Financial data intelligent checking and automatic filling integrated method and system

Abstract

The application provides a financial data intelligent verification and automatic filling integrated method and system, comprising: acquiring input data, wherein the input data comprises financial data and corresponding demand information; constructing a knowledge graph, wherein the knowledge graph comprises a rule graph for storing financial rules and constraint associations and a template graph for storing business logic and template associations; intelligently verifying the input data through the rule graph and a large model to obtain an intelligent verification result, filtering and extracting the intelligent verification result according to the template graph to obtain template data, and filling in the template data and the intelligent verification data through the large model to obtain a final financial data filling result. The intelligent integrated processing method from data verification to filling is realized.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to an integrated method and system for intelligent verification and automated data entry of financial data. Background Technology

[0002] Current corporate financial data processing faces core pain points such as diverse original voucher formats, complex and ever-changing compliance rules, and low efficiency and error-proneness of manual operations. Existing solutions either focus on automation based on fixed rules or attempt to apply large models for flexible interpretation. However, large models have poor controllability and unstable output, making it difficult to strike a balance between flexibility, accuracy, and compliance rigidity. In particular, there is a lack of an intelligent, integrated processing method that can deeply integrate corporate proprietary knowledge, real-time external regulations, and span the entire process from data verification to data entry. Summary of the Invention

[0003] In view of this, the present invention proposes an integrated method and system for intelligent verification and automated data entry of financial data to solve the problems existing in the prior art.

[0004] To achieve the above objectives, this invention proposes an integrated method and system for intelligent verification and automated data entry of financial data, comprising: Obtain input data, which includes financial data and corresponding requirements information; Construct a knowledge graph, wherein the knowledge graph includes a rule graph for storing financial rules and constraint associations and a template graph for storing business logic and template associations; The input data is intelligently validated using rule graphs and large models to obtain intelligent validation results. Template data is obtained by filtering and extracting the intelligent validation results using template graphs. Finally, the financial data is filled in using large models based on the template data and intelligent validation data to obtain the final financial data filling results.

[0005] Optionally, input data can be obtained through a user interface via data upload. The user interface includes an upload area for uploading financial data and a dialog box for inputting and prompting information. The user interface also includes an area for uploading basic data for building the knowledge graph and a component for viewing the knowledge graph. Additionally, the user interface includes a component for previewing and downloading the final financial data entry results.

[0006] Optionally, the financial data may be organized documents or documents after different financial data have been identified.

[0007] Optionally, the rule graph includes different entities with different associations, wherein the entities include invoice entities, account entities, amount entities, date entities, and calculation entities; there are calculation rules between the calculation entities, wherein the calculation rules have constraint rules, wherein the constraint rules include time validity constraints, data status constraints, and business process triggering constraints.

[0008] Optionally, the template graph includes process entities with different transition conditions, wherein each process entity is associated with a different template entity, and the template entity is associated with template data.

[0009] Optionally, the intelligent verification process includes: The input data is retrieved from the rule graph to obtain rule retrieval results, which include the corresponding rules, formulas, and parameters. The input data and rule retrieval results are combined to form a complete rule context, which is then processed by a large model to obtain intelligent verification results. The large model adopts a lightweight RAG model.

[0010] Optional, the form completion process includes: Based on the cleanliness data in the intelligent verification results, a search is performed in the template graph to generate associated template nodes. The corresponding template data is extracted based on the associated template nodes. The template data, cleanliness data, and requirement information are combined. The combined data is processed through a large model to generate the final financial data reporting results. The large model adopts a lightweight RAG model. The template data includes the template structure, the template field mapping rules, the template detailed data, and the context information of the template entity.

[0011] On the other hand, the present invention provides an integrated system for intelligent verification and automated data entry of financial data, for performing the above-mentioned methods.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves a leap in financial data processing from manual rule configuration to intelligent semantic understanding and execution by constructing a dynamically evolving financial knowledge graph and deploying a task-oriented, lightweight large-scale model. The system can automatically adapt to diverse financial documents and customized enterprise processes, improving the overall efficiency of data preparation, verification, and entry while significantly reducing human error, all while ensuring compliance. Attached Figure Description

[0013] 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. In the drawings: Figure 1 This is a schematic diagram of the user interface in an embodiment of the present invention; Figure 2 This is a schematic diagram of the data processing flow in an embodiment of the present invention. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] This embodiment proposes an integrated method for intelligent verification and automated data entry of financial data, such as... Figure 1-2 As shown, it includes: Users can upload financial documents such as invoices. On the user interface, they can choose to drag and drop files or click to upload them. The user interface also has a dialog box for inputting requirements, which indicates "Enter requirements". The user interface also has progress and status prompts. After data processing in the backend, the user interface generates corresponding files based on the aforementioned financial documents and requirements. Users can preview the generated files using the file preview option and download the files using the generated financial document download button.

[0016] The user interface also includes a custom knowledge base file upload function. During the upload process, a progress bar indicates the upload status, and options for knowledge base file status and position are provided. Clicking on the knowledge base file status and position allows users to view the file names in the knowledge base and the learning status of the large model.

[0017] The knowledge base file upload is used for the construction of the knowledge graph and the identification of template documents of different template types.

[0018] To facilitate the organization of financial data, it is also possible to connect to the company's OA financial system as needed and perform automated data entry through the company OA's automatic execution options.

[0019] The specific details are as follows: 1. Preload some financial header or data format files within the node machine; 2. Users can customize and upload financial files to meet specific task requirements based on actual scenario needs; 3. Users can confirm the upload status of customized files using the progress bar; clicking on "Knowledge Base File Status" allows viewing the file names in the knowledge base and the learning status of the large model; 4. Upload invoices or other financial files, entering the desired financial file type in the dialog box below; 5. Click "Confirm" to display a preview of the generated file; 6. Download the generated file; 7. Automated data entry can also be performed by connecting to the company's OA financial system as needed.

[0020] The data processing backend of this invention acquires relevant financial data documents, combines them with user-input requirements, and uses a knowledge graph and a large-scale model API to intelligently validate the documents. The validated data is then cached in the business service. Based on the validated data, corresponding template data is selected from the document templates and cached. The cached template data, the validated data, and the user-input requirements are used as re-input information. This re-input information is then analyzed and automatically populated using a knowledge graph and a large-scale model API to generate the final financial data filled with a standard template.

[0021] The system uses programmatically identified documents as template content, calls LightRAG's API to answer questions, and then precisely fills the template fields using a customized program. The document format is accurate, controllable, and reliable, while the question content is also relatively controllable and reliable.

[0022] The document materials can be selected from organized data-related documents, or from corresponding electronic invoices, invoice images, or other document data in the financial system. The text recognition will extract and save the financial data as a template for filling in the content.

[0023] The above technical solution is described in detail below: Regarding the aforementioned user interface, which serves as the front-end entry point for the entire system, its core design aims to guide users efficiently through the entire process from data upload to result retrieval, while simultaneously making the complex internal processing states of the system transparent. The main interface is divided into four logically coherent areas.

[0024] The first area is the document and knowledge base management area. Users can upload raw financial documents to be processed, such as invoice images, PDF bank statements, or Excel spreadsheets, by dragging and dropping or selecting. Users can choose to upload or manage "custom knowledge bases" synchronously or asynchronously, such as the company's financial policy manual, historical expense reimbursement templates, or specific tax rule documents. When a file upload begins, the system provides real-time feedback through a dynamic progress bar and clear status labels (such as "Uploading," "Parsing," and "Loaded into Knowledge Base"). For documents already uploaded to the knowledge base, users can click "Knowledge Base Status" to view their name and the learning status of their content by the system's backend big data model.

[0025] After the file upload is complete, the interface focus naturally transitions to the second area—the requirements input area. The smart input box is always available, prompting the user to enter detailed requirements including business processes and types. This design allows users to flexibly combine preset tasks with free instructions to define the final data processing goal.

[0026] Once the background processing is complete, the third area—the Result Preview and Interactive Correction Area—will be activated. In this area, the system displays a preview of the generated standard financial documents, such as neatly formatted expense reports or statements. Crucially, all data fields automatically populated by the system or corrected based on rule validation are highlighted with a gentle color, facilitating quick user review. Users can edit any field directly in the preview interface if they have any questions.

[0027] Ultimately, all processes converge in the fourth area—the Execution and Feedback Area. Here, users can decide on the final action, such as confirming and downloading the generated file, or initiating automated data entry with a single click to push the results to the company's OA or financial system. The execution log panel at the bottom of the interface provides transparent process tracking, scrolling through key step information such as "Knowledge graph matching completed, template confirmed as 'Travel Expense Reimbursement Form'", "Large model API verification passed, no logical conflicts", and "Data successfully cached and ready to be pushed," building user trust in the system.

[0028] In the data processing backend, knowledge graphs and large-scale models are mainly used to analyze the input documents and requirements information, and these are used as the basis for subsequent validation, filling, and generation. The relevant content of the knowledge graph is as follows: The graph adopts a two-level structure design, corresponding to data understanding and business logic respectively.

[0029] To enhance data understanding, a financial entity and rule graph was constructed. During system initialization or when users upload knowledge base files, the backend invokes a large model to perform deep semantic analysis on financial documents, extracting key entities such as "Taxpayer Identification Number," "Total Price and Tax," and "Budget Items," and establishing attribute relationships and constraint rules between them. For example, the graph records that the "Value-Added Tax Invoice" entity is associated with the "Deductible Tax Amount" attribute, and the calculation of this attribute must follow the rule "Tax Amount = Amount × Tax Rate." This graph enables the system to understand the basic structure and inherent relationships of financial data.

[0030] During the construction of the financial entity and rule graph, a pre-built basic financial knowledge framework is loaded during initialization. This framework includes core entities and basic relationships defined in national accounting standards and general tax rules. When users upload enterprise-specific knowledge base files, such as financial system manuals and historical compliance reports, the knowledge graph is constructed based on the above data.

[0031] First, the backend calls a large model to perform deep semantic analysis on the uploaded documents. The large model, fine-tuned through instructions, then performs structured information extraction. Based on predefined entity type schemas, such as invoice entities, account entities, amount entities, date entities, and constraint entities, the model identifies and extracts entity instances from unstructured document text. For example, from a financial policy document, the model can identify a VAT invoice as an invoice entity and extract its key attributes, such as the requirement for separate storage of the deduction copy and a 360-day authentication period.

[0032] Next, the large model is further driven to mine relationships and rules. The large model analyzes statements in the document describing interactions, constraints, or computational logic between entities, formalizing them into structured rules. For example, when parsing the policy requiring round-trip transportation tickets for travel expense reimbursement, the model constructs a rule: associating the travel expense reimbursement form (template entity) with the transportation ticket (ticket entity) through a required relationship, adding a mandatory "yes" attribute. Simultaneously, basic rules regarding positive / negative amounts, contract consistency verification, and single-transaction amount limits are also structurally constructed. For computational rules, such as parsing the current period's tax payable = output tax - input tax from tax law clauses, the large model constructs this as a computational node centered on the tax payable, clearly defining its input variables (output tax, input tax) and mathematical relationships. The large model employs existing models in the knowledge graph field.

[0033] In addition to the structured rules and computational rules mentioned above, constraint rules are also included: A clear time validity boundary is established for each calculation rule. In the knowledge graph, calculation rule nodes are associated with effective and expiration date attributes and dynamically linked to time entity nodes such as tax law versions and accounting policy changes. During calculation execution, it first verifies whether the current business date is within the rule's validity period, and simultaneously checks whether the time attributes of all input data meet the calculation caliber requirements. For example, VAT deduction calculation requires that output and input invoices belong to the same tax period; the consistency of time attributes is verified to ensure the compliance basis of the calculation. The time constraint layer prevents the application of expired rules and erroneous calculations of cross-period data.

[0034] Input variables for computational rules must meet specific data state conditions to participate in the calculation. A state machine model is established for each computational variable, and its legal state transition paths are explicitly defined in the graph. Taking input tax deduction as an example, each input invoice participating in the calculation must simultaneously meet multiple state conditions such as "certified," "certification not expired," and "purpose complies with regulations." These state conditions themselves exist as verification rule nodes in the graph, forming a prerequisite dependency relationship with the computation nodes. The system automatically triggers these state verification rules before executing the calculation; only clean data that passes all verifications can flow into the computation process. At the same time, legal value range constraints are also set for numerical variables to prevent abnormal data from interfering with the calculation results.

[0035] All calculation rules must be triggered in the correct business process. The system achieves scenario-based scheduling of computing power by associating calculation nodes with state nodes in the business process graph. For example, tax payable calculation is only activated when bound to the "tax declaration" process node, while in the "daily accounting processing" scenario, only tax amount recording is executed without initiating summary calculation. This design ensures the business rationality of calculation actions. For complex multi-step calculations, the system establishes logical dependencies between rules in the graph, forming a calculation workflow. Downstream calculation rules are only triggered and executed when the outputs of all precondition rules (such as enterprise type determination and tax incentive eligibility verification) meet the requirements.

[0036] All extracted entities and rules are fed into the graph engine, where they are integrated with existing knowledge in the underlying framework through entity alignment and relationship fusion algorithms. For example, a taxpayer identification number extracted from document A and a tax number extracted from document B will be identified as the same entity and merged. Ultimately, a machine-readable and reasonable semantic network is formed, where nodes represent entities and concepts, and edges represent attributes, categories, or rule relationships. This graph provides a unified and accurate foundation for financial semantic understanding for all subsequent processes.

[0037] To support business logic, a business process and template graph is constructed, overlaying the enterprise's specific workflow on top of the entity graph. This graph models process nodes defined in the enterprise's financial management system, such as employee submission → department manager approval → finance department review → cashier payment, as a series of state nodes and transition conditions. Simultaneously, various financial document templates, such as corporate payment application forms and fixed asset purchase orders, are also modeled as nodes in the graph and strongly associated with process nodes and first-level financial entities. For example, the template node for a travel expense reimbursement form links to entities such as transportation expenses and accommodation expenses, indicating that it must be completed before the department manager approval node. When a user uploads a new template file, the system automatically parses its structure and integrates it as a new subgraph into this layer of the graph, enabling dynamic expansion of the template library.

[0038] The construction of business processes and template diagrams involves modeling the unique operational processes and document specifications of an enterprise. The starting point for this construction is the analysis of the enterprise's written financial management systems, approval authority documents, and archived standard template libraries.

[0039] The system utilizes a large model to perform process mining on policy documents. The model identifies process descriptions, role definitions, and conditional statements in the documents, transforming them into a flowchart graph containing state nodes, action edges, and transition conditions. For example, when it detects that an expense reimbursement form exceeding 5000 yuan requires approval from the supervising leader, the system creates two nodes in the graph: "Department Manager Approval" and "Supervising Leader Approval," adding the conditional attribute "Amount > 5000" to the edge connecting them. Simultaneously, each node in the process is associated with role entities (such as department manager and financial auditing personnel) in the financial entity graph.

[0040] For template modeling, the system employs a dual approach of structural parsing and semantic association. When a new standard template file (such as an Excel-formatted fixed asset purchase requisition) is introduced, the system first parses its physical structure: identifying the header, field names, data types, and required field identifiers. Subsequently, it calls the larger model and, based on the financial entity graph, semantically annotates each field. For example, the asset name field in the template is associated with the fixed asset concept in the entity graph, the supplier field is associated with the supplier entity, and the application amount field is connected to the budget item entity through affiliation relationships. The template itself is added to the graph as a node and associated with the aforementioned semantically annotated field child nodes.

[0041] The relevant content of the large model is as follows: The system's intelligent processing capability is realized through the large model API, and the models at each stage work together.

[0042] In the document understanding and initial classification stage, a general-purpose large-scale model API with powerful multimodal recognition capabilities is first invoked. At this point, uploaded invoice images or messy PDF files are fed into the model, accompanied by precise prompts such as "Please identify this document type and extract all fields containing numbers, dates, and company names." The model's task is to perform preliminary perception and coarse processing, outputting structured data containing document type identification, key field text, and their positions within the original text. This model is used as a programmatic recognition model.

[0043] After completing the initial document understanding and data extraction, the process moves into the deep verification and logical reasoning stage. The core technical solution for this stage employs a lightweight retrieval-enhanced generation (Lightweight RAG) model architecture based on knowledge graphs to achieve accurate, efficient, and traceable financial data rule verification. This solution constructs the verification process as a rule-driven, closed-loop, auditable, deterministic computation and matching workflow.

[0044] First, the raw data fields extracted from the front end are structurally reorganized to form a set of data propositions to be verified. For example, for the processing of VAT invoices, the system generates core propositions such as "the invoice amount is X yuan," "the applicable tax rate is Y%," and "the tax amount recorded on the invoice is Z yuan." These propositions then trigger concurrent targeted searches of the first-layer financial entities and rule graph. The search mechanism adopts a hybrid search strategy, combining semantic similarity matching based on embedded vectors with structured traversal based on graph relationships to accurately recall all verification rules, mathematical calculation formulas, compliance parameters, and business constraints related to each data proposition. For example, for the "invoice amount" entity, it may retrieve "amount positive / negative rules," "contract consistency verification rules," and "single transaction amount upper limit threshold" in parallel; for "tax rate" and "tax amount," it will necessarily retrieve "VAT payable calculation formula" and "the list of currently applicable tax rates in national tax regulations."

[0045] The retrieved rules, formulas, and parameters, together with the original data propositions, constitute the complete rule context for this verification task. This context is input into a specially optimized lightweight RAG model. The model's instructions are strictly limited to rule execution and logical verification, such as: "Please strictly verify the following data propositions item by item according to the provided rules and parameters. For calculation-based propositions, the calculation process must be shown and compared with the claimed value." Within this framework, the model acts as an "automated compliance engine," with a highly deterministic workflow: First, it parses the calculation formulas and parameters in the rule context; then, it substitutes the specific values ​​from the propositions to perform calculations; next, it precisely compares the calculation results with the proposition's claimed value, or checks whether the values ​​are within the legal range defined by the compliance parameters; finally, it outputs a standardized, structured intelligent verification conclusion report, while also including the data without anomalies in the report as a clean data set.

[0046] Following the generation of intelligent verification conclusions in the first stage, the final generation stage of the financial data processing workflow transforms the rigorously verified structured data into standard financial documents that conform to corporate standards. This stage employs a lightweight retrieval-enhanced generation model technology solution driven by knowledge graphs. Through rule-constrained intelligent assembly and semantic generation, it ensures that the output documents meet enterprise-level application standards in terms of format accuracy, content completeness, and business compliance.

[0047] In the second phase of automatic completion, template selection is first performed using a business function model. A multi-dimensional, structured completion generation context is constructed, including the clean data set confirmed in the deep validation phase, the personalized requirement description text submitted by the user on the front end, and the relevant business process identification information associated with the current task within the requirement, serving as the raw data input for the second phase. These elements collectively trigger a deep retrieval and matching process between the second-layer business process and template graph. The retrieval mechanism employs a hybrid strategy: on the one hand, semantic vectorization technology is used to calculate the similarity between the user requirement description and various business scenario nodes in the graph; on the other hand, based on the key entity types in the clean data, a structured traversal is performed along predefined relational paths in the graph. For example, when the system identifies data entities containing "airfare receipts" and "hotel invoices" and the user's request mentions "international travel reimbursement," it will accurately locate the "international travel expense reimbursement form" template node. Through the content in the knowledge graph, the system will extract the complete content of the template through business functions. That is, after extracting the corresponding template entity and other entity content from the knowledge graph through business functions, the system will further obtain the complete structural data, fields, field mapping rules, enterprise-specific filling specifications, associated multi-level approval process nodes, and special policy requirements for international travel associated with the template in the detailed database based on the template entity and other entity content.

[0048] After the business function module retrieves the relevant template content, the retrieved template structure, template field mapping rules, detailed template data, and contextual information associated with template entities, together with the original input data, form a structured instruction set, which is then fed into the lightweight RAG model for processing. The large model acts as a rule executor in this stage, and its operation is strictly constrained. It must completely follow the provided template structure for data mapping and cannot arbitrarily create or change field definitions. The population process is manifested in two closely connected sub-stages: First, the precise value mapping stage, where the model accurately places atomic data units from the clean data into the predefined positions of the template according to the field mapping rules. For example, numerical and coded data such as invoice numbers, transaction dates, and currency amounts are filled into the corresponding fields strictly according to the format standards and precision requirements specified in the template. Second, the derived value calculation and population stage, where for fields that need to be calculated based on business rules, the model performs calculations based on the calculation formulas retrieved from the knowledge graph and business logic. For example, it automatically summarizes the "total expenses" based on various expense details, or calculates the daily subsidy amount based on travel allowance standards, ensuring that all calculation processes comply with the company's financial system requirements.

[0049] For fields in the template that require natural language description, the system initiates a context-aware, restricted text generation process. This generation process is not an open-ended, free creation, but rather an information synthesis and expression optimization within strictly defined boundaries. The generation mechanism is constrained by three constraints: data constraints require that the generated text faithfully reflect key factual information in clean data, such as time series, location distribution, personnel composition, project relationships, and other core elements; business context constraints require that the text expression conform to the specific business scenario specifications and corporate writing conventions retrieved from the graph, for example, the description of the reason for project reimbursement needs to reflect the R&D purpose and output relationship, and the format often requires stating the project background first, followed by explaining the cost composition; and requirement instruction constraints require that the generated content accurately respond to the key concerns emphasized by the user in the requirement description, for example, if the user specifically specifies that the correspondence between costs and milestones should be highlighted, then the generated text needs to clearly reflect the temporal relationship between project stage information and cost occurrence.

[0050] After data entry and text generation are completed, the system performs multi-level output validation and quality assurance. Format integrity validation ensures all required fields are accurately filled and that all data formats fully comply with the template's technical specifications. Data consistency validation checks the internal data logic of the generated document, verifies the matching relationship between item summaries and total amounts, and checks the validity of various coding systems. Compliance review performs keyword scanning and semantic analysis on the generated text descriptions to ensure that they do not contain any prohibited reimbursement items or sensitive information. Finally, the system produces a fully formatted standard financial document that can be directly submitted for approval, along with a detailed process traceability report that fully records the template version used, data mapping relationships, and the basis for the generation rules, providing a complete technical archive for subsequent auditing and quality traceability.

[0051] Using the above technical solution, the present invention has the following technical effects: Improved accuracy and reliability: By solidifying financial entities, rules, and business processes through knowledge graphs, precise domain knowledge boundaries are provided for large models, ensuring that all understanding, verification, and generation actions are carried out within a controllable range. This fundamentally avoids the uncertainty and "illusion" problems of general large models, and the output results have extremely high business accuracy and compliance reliability.

[0052] A qualitative leap in processing efficiency: The integrated process achieves seamless automation from uploading original files to generating final forms or filling in system data, completely freeing finance personnel from tedious and repetitive data handling, verification, and entry work, reducing processing time from hours to minutes or even seconds.

[0053] Strong adaptability and scalability: The knowledge graph upon which the system's core depends supports dynamic updates. New financial regulations, templates, or tax policies of an enterprise can be semi-automatically integrated into the graph by uploading documents, enabling the system to gain new capabilities without modifying the core code, greatly reducing system maintenance costs and iteration cycles.

[0054] Excellent explainability and auditability: All key system decisions (such as why a certain template is selected, why a certain invoice is deemed abnormal, and how a certain amount is calculated) are generated based on traceable nodes and rules in the knowledge graph, producing detailed process logs. This provides a complete digital track for financial auditing, internal control, and problem investigation, meeting the stringent requirements of enterprise internal control and compliance.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method integrating intelligent verification and automated data entry for financial data, characterized in that: include: Obtain input data, which includes financial data and corresponding requirements information; Construct a knowledge graph, wherein the knowledge graph includes a rule graph for storing financial rules and constraint associations and a template graph for storing business logic and template associations; The input data is intelligently validated using rule graphs and large models to obtain intelligent validation results. Template data is obtained by filtering and extracting the intelligent validation results using template graphs. Finally, the financial data is filled in using large models based on the template data and intelligent validation data to obtain the final financial data filling results.

2. The method according to claim 1, characterized in that, Input data is obtained through data upload via a user interface. The user interface includes an upload area for uploading financial data and a dialog box for inputting and prompting information. The user interface also includes an area for uploading basic data for building the knowledge graph and a component for viewing the knowledge graph. Additionally, the user interface includes a component for previewing and downloading the final financial data entry results.

3. The method according to claim 1, characterized in that, The financial data refers to organized documents or documents that have been identified from different financial data.

4. The method according to claim 1, characterized in that, The rule graph includes different entities with different relationships, including invoice entities, account entities, amount entities, date entities, and calculation entities; there are calculation rules between the calculation entities, and there are constraint rules between the calculation rules, including time validity constraints, data status constraints, and business process triggering constraints.

5. The method according to claim 1, characterized in that, The template graph includes process entities with different transition conditions, wherein each process entity is associated with a different template entity, and the template entity is associated with template data.

6. The method according to claim 1, characterized in that, The intelligent verification process includes: The input data is retrieved from the rule graph to obtain rule retrieval results, which include the corresponding rules, formulas, and parameters. The input data and rule retrieval results are combined to form a complete rule context, which is then processed by a large model to obtain intelligent verification results. The large model adopts a lightweight RAG model.

7. The method according to claim 1, characterized in that, The process of filling out the form includes: Based on the cleanliness data in the intelligent verification results, a search is performed in the template graph to generate associated template nodes. The corresponding template data is extracted based on the associated template nodes. The template data, cleanliness data, and requirement information are combined. The combined data is processed through a large model to generate the final financial data reporting results. The large model adopts a lightweight RAG model. The template data includes the template structure, the template field mapping rules, the template detailed data, and the context information of the template entity.

8. An integrated system for intelligent verification and automated data entry of financial data, characterized in that: Used to perform the method described in any one of claims 1-7.

Images