A multi-agent-based natural language, invoice, bank flow, payroll automatic accounting method and system

Abstract

This invention discloses a method and system for automatic accounting based on natural language, invoices, bank statements, and payroll, belonging to the field of financial automation technology. This invention utilizes four functional modules—intent parsing, voucher generation, risk assessment, and report generation—to collaboratively generate accounting vouchers from at least one data source, including natural language business descriptions, invoices, bank statements, and payroll. After compliance and risk assessment, it automatically generates financial reports, achieving a fully automated accounting loop from user instructions to financial statements. This invention decouples intelligent decision-making from data execution in a layered manner, improving accounting efficiency and accuracy, reducing errors and tax risks, and can be widely applied in enterprise financial software and financial automation scenarios.

Description

Technical Field

[0001] This invention relates to the fields of financial automation, artificial intelligence applications, and computer software technology, specifically to a method and system for automatic accounting of natural language, invoices, bank statements, and payroll based on multi-agent systems. Background Technology

[0002] Most existing accounting software relies on manual input of original documents such as invoices, bank statements, and payroll, as well as manual determination of accounting subjects and preparation of accounting vouchers. This results in problems such as low efficiency, high error rate, high cost, and poor standardization.

[0003] While some automated accounting tools exist, they generally suffer from the following drawbacks: 1. It only supports processing a single data source and cannot simultaneously perform integrated automatic accounting for multiple data sources such as natural language business descriptions, invoices, bank statements, and payroll. 2. It cannot automatically generate accounting vouchers based directly on conversational business descriptions (such as "purchased office desks for 150 yuan"); 3. Bookkeeping, auditing, and reporting are independent of each other, and a fully automated closed loop has not been formed; 4. The high degree of coupling between AI decision-making and data execution results in poor system stability, security, and maintainability; 5. Lack of professional and modular intelligent division of labor, resulting in insufficient accuracy in bookkeeping and risk control capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for automatic accounting of natural language, invoices, bank statements, and payroll based on multi-agent systems, so as to achieve full-process, highly reliable, and highly secure automatic accounting.

[0005] The technical solution of this invention is as follows: An automatic accounting method based on multi-agent intelligence, comprising: 1) receiving user natural language instructions or natural language business descriptions, and performing intent parsing and task distribution; 2) automatically generating accounting vouchers based on at least one data source selected from invoices, bank statements, payrolls, and natural language business descriptions; 3) conducting compliance and risk audits on the accounting vouchers; and 4) automatically generating financial reports based on the accounting and audit results. In this method, intent parsing, voucher generation, risk auditing, and report generation are performed by different functional modules, with intelligent decision-making and data execution decoupled in a layered manner, forming a fully automatic accounting closed loop.

[0006] Furthermore, the intent parsing module identifies the user's natural language commands, extracts parameters such as account set, company, year, month, account, amount, summary, and business item, and distributes the task to the corresponding module.

[0007] Furthermore, the voucher generation module identifies and matches accounts based on at least one of the following: natural language business descriptions, invoices, bank statements, and payroll, automatically generating standardized accounting vouchers. The natural language business descriptions include colloquial business transactions, and the system automatically identifies the account, debit / credit direction, amount, and summary.

[0008] Furthermore, the risk audit module checks the account of the voucher, the balance of debits and credits, the compliance of invoices, the rationality of cash flow, the payroll logic, tax risks, and the rationality of natural language business, and outputs errors and suggestions.

[0009] Furthermore, the intelligent decision-making module only outputs structured results and does not directly manipulate the database. Data storage and execution are handled by an independent execution layer, ensuring system security and stability.

[0010] Furthermore, if the review fails, the system will automatically correct the error or prompt manual intervention until the review is passed, at which point the report generation process will begin.

[0011] This invention also provides an automated accounting system based on multiple agents, including: an intent parsing module, a voucher generation module, a risk auditing module, and a report generation module; each module has its own division of labor and cooperation, and intelligent decision-making and data execution are decoupled in a layered manner to realize the above-mentioned automated accounting method. Detailed Implementation

[0012] The user inputs a natural language business description: "I spent 150 yuan to buy an office desk." The intent parsing module identifies this as an office expense, extracts information such as the account, amount, and summary, and distributes it to the voucher generation module. The voucher generation module automatically determines the accounting account, debit / credit direction, and amount, generating an accounting voucher. The risk review module reviews the reasonableness of the account, the amount, and the debit / credit balance. After the review is approved, the report generation module generates financial statements and risk analysis based on the complete ledger.

[0013] For invoices, bank statements, and payroll data sources, the system automatically identifies, extracts, matches accounts, generates vouchers, reviews risks, and outputs reports, achieving integrated automatic accounting across multiple data sources. In this embodiment, the functional modules work independently yet collaboratively, with intelligent decision-making and data execution decoupled in a layered manner to ensure system security, stability, and scalability.

[0014] Beneficial effects 1. It supports automatic accounting for four types of data sources: natural language business descriptions, invoices, bank statements, and payroll, covering all core daily accounting scenarios for enterprises; 2. It can automatically generate vouchers based on conversational business descriptions (such as "purchase of office desk for 150 yuan"), making it extremely easy to use; 3. Multi-module specialized division of labor ensures high accounting accuracy and strong risk control capabilities; 4. The intelligent layer and the execution layer are decoupled, making the system secure, stable, easy to maintain, and easy to upgrade; 5. It achieves a fully automated closed loop from user instructions to financial statements, significantly improving bookkeeping efficiency and reducing labor costs and the risk of errors.

Claims

1. An automatic accounting method based on multi-agent systems, characterized in that, include: 1) Receive user natural language commands or natural language business descriptions, and perform intent parsing and task distribution; 2) Automatically generate accounting vouchers based on at least one data source, including invoices, bank statements, payrolls, and natural language business descriptions; 3) Perform compliance and risk audits on the generated accounting vouchers; 4) Automatically generate financial reports based on the accounting and audit results; Among these, intent parsing, voucher generation, risk auditing, and report generation are performed by different functional modules, and intelligent decision-making and data execution are decoupled in a layered manner to form a fully automated accounting closed loop.

2. The method according to claim 1, characterized in that, The intent parsing and task distribution include: recognizing user natural language commands, extracting parameters such as account set, company, year, month, subject, amount, summary, and business item, and assigning tasks to corresponding functional modules.

3. The method according to claim 1, characterized in that, The natural language business description voucher generation process includes: recognizing business matters in colloquial and natural language form input by users, automatically determining the accounting subject, debit / credit direction, amount, and summary, and generating accounting vouchers.

4. The method according to claim 1, characterized in that, The automatic invoice voucher generation function includes: identifying invoice type, input / output tax, tax rate, and amount; automatically matching accounting subjects; and generating vouchers for revenue, expenses, input tax, and output tax.

5. The method according to claim 1, characterized in that, The automatic bank statement voucher generation function includes: recognizing receipts, payments, transfers, handling fees, and interest, and automatically generating vouchers for bank deposits, accounts receivable, accounts payable, and expenses.

6. The method according to claim 1, characterized in that, The automatic payroll voucher generation function includes: identifying gross salary, social security contributions, housing provident fund contributions, individual income tax, and net salary, and automatically generating vouchers for accrued wages, wage payments, and social security and housing provident fund contributions.

7. The method according to claim 1, characterized in that, The risk audit includes at least one of the following: account correctness audit, debit and credit balance audit, invoice duplication verification, transaction anomaly verification, payroll logic verification, tax risk verification, and natural language business rationality verification.

8. The method according to claim 1, characterized in that, The intelligent decision-making layer only outputs structured results and does not directly manipulate the database; the storage, retrieval, and execution of data are handled by an independent execution layer.

9. The method according to claim 8, characterized in that, After the voucher is generated, the execution layer automatically sends it to the risk review module. If the review fails, it will automatically correct the voucher or prompt manual processing until the review is passed.

10. The method according to claim 1, characterized in that, The financial reports include at least one of the following: balance sheet, income statement, cash flow statement, tax risk report, and business analysis report.

11. An automated accounting system based on multi-agent systems, characterized in that, include: Intent parsing module: used for understanding, parsing, and task distribution of natural language instructions and business descriptions; Voucher generation module: Used to generate accounting vouchers based on at least one of the following: invoices, bank statements, payroll, and natural language business descriptions; Risk audit module: used for voucher compliance and risk checks; Report generation module: Used to generate financial reports; Each module has its own division of labor and collaboration, and intelligent decision-making and data execution are decoupled in a layered manner to execute the automatic accounting method described in any one of claims 1 to 10.