CN-121981669-A - Collaborative auditing method and system for financial document

CN121981669ACN 121981669 ACN121981669 ACN 121981669ACN-121981669-A

Abstract

The invention provides a collaborative auditing method and a collaborative auditing system of a financial document, wherein the method comprises the steps of obtaining the financial document to be audited, judging whether the document is a mixed-adhesion document, if yes, dividing the document into a plurality of Shan Zhanglei documents through a target detection module, if not, directly taking the document as a bill Zhang Lei document, extracting identification text from the Shan Zhanglei document, completing business classification through a classification model based on the identification text, extracting a preliminary auditing element through a large language model in combination with a business type, completing standardized processing according to an auditing element term dictionary, adopting a double auditing mechanism, calling a rule engine to execute rigid check on a strong compliance element, calling the large language model to execute flexible judgment on a subjectivity element, and obtaining an auditing result. According to the method, the automatic segmentation, element standardization and dual-core collaborative audit of the mixed-pasting bill are adopted, so that the automation level and coverage range of bill audit are improved, the quality of audit data is ensured, and the system deployment and maintenance cost is reduced.

Inventors

DING WENJING
ZHAO YAN
Yuan Fangqi
YANG AIPING
CHEN XIAN
FANG LIANG

Assignees

上海宝信软件股份有限公司

Dates

Publication Date: 20260505
Application Date: 20251222

Claims (10)

1. A collaborative auditing method of a financial document is characterized by comprising the following steps: Step S1, acquiring a financial document to be audited; Step S2, judging whether the financial document is a mixed-pasting type document comprising a plurality of independent documents, if so, carrying out region segmentation processing on the mixed-pasting type document through a preset target detection module to obtain a plurality of Shan Zhanglei documents, and if not, directly taking the financial document as a document Zhang Lei; S3, extracting document identification text from the Shan Zhanglei documents; S4, based on the document identification text, carrying out service classification through a preset classification model to obtain service types; Step S5, extracting a preliminary audit element from the receipt identification text through a large language model based on the service type, and carrying out standardized processing on the preliminary audit element by combining a preset audit element term dictionary to obtain a standardized audit element; and S6, performing auditing on the standardized auditing elements by adopting a double auditing mechanism to obtain auditing results, wherein the double auditing mechanism comprises rigid check and flexibility judgment.
2. The collaborative audit method according to claim 1 wherein in step S2, the target detection module employs a convolutional neural network based detection model.
3. The collaborative auditing method according to claim 1, wherein in the step S3, an optical character recognition module is called for a bill Zhang Lei in an image format, and a document parsing module is called for a bill Zhang Lei in an electronic file format to directly extract text streams.
4. A collaborative auditing method according to claim 3, wherein in step S3, an optical character recognition enhancement processing sub-step is added to a sheet Zhang Lei document in image format: a rotation correction step, namely detecting the character line direction of the original image corresponding to the Shan Zhanglei bill, and correcting the image deflection through an angle regression algorithm; A layout analysis step, namely carrying out layout analysis on the corrected image, and carrying out regional blocking according to the spatial distribution and boundary characteristics of the text region and the table region to divide the text region and the table region; And reconstructing a table structure according to the geometric layout of the horizontal lines, the vertical lines and the cells in the image aiming at the table area, reconstructing a table row and column structure and outputting structured table data.
5. A collaborative audit method according to claim 1 wherein step S4 includes: Based on the bill identification text, combining a preset accessory corpus, carrying out preliminary classification by adopting a mode of combining vector semantic retrieval and keyword matching, outputting a preliminary service type and a corresponding confidence coefficient value, if the confidence coefficient value meets a preset confidence coefficient threshold value, directly determining the preliminary service type as the service type of the bill, and if the confidence coefficient value does not meet the preset confidence coefficient threshold value, calling a large language model to carry out deep semantic classification on the bill identification text, and determining a semantic classification result as the service type of the bill.
6. The collaborative auditing method according to claim 1, wherein in the step S6, the rigidity verification adopts a rule engine, and performs logic verification on a strong compliance element in the standardized audit elements according to a preset financial rule, wherein the strong compliance element is an element verified based on deterministic logic; and the flexibility judgment adopts the large language model to execute semantic understanding and rationality judgment on subjective elements in the standardized auditing elements, wherein the subjective elements are elements which need to be combined with a business background to carry out rationality inference.
7. The collaborative auditing method according to claim 6, wherein the rule engine adopted in the step S6 has a visual configuration function, and the user completes the creation and modification of the financial system rule on a graphical operation interface by dragging a functional component and setting parameters.
8. A collaborative audit method for financial documents according to claim 1 further including: And S7, summarizing the auditing result, generating an auditing opinion report, and pushing the auditing opinion report to an enterprise business system through a standardized application program interface.
9. The collaborative audit method of financial documents according to claim 8, wherein in step S7, the summarized audit result is encapsulated into a structured JSON data packet, the structured JSON data packet includes an overall audit status and a problem list, the problem list includes a problem element, a corresponding rule ID, a hint information, and a severity level, and the structured JSON data packet is pushed to an enterprise business system through a standardized application program interface.
10. A collaborative audit system for financial documents employing a collaborative audit method for financial documents according to any of claims 1-9 comprising: the method comprises the steps of (1) acquiring a financial document to be audited; The module M2 judges whether the financial document is a mixed-pasting document comprising a plurality of independent documents, if so, the mixed-pasting document is subjected to region segmentation processing through a preset target detection module to obtain a plurality of Shan Zhanglei documents, and if not, the financial document is directly used as a document Zhang Lei; A module M3 extracts a bill identification text for the Shan Zhanglei bill; The module M4 is used for carrying out service classification through a preset classification model based on the document identification text to obtain service types; The module M5 extracts a preliminary audit element from the bill identification text through a large language model based on the service type, and performs standardized processing on the preliminary audit element by combining a preset audit element term dictionary to obtain a standardized audit element; And the module M6 is used for performing audit on the standardized audit elements by adopting a double audit mechanism to obtain audit results, wherein the double audit mechanism comprises rigid check and flexibility judgment.

Description

Collaborative auditing method and system for financial document Technical Field The invention relates to the field of computer technology and artificial intelligence application, in particular to a collaborative auditing method and a collaborative auditing system for financial documents. Background In the field of enterprise financial management, financial document auditing is a key link for controlling cost compliance and guaranteeing fund safety. Along with the continuous expansion of enterprise business scale, the source channel, format standard and business type of financial documents are diversified and complex, and higher requirements are put on the efficiency and precision of the audit work. The traditional financial auditing mode completely depends on manual operation, so that auditing efficiency is low, labor cost is high, auditing errors are easily caused by subjective judgment differences and fatigue states of staff, and the requirement of large-scale development of enterprises is difficult to adapt. In order to improve auditing efficiency, the prior art introduces an automatic auditing scheme. The scheme adopts an optical character recognition technology to extract ticket characters and then checks the ticket characters through a preset rule base, but has the obvious defects that firstly, mixed-pasting type documents cannot be automatically identified and segmented, manual pre-splitting is needed, the degree of automation is limited, secondly, the maintenance cost of the rule base is high, frequent manual updating rules are needed for documents with changeable formats or new types, adaptability is poor, thirdly, the deep semantic understanding capability is lacked, only the checking of structural elements can be completed, and the subjective elements such as business recruitment rationality, cost and business activity correlation and the like cannot be effectively checked. Another type of scheme attempts to introduce a large language model to improve the auditing intelligence level, for example, the large language model is utilized to extract the document key fields, and then the auditing decision is completed by combining a knowledge graph system. The scheme still does not solve the problem of automatic processing of the mixed-pasting bill, an auditing mechanism is still single, flexible judgment of complex subjective business scenes cannot be realized only by means of structured query logic of a knowledge graph, and meanwhile, the scheme completely depends on a large-scale multi-mode model, so that the problems of high deployment cost and severe requirement on computing resources exist, and popularization and application of the scheme in small and medium enterprises are limited. The invention patent with the publication number of CN120725813A is found through the search of a patent document, and discloses an intelligent auditing method, device, equipment and storage medium, wherein the method comprises the steps of collecting a financial auditing bill file; the method comprises the steps of converting a financial auditing bill file into a structured text by using an OCR model, classifying the structured text by using a classification model to obtain a classification result, extracting key fields from the structured text by using a large language model and a multi-modal model based on the classification result, acquiring auditing rules and constructing a knowledge graph system based on the auditing rules, and inputting the key fields into the knowledge graph system to carry out auditing decision to obtain an auditing result. According to the invention, the document text is accurately identified through the OCR model, the document is intelligently classified through the classification model, the document content is deeply analyzed through the large language model and the multi-mode model, and meanwhile, the document is automatically checked by combining with the knowledge graph system, so that the automatic check of the document is realized. The patent cannot process mixed-pasting documents, has limited auditing precision due to non-standardized elements, only depends on knowledge graph auditing, lacks dual checking capability of rigidity and flexibility, has irregular flexible configuration and multi-system adaptation design, and has poor scene adaptability and usability. In summary, in order to solve the above-mentioned problems in the prior art, research on a collaborative auditing method and system for financial documents is a critical task to be solved. Disclosure of Invention Aiming at the defects in the prior art, the invention aims to provide a collaborative auditing method and a collaborative auditing system for financial documents. The collaborative auditing method of the financial document provided by the invention comprises the following steps: Step S1, acquiring a financial document to be audited; Step S2, judging whether the financial document is a mixed-pasting type document compr