedback from associated modules, and quality assessment feedback, forming a feedback dataset. The large language model enhanced by retrieval enhancement generation technology is used to analyze the feedback dataset, identify the optimization direction of accounting rules and declaration rules, generate optimization instructions, and perform automatic iteration.

[0017] Furthermore, the method is executed automatically by a computer program; when the core data source is missing, resulting in the inability to automatically calculate, the application is rejected and automatic correction fails, or unstructured regulatory updates involve significant rule adjustments, a manual intervention node is triggered.

[0018] The role and effect of invention This invention presents a tax accounting and declaration method based on a tax ontology. Using a tax ontology as the core semantic constraint framework, it automatically maps intelligent accounting data to ontology instances, eliminating data disconnect and manual secondary entry in the accounting and bookkeeping stages, and achieving integrated automatic flow from data acquisition to accounting processing. Through the collaborative work of the tax ontology and tax knowledge graph, semantic verification is performed synchronously during the accounting process, automatically identifying deep-seated problems such as accounting logic anomalies and data association errors. When errors occur in the accounting results, a large language model enhanced by retrieval enhancement generation technology can automatically generate and execute correction schemes in conjunction with historical amendment examples, significantly improving the system's fault tolerance. In the declaration stage, a multi-dimensional verification system ensures the accuracy of declaration data from multiple levels, including data consistency, compliance, and semantic logic, effectively reducing the risk of declaration rejection. Simultaneously, a dynamic rule update mechanism and a continuous optimization loop based on external real feedback enable the accounting and declaration rules to respond promptly to changes in tax policies, achieving sustainable improvement in processing quality. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a tax accounting and declaration method based on the financial and tax ontology of the present invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following embodiments are described in detail with reference to the accompanying drawings.

[0021] like Figure 1 As shown, this invention provides a tax accounting and declaration method based on the tax and finance ontology. This method is automatically executed by a computer program, with the entire process proceeding sequentially and setting up verification branch logic. When the core data source is missing, resulting in the inability to automatically calculate, the declaration is rejected and automatic correction fails, or unstructured regulations are updated, involving significant rule adjustments, a manual intervention node is triggered. All other nodes are executed automatically by default, while retaining the right to manual intervention.

[0022] This example uses a VAT general taxpayer enterprise that applies corporate accounting standards to process a transaction involving the purchase of office supplies. In actual operation, this method supports automatic accounting for batch vouchers throughout the entire period and declaration of all tax types.

[0023] S1. Obtain intelligent accounting data. Based on the pre-built financial and tax domain ontology, map the intelligent accounting data to ontology instances and extract the core accounting data source.

[0024] In this step, the tax and finance domain ontology is pre-constructed and responsible for defining the hierarchy, attributes, and hard rule constraints of domain concepts. This ontology is organized in a structured form, and its core components include: concept classes, used to define basic entities and business categories in the tax and finance domain, such as enterprises, taxpayers, invoices, administrative expenses, input tax, output tax, and procurement transactions; object attributes, used to describe the relationships between concepts, such as an enterprise being a taxpayer, a taxpayer generating an invoice, an invoice being mapped to an original voucher, and procurement transactions generating input tax; data attributes, used to describe the characteristic parameters of the concept itself, such as an invoice having a tax-exclusive amount and an invoice having a tax rate; and rule constraints, used to define the conditions and conclusions based on concepts and attributes, forming the core of the ontology's semantic constraints. For example, if the invoice's product name includes "printing paper" and the department using it is the administrative department, then the expense category is administrative expenses - office supplies expenses, and if it meets the pre-tax deduction requirements at the tax level, the pre-tax deduction status is "fully deductible." The ontology concepts support multi-level expansion; the administrative expenses general ledger account can be expanded with detailed concepts such as office supplies expenses and travel expenses.

[0025] The tax knowledge graph and the tax domain ontology work together. The tax domain ontology is responsible for defining the hierarchy, attributes, and hard rule constraints of domain concepts, while the tax knowledge graph is responsible for storing the relationships between entity instances and business experience rules, such as the correspondence between specific companies and specific invoices, and common tax risk points in a certain industry. The combination of the two provides dual protection of "rigid rule verification and flexible logical reasoning" in subsequent accounting and verification processes.

[0026] The large language model invoked in this method is enhanced using Retrieval Enhancement Generation (RAG) technology. The RAG knowledge base externally stores structured and unstructured documents such as tax and financial regulations, accounting systems, and historical expert amendment examples. When the large language model needs to generate a revised solution, the system first retrieves the historical amendment example library to provide the correct handling methods for similar problems; when the large language model needs to perform deep semantic verification, the system first retrieves the regulations library to provide the original legal texts, thereby improving the accuracy and interpretability of the output.

[0027] The financial and tax intelligent agent is an intelligent unit for financial and tax business processing with autonomous decision-making, task execution and autonomous learning capabilities. In this method, it can serve as a scheduling center to coordinate and call on the capabilities of ontology, knowledge graph and large language model to complete intelligent operations such as anomaly diagnosis and rule optimization.

[0028] After the program starts, it first performs system initialization, loading the aforementioned tax and finance domain ontology, tax and finance knowledge graph, RAG-enhanced large language model, tax and finance intelligent agent configuration data, as well as verified agency accounting voucher data, tax and finance law database, tax type database, historical accounting and declaration data, etc. After initialization, the program establishes semantic relationships between various data to ensure seamless integration between various technical modules.

[0029] Subsequently, the program configures adapted accounting and declaration rules based on the user's personalized needs. The program extracts personalized requirements from the user's configuration data and business needs, such as the need to focus on the additional deduction of R&D expenses and adapt to the specific declaration form format of the local electronic tax bureau. Combining the tax and financial system and expert rules, the program generates a personalized accounting and declaration rule set and pushes it to the user's terminal for confirmation.

[0030] After completing the rule configuration, the program calls the multimodal data fusion module and the voucher generation module to receive intelligent accounting data and perform adaptation and preprocessing. In this step, the multimodal data fusion module has pre-completed OCR recognition, structuring, and preliminary ontology annotation of raw data such as invoice images and bank receipts, for example, marking text areas in the images as fields such as invoice number, amount excluding tax, and tax amount. This step directly calls its output standardized structured data to perform secondary ontology mapping and core field extraction for accounting dimensions.

[0031] Taking the voucher with the summary "Purchase of A4 Printing Paper" as an example in this embodiment, the program maps and associates the structured invoice information with the ontology of the financial and tax domain: the invoice is mapped to the ontology concept "Invoice", the product name "A4 Printing Paper" is mapped to the ontology concept "Office Supplies Expenses" (a sub-concept of management expenses), and the tax amount is mapped to the ontology concept "Input Tax". After completing the mapping, the program accurately extracts the core data source required for accounting from the adapted data: 1000 yuan excluding tax, 130 yuan in tax, and 1130 yuan including tax.

[0032] S2. Based on the core accounting data source and preset accounting rules, combined with the accounting rule constraints in the tax and finance domain ontology and the reasoning rules in the tax and finance knowledge graph, it performs automatic accounting for multiple tax types and performs semantic verification synchronously during the accounting process.

[0033] In this step, the program first automatically adapts the tax types to be calculated based on user needs and accounting data, clarifying the accounting priority and requirements for each tax type. Then, it loads personalized accounting rules, tax ontology accounting rules, knowledge graph reasoning rules, and tax regulations accounting requirements, forming a multi-tax accounting rule system. The accounting process adopts a layered accounting strategy: first, it completes the VAT accounting, calling the rule "VAT general taxpayers' tax payable equals current output tax minus current input tax." This transaction only involves input tax of 130 yuan, which is temporarily included in the current input tax pool. Next, it completes the income tax accounting, recording the office supplies expense as current management expenses according to the ontology rules, including it in the deductions from the current taxable income, completing the tax collection of individual vouchers, and at the end of the period, combining the total revenue and cost data to complete the complete income tax payable calculation. Finally, it completes the surcharge accounting, automatically calculating urban construction tax and education surcharge based on the VAT accounting results; this transaction does not require calculation at this time.

[0034] During the accounting process, the program simultaneously performs semantic verification. Specifically, the program checks whether the accounting logic, data relationships, and deduction standards conform to the rules and constraints of the tax and finance ontology and the reasoning results of the tax and finance knowledge graph. For example, if the knowledge graph stores the triple relation "A4 printing paper - applicable tax rate - 13%", the program verifies whether the 13% tax rate on the invoice is consistent with this reasoning result, thereby ensuring the accuracy of the accounting logic.

[0035] After the calculation is completed, the program performs a preliminary verification of the calculation results for each tax type. If the calculation results pass the preliminary verification, the program proceeds to the subsequent data conversion stage.

[0036] If the accounting results fail the initial verification, for example, due to a data mapping error that incorrectly maps input tax to output tax, resulting in abnormal VAT accounting, the program will execute an automatic error identification and correction process. This process can be uniformly scheduled and executed by the financial and tax intelligent agent.

[0037] During the error identification phase, the program uses ontology constraints to determine whether the purchase transaction generates input tax instead of output tax. It also uses knowledge graph reasoning to confirm the invoice type is input tax, thus automatically identifying the error type as "incorrect application of accounting rules." In the solution generation phase, the program calls a large language model enhanced by RAG. It first retrieves similar historical cases from the external error correction example library, such as a successful case of "correcting an incorrect tax mapping to input tax." Then, it passes the retrieved case and the current error information to the large language model, which generates a precise correction instruction to "reclassify the tax amount of 130 yuan to the input tax account." In the automatic correction phase, the program automatically executes this correction instruction, adjusting parameters such as accounting logic, data source, and deduction standards. After correction, it performs preliminary and semantic checks again until the checks pass. Only in complex anomalies that cannot be automatically corrected, such as missing core data sources, does the program push the abnormal accounting results and correction suggestions to a human terminal for manual correction and re-verification.

[0038] S3. Convert the verified accounting results into declaration data, generate declaration forms, and perform multi-dimensional verification on the declaration forms.

[0039] The program first converts the accounting results into a suitable declaration data format based on personalized declaration rules and local tax system requirements. For example, it summarizes all input tax amounts collected in the current period into the data in the "Certified and Deducted in the Current Period" column of the supplementary information to the VAT and surcharges declaration form. Then, the program calls a RAG-enhanced large language model to retrieve the standard template for the tax type's declaration form, fills in the summarized data into the corresponding positions in the template, and generates a complete declaration form. After the form is generated, the program first performs format validation and data consistency checks on the declaration form, verifying the completeness of the form items and the consistency between the declaration data and the accounting data. If any checks fail, the conversion logic is readjusted.

[0040] Subsequently, the program performs a comprehensive multi-dimensional verification of the declaration form. For data consistency verification, it checks whether the input tax amount in the declaration form matches the total amount in the accounting module. For compliance verification, it uses a large language model enhanced by RAG to retrieve relevant tax regulations and verify whether the deduction items entered meet the deduction standards stipulated by tax law. For semantic logic verification, it performs deep logic verification through a knowledge graph, such as verifying whether the business corresponding to the input tax amount for office supplies falls under the circumstances where deduction is not allowed under tax law. The program, combined with the business attributes of the purchasing department in the knowledge graph, confirmed that it is a management department rather than a welfare department, determining that it is for normal business purposes, and the semantic verification passed. In addition, the program also performs a time limit verification, automatically comparing the current declaration date with the statutory declaration deadline for the tax type. If the deadline is approaching, an expedited processing flag is triggered; if it is overdue, an overdue warning is generated and automatic submission is blocked, prompting manual confirmation.

[0041] S4. Automatically submit verified tax return forms to the tax return system electronically, receive and parse the returned return results, trigger corresponding processing flows based on the results, and feed back the entire process data to related modules to drive iterative updates of accounting and return rules.

[0042] The program connects to the tax filing system through the official API interface, automatically submits the filing form after identity authentication, monitors the submission status in real time, and records the filing submission log.

[0043] When the returned declaration result is successful, the program records the success status, archives the declaration form and receipt, links the accounting data and bookkeeping data, and establishes an archive index of the declaration result to support subsequent queries and traceability.

[0044] When the returned application result is an application rejection, the program automatically parses the rejection reason code and matches it with error correction rules to adjust the application form. If a corresponding correction rule can be matched, the application data is automatically adjusted and resubmitted; if the rejection reason is a qualification-related or policy-related issue that cannot be automatically corrected, an anomaly report is generated and pushed to the manual terminal, and the corresponding accounting rule is simultaneously marked as pending optimization.

[0045] After the declaration is completed, the program executes cross-module collaborative application and feedback collection. The structured data and results of this accounting and declaration are pushed to various related modules through a unified data interface: to the knowledge graph module for node attribute updates and inference rule optimization; to the financial and tax intelligent agent module as training samples for its autonomous learning and behavior optimization; and to the voucher generation module for voucher generation rule optimization. Simultaneously, the program receives feedback from each related module regarding the application status of the accounting and declaration data and any anomalies, compiling this into feedback data to provide a basis for subsequent optimization.

[0046] Regarding dynamic rule updates, the trigger sources include three categories: updates to tax and financial regulations, iterations of the tax and financial ontology version, and updates to the tax and financial knowledge graph rule base. When the program detects update trigger conditions, it adopts different processing strategies for structured and unstructured changes. For example, when the VAT collection rate for small-scale taxpayers is reduced from 3% to 1%, this change is a structured change. The program automatically parses the change content, generates a rule update plan and impact assessment report to update the collection rate from 3% to 1%, and automatically applies the update. The entire process is fully recorded and supports manual review and rollback at any time. When complex, unstructured changes to interpretive clauses of regulations are detected, the program extracts a change summary, generates an impact analysis report on the potential impact on accounting rules, and pushes it along with the original change text to a human expert terminal for confirmation before executing the update. The system synchronously updates the filing deadlines for each tax type and links with the time limit verification module.

[0047] In terms of feedback optimization and iteration, the program continuously collects feedback from application results, related modules, users, quality assessments, error correction records, and dynamic updates, compiling this into a feedback dataset. Subsequently, a large language model enhanced with RAG is used to analyze the feedback dataset, identifying deficiencies in accounting rules, application rules, and data adaptation processes. Optimization instructions are generated, and these instructions are automatically updated accordingly. Simultaneously, optimization suggestions are pushed to related modules such as the ontology construction module, model optimization module, data fusion module, knowledge graph module, and voucher generation module, achieving collaborative optimization across the entire system. The optimized system is then applied to subsequent accounting and application processes, enabling dynamic iteration of the technical solution through continuous feedback data collection, forming a complete optimization loop centered on real external feedback.

[0048] In this embodiment, all the above steps are automatically executed by a computer program. From the access of voucher data, ontology mapping, automatic calculation of multiple taxes, generation of declaration forms, multi-dimensional verification to automatic declaration submission and subsequent feedback iteration, no manual intervention is required. The system only triggers a manual intervention node and notifies relevant personnel to handle the situation when preset scenarios occur, such as missing core data sources, declaration rejection and failure of automatic correction, or unstructured regulatory updates involving significant rule adjustments. This embodiment is an exemplary illustration of a single-voucher accounting scenario. In actual enterprise application environments, this method supports continuous automatic processing of batch voucher data across the entire period, as well as simultaneous declaration of multiple taxes.

[0049] The role and effect of the embodiments This invention presents a tax accounting and declaration method based on a tax ontology. By accurately mapping voucher data for "A4 printing paper" procurement transactions to ontology instances, it ensures high consistency between expense classification and accounting regulations and tax attributes, avoiding classification errors common in traditional keyword matching methods. The collaborative work of rigid ontology constraints and flexible reasoning from knowledge graphs successfully verifies the matching relationship between invoice tax rates and product categories, effectively intercepting potential accounting anomalies. When data mapping errors occur, a large language model enhanced by RAG, combined with historical amendment examples, automatically generates and executes a correction scheme, requiring no manual intervention throughout the process. A multi-dimensional verification system covering data consistency, compliance, and semantic logic ensures the accuracy and compliance of declared data from multiple technical levels. A dynamic rule update mechanism and feedback optimization loop enable the system to continuously improve itself in practical use.

[0050] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. 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. A tax accounting and declaration method based on the subject of finance and taxation, characterized in that, Includes the following steps: S1. Obtain intelligent accounting data. Based on the pre-built financial and tax domain ontology, map the intelligent accounting data to ontology instances and extract the core accounting data source. S2. Based on the core accounting data source and the preset accounting rules, combined with the accounting rule constraints in the tax domain ontology and the reasoning rules in the tax knowledge graph, perform automatic accounting for multiple taxes, and simultaneously perform semantic verification during the accounting process; S3. Convert the verified accounting results into declaration data, generate a declaration form, and perform multi-dimensional verification on the declaration form; S4. Automatically submit verified tax return forms to the tax return system electronically, receive and parse the returned return results, trigger the corresponding processing flow based on the return results, and feed back the entire process processing data to the related modules to drive the iterative update of accounting and return rules.

2. The method according to claim 1, characterized in that: The tax domain ontology described in steps S1 and S2 predefines the concepts, attributes, relationships between concepts, and rule constraints based on the concepts and attributes in the tax domain. The tax knowledge graph stores the association relationships of entity instances and business experience rules. The tax domain ontology and the tax knowledge graph work together to provide rigid rule verification and flexible logical reasoning in the accounting and verification process.

3. The method according to claim 1, characterized in that: The semantic verification performed synchronously during the accounting process in step S2 includes checking whether the accounting logic, data associations, and deduction standards used in the accounting process comply with the rule constraints of the tax domain ontology and the reasoning results of the tax knowledge graph.

4. The method according to claim 1, characterized in that: After performing automatic multi-tax calculation as described in step S2, the method also includes automatically identifying the type of calculation error when the calculation result fails the preliminary verification. The large language model enhanced by retrieval enhancement generation technology is invoked, and combined with the tax and finance domain ontology, the tax and finance knowledge graph, and the historical amendment example library, to generate an automatic correction scheme and execute the correction until the verification is passed.

5. The method according to claim 1, characterized in that: Step S3 describes performing multi-dimensional verification on the declaration form, including data consistency verification, which checks the consistency between the declared data and the accounting data. Compliance verification: Verify whether the declared data complies with the financial and tax regulations and the rules and constraints of the financial and tax domain itself; as well as Semantic logic verification, combined with the aforementioned financial and tax knowledge graph, verifies the logical consistency between the declared data and the business semantics and tax attributes.

6. The method according to claim 1, characterized in that: Step S4, which involves triggering the corresponding processing flow based on the application result, includes automatically parsing the rejection reason code, matching the error correction rules to adjust the application form, and resubmitting it if the application result is an application rejection. If the reason for rejection cannot be automatically corrected, an exception report will be generated and pushed to the human terminal.

7. The method according to claim 1, characterized in that: The iterative update of the driving accounting and reporting rules mentioned in step S4 includes monitoring the triggering conditions for updates to financial and tax regulations, iterations of the financial and tax domain ontology version, and updates to the financial and tax knowledge graph rule base. When structured updates are detected, the updated content is automatically parsed, a rule update scheme is generated and automatically applied, and an impact assessment report is generated and manual rollback is supported. When unstructured updates are detected, update suggestions and impact analysis reports are generated and pushed to a human terminal for confirmation before the update is executed.

8. The method according to claim 1, characterized in that: Step S4 describes feeding back the entire process data to the associated modules to drive the iterative updates of accounting and reporting rules. This includes collecting feedback on reporting results, feedback from associated modules, and quality assessment feedback to form a feedback dataset. The feedback dataset is analyzed by calling a large language model enhanced by retrieval enhancement generation technology, which identifies the optimization direction of accounting rules and reporting rules, generates optimization instructions, and performs automatic iteration.

9. The method according to any one of claims 1 to 8, characterized in that: The method is executed automatically by a computer program; a manual intervention node is triggered when the core data source is missing, resulting in the inability to automatically calculate, the application is rejected and automatic correction fails, or unstructured regulations are updated, involving significant rule adjustments.

Images