Intelligent voucher auditing method and system based on large language model

Through the intelligent credential auditing method based on the large language model, the problem of poor flexibility and slow processing speed of rule engines in the existing technology is solved, automatic classification and auditing is realized, the system's self-learning and optimization capabilities are enhanced, and efficiency and security are improved. sex.

CN119027251AInactive Publication Date: 2024-11-26FUJIAN BOSS SOFTWARE

Patent Information

Application Number
CN202411137849.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art has problems such as poor flexibility of rule engines, slow processing speed and lack of semantic understanding in voucher audits, which makes it difficult to adapt to rapidly changing business needs and diversified voucher formats, affecting efficiency and decision-making accuracy.

Method used

Using intelligent credential auditing method based on large language model, structured text information is extracted through optical character recognition, semantic analysis and in-depth understanding, and classified and audited credentials with preset rules, and self-learning and optimization are used for knowledge base to adapt to Business changes.

Benefits of technology

It realizes automated classification and auditing, improves work efficiency, enhances the system's self-learning and optimization capabilities, can detect abnormalities and compliance, and improves the system's security and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027251A_ABST
    Figure CN119027251A_ABST
Patent Text Reader

Abstract

The invention relates to an intelligent voucher auditing method and system based on a large language model. The method comprises the following steps: collecting a voucher digital copy uploaded by a user, and adding a label to the uploaded digital copy; using an OCR technology to extract the uploaded digital copy structured text information; performing semantic analysis on the text information extracted by the OCR by using a large language model; classifying the voucher digital copy into a preset category based on a semantic analysis result; using a preset rule and deep semantic understanding of a large language model, and referring to the collected original bill structured data to audit the classified voucher digital copies; a final result of the certificate digital copy audit is obtained and output to the user, and the user manages and adjusts the audit process according to the output result; and according to the feedback of the user to the output result, storing the classification and auditing result into a knowledge base to continuously carry out large language model training and optimization, and automatically adjusting the classification and auditing rules.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an intelligent credential auditing method and system based on a large language model. Background Art

[0002] At present, enterprises generally use the following two mainstream technical means in the process of auditing various voucher documents:

[0003] Manual review: Many companies rely on professional financial personnel to check the vouchers one by one to ensure the accuracy and compliance of the data. In this process, financial personnel can keenly discover potential problems and errors with their rich experience and expertise. For example, for some complex financial transactions, manual review can make comprehensive judgments based on context and business logic to avoid misjudgments based solely on data. However, manual review also has some obvious limitations. For example, it is highly dependent on personal energy and concentration, and long hours of work can easily lead to fatigue and omissions. Moreover, when the business volume increases significantly, the efficiency of manual review is difficult to keep up, which may cause the review cycle to be extended and affect the company's operational efficiency.

[0004] OCR combined with rule engine: To improve efficiency, some companies have implemented an automation engine based on OCR (optical character recognition) and preset rules to classify and audit vouchers. OCR technology can quickly convert paper vouchers or image vouchers into processable electronic data, while the rule engine screens and judges these data according to pre-set conditions and logic. This greatly reduces the burden of manual work and improves processing speed. Taking the identification and verification of invoices as an example, OCR can accurately read the key information on the invoice, and the rule engine can quickly determine the authenticity and compliance of the invoice. However, this approach is not perfect. Due to the preset nature of the rules, effective identification and processing may not be possible for some special cases or emerging business models. Moreover, the accuracy of the OCR technology itself will also be affected by factors such as the quality of the voucher and the clarity of the font, which may lead to misjudgment.

[0005] These methods have met the basic needs of enterprises to a certain extent, but with the growth of business volume and the increase of business complexity, existing technical means are facing new challenges.

[0006] Prior art, such as a Chinese patent with the patent number "CN117876139A", discloses a financial information management system. The system includes intelligent form filling, intelligent auditing, risk management, expense analysis, budget planning, notification approval and system management modules to achieve all-round financial management. The intelligent form filling module uses OCR technology to quickly input financial data, connects with external data sources, and automatically fills in forms. The intelligent audit module automatically audits financial information through logical verification, and can directly connect with the tax authorities to verify the authenticity of invoices. Risk management automatically intercepts abnormal reimbursement, and the expense analysis module digitally analyzes reimbursement data and generates intelligent reports. The budget and planning module provides financial advice based on historical data, the notification and approval module manages the approval process, and the system management module is responsible for user permissions and system logs. The present invention improves work efficiency, reduces risks, performs digital analysis of financial data, and provides forecasting and planning suggestions, while simplifying process management.

[0007] The problems existing in the above prior art are:

[0008] Poor flexibility of rule engines: rule engines often have difficulty adapting to rapidly changing business needs and diverse credential formats, which limits their scope of application.

[0009] Slow processing speed: Faced with a large number of vouchers, the processing speed of the rule engine is difficult to meet the company's efficiency requirements, becoming a bottleneck in the business process.

[0010] Lack of semantic understanding: Existing technologies are usually unable to deeply understand the semantics of the credential content, which may lead to misunderstanding or incorrect processing of the credential and affect the accuracy of decision-making. Summary of the invention

[0011] In order to solve the above technical problems, the present invention proposes an intelligent credential auditing method and system based on a large language model.

[0012] The technical solution of the present invention is as follows:

[0013] The present invention proposes a method for intelligent voucher auditing based on a large language model, comprising the following steps:

[0014] Step S1, collecting digital copies of user uploaded credentials and adding tags to the uploaded digital copies, and tracking and sorting the processing flow through the added tags;

[0015] Step S2, using optical character recognition (OCR) technology to extract structured text information from the uploaded digital copy;

[0016] Step S3, using a large language model to perform semantic analysis on the text information extracted by optical character recognition (OCR);

[0017] Step S4, classifying the digital copy of the voucher into a preset category based on the result of the semantic analysis;

[0018] Step S5, using preset rules and deep semantic understanding of the large language model, auditing the classified voucher digital copies with reference to the collected original bill structured data;

[0019] Step S6, the classification and audit results of the voucher digital copies are summarized according to the added tags, and the final audit results of the voucher digital copies are obtained and output to the user, and the user manages and adjusts the audit process according to the output results;

[0020] Step S7, based on the user's feedback on the output results, the classification and audit results are stored in the knowledge base to continuously train and optimize the large language model, and automatically adjust the classification and audit rules.

[0021] As a preferred embodiment, the use of optical character recognition (OCR) technology to extract structured text information from the uploaded digital copy includes the following steps:

[0022] removing background noise present in the digital copy of the user uploaded credentials, converting the digital copy image to black and white, and adjusting the brightness and contrast of the digital copy image;

[0023] Locating text regions in the digital copy using a text localization algorithm;

[0024] Perform character segmentation on each text area, use the optical character recognition (OCR) engine to recognize each character, and convert the characters in the image into text data;

[0025] The recognition results are post-processed and text data is output; the post-processing operation includes character connection, error correction, and removal of extra spaces and special characters.

[0026] As a preferred implementation, the digital copies of the vouchers are classified into preset categories based on the results of semantic analysis, where the first-level classification includes: scientific research travel expense reimbursement business, general reimbursement business, conference and training reimbursement business, labor fees, equipment procurement, and human cost reimbursement business; and several second-level classifications are set under each first-level classification.

[0027] As a preferred implementation, the classified digital copies of the vouchers are audited with reference to the collected structured data of the original bills; wherein, the item names of the digital copies of the vouchers are audited, the audited item names are vectorized by the large-scale Chinese text vectorization model text2vec-large-chinese, and the cosine similarity is calculated to measure the similarity with the original bill names. If the calculated similarity is greater than a preset threshold, the voucher item names that are highly similar and semantically identical are retained.

[0028] On the other hand, the present invention also provides an intelligent voucher auditing system based on a large language model, comprising:

[0029] The credential collection module collects digital copies of credentials uploaded by users and adds tags to the uploaded digital copies, and tracks and sorts the processing flow through the added tags;

[0030] Optical character recognition (OCR) processing module, using optical character recognition (OCR) technology to extract structured text information from uploaded digital copies;

[0031] The semantic analysis module uses a large language model to perform semantic analysis on text information extracted by optical character recognition (OCR);

[0032] A classification module, which classifies the digital copy of the voucher into a preset category based on the result of semantic analysis;

[0033] The intelligent audit module uses preset rules and deep semantic understanding of large language models to audit classified digital copies of vouchers with reference to the collected original bill structured data;

[0034] The result output and feedback adjustment module summarizes the classification and audit results of the digital copies of the vouchers according to the added tags, obtains the final audit results of the digital copies of the vouchers and outputs them to the user. The user manages and adjusts the audit process according to the output results;

[0035] The knowledge base learning module stores the classification and audit results into the knowledge base for continuous large language model training and optimization based on user feedback on the output results, and automatically adjusts the classification and audit rules.

[0036] As a preferred implementation, the optical character recognition (OCR) processing module uses optical character recognition (OCR) technology to extract structured text information of the uploaded digital copy, including the following steps:

[0037] removing background noise present in the digital copy of the user uploaded credentials, converting the digital copy image to black and white, and adjusting the brightness and contrast of the digital copy image;

[0038] Locating text regions in the digital copy using a text localization algorithm;

[0039] Perform character segmentation on each text area, use the optical character recognition (OCR) engine to recognize each character, and convert the characters in the image into text data;

[0040] The recognition results are post-processed and text data is output; the post-processing operation includes character connection, error correction, and removal of extra spaces and special characters.

[0041] As a preferred implementation, the classification module classifies the digital copies of the vouchers into preset categories based on the results of semantic analysis, where the first-level classification includes: scientific research travel expense reimbursement business, general reimbursement business, conference and training reimbursement business, labor fees, equipment procurement, and human cost reimbursement business; and several second-level classifications are set under each first-level classification.

[0042] As a preferred implementation, the classification module audits the classified digital copies of the vouchers with reference to the collected original invoice structured data; wherein, the project names of the digital copies of the vouchers are audited, and the audited project names are vectorized by the large-scale Chinese text vectorization model text2vec-large-chinese, and the cosine similarity is calculated to measure the similarity with the original invoice names. If the calculated similarity is greater than a preset threshold, the voucher project names that are highly similar and semantically identical are retained.

[0043] On the other hand, the present invention further provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, the computer program implements an intelligent credential auditing method based on a large language model as described in any embodiment of the present invention.

[0044] On the other hand, the present invention also provides a computer-readable medium for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement an intelligent credential auditing method based on a large language model as described in any embodiment of the present invention.

[0045] The present invention has the following beneficial effects:

[0046] 1. Automated classification and auditing: Through the analysis results of the large language model, the system can automatically classify the vouchers and conduct intelligent auditing, reducing manual intervention and improving work efficiency.

[0047] 2. Self-learning and optimization mechanism: The system has a knowledge base that can store verified classification and audit results, continuously optimize models and rules, and adapt to business development and changes.

[0048] 3. Anomaly detection and compliance assurance: The intelligent audit module can not only detect the consistency of credentials, but also identify anomalies and non-compliance, enhancing the security and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A schematic diagram of a method flow in an embodiment of the present invention;

[0051] Figure 2 is a classification flow chart in an embodiment of the present invention;

[0052] Figure 3 Schematic diagram of the project name audit process in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0055] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0056] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0057] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0058] Embodiment 1:

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be combined with the specific embodiments of the present application and refer to the attached Figure 1 , clearly and completely describe the technical solution of the present invention.

[0060] To solve the problems in the prior art, the present invention provides an intelligent credential auditing method based on a large language model, comprising the following steps:

[0061] Step S1, collecting digital copies of user uploaded credentials and adding tags to the uploaded digital copies, and tracking and sorting the processing flow through the added tags;

[0062] Step S2, using optical character recognition (OCR) technology to extract structured text information from the uploaded digital copy;

[0063] The OCR module is the basic input module of this system, which is responsible for extracting text information from the uploaded digital copy image of the credential. This module uses the latest optical character recognition technology, which can recognize and convert text in different fonts and formats to generate high-precision structured data. In addition, the OCR module can process text in multiple languages ​​to ensure the wide applicability of the system. The specific processing steps are as follows: remove the background noise in the digital copy of the user's uploaded credential, convert the digital copy image to black and white, and adjust the brightness and contrast of the digital copy image;

[0064] Locating text regions in the digital copy using a text localization algorithm;

[0065] Perform character segmentation on each text area, use the optical character recognition (OCR) engine to recognize each character, and convert the characters in the image into text data;

[0066] The recognition results are post-processed and text data is output; the post-processing operation includes character connection, error correction, and removal of extra spaces and special characters.

[0067] Step S3, using a large language model, performs semantic analysis on the text information extracted by optical character recognition (OCR) to ensure a comprehensive understanding of the voucher content, including but not limited to key information such as date, amount, and item description.

[0068] Step S4, classifying the digital copy of the voucher into a preset category based on the result of the semantic analysis;

[0069] The specific classification categories are as follows:

[0070]

[0071]

[0072] The logistics service fees include: water, electricity, gas fees, air conditioning maintenance fees, security and cleaning fees, postage and telecommunications fees, network service fees, plant rental fees, sewage treatment fees, and scientific research funding reimbursement fees.

[0073] The specific classification steps are as follows:

[0074] File content: input the text content analyzed in step S4 into the large language model;

[0075] Construct the first-level classification prompt: "You are a credential file classification assistant. Please classify the file into one of the following categories according to the file content {fill in the first-level classification category here}. The file content is as follows: {fill in the text content here}"

[0076] Large language model: Input the prompt words constructed above into the large language model for understanding and classification.

[0077] Get the first-level classification: Get the first-level classification result based on the large language model.

[0078] Construct secondary classification prompt words: You are a {fill in the primary classification} voucher file classification assistant. You need to classify the voucher into one of the following: {fill in the secondary classification corresponding to the primary classification} according to the voucher content. Please return the type directly without any explanation. The file content is as follows: {fill in the file content}.

[0079] Large language model: Input the secondary classification prompt words into the large language model to understand the classification.

[0080] Obtain secondary classification: Obtain secondary classification results based on the large language model.

[0081] Step S5, using preset rules and deep semantic understanding of the large language model, auditing the classified voucher digital copies with reference to the collected original bill structured data;

[0082] Abnormal and non-compliant audit: Different digital copies of vouchers use different preset audit rules according to different semantically recognized keywords, so that the intelligent audit module can not only detect the consistency of vouchers, but also identify abnormalities and non-compliant situations, enhancing the security and reliability of the system. For example: if the uploaded voucher is a special VAT invoice and it is audited by the intelligent audit module, it is found that the invoice for this project has only one copy (normally it should be three or six copies), then an abnormal report will be generated.

[0083] Consistency Audit:

[0084] Project Name Audit: Figure 3 As shown, the audited voucher item names and the item names of the original bills are vectorized through the large-scale Chinese text vectorization model text2vec-large-chinese, and the similarity is calculated using cosine similarity. If the calculated similarity is greater than the preset threshold, the voucher item names that are highly similar and semantically identical are retained, such as: drug expiration date-drug expiration date, batch number-approval question mark, production date-production date.

[0085] Project content audit: Use the large language model to extract key information of the voucher, such as amount, time, unit price, etc., and compare and review it with the collected original bill structured data through the embedding model.

[0086] Step S6, summarizes the classification and audit results of the digital copies of the vouchers according to the added labels, obtains the final audit results of the digital copies of the vouchers and outputs them to the user, who manages and adjusts the audit process based on the output results; the user can not only view the processing results, but also manage and adjust them to ensure the transparency of the process and the user's control.

[0087] Step S7, based on user feedback on the output results, the classification and audit results are stored in the knowledge base for continuous large language model training and optimization, and the classification and audit rules are automatically adjusted to adapt to changing business needs and standards.

[0088] Embodiment 2:

[0089] This embodiment provides an intelligent voucher audit system based on a large language model, including:

[0090] The credential collection module collects digital copies of credentials uploaded by users and adds tags to the uploaded digital copies, and tracks and sorts the processing flow through the added tags;

[0091] Optical character recognition (OCR) processing module, using optical character recognition (OCR) technology to extract structured text information from uploaded digital copies;

[0092] The semantic analysis module uses a large language model to perform semantic analysis on text information extracted by optical character recognition (OCR);

[0093] A classification module, which classifies the digital copy of the voucher into a preset category based on the result of semantic analysis;

[0094] The intelligent audit module uses preset rules and deep semantic understanding of large language models to audit classified digital copies of vouchers with reference to the collected original bill structured data;

[0095] The result output and feedback adjustment module summarizes the classification and audit results of the digital copies of the vouchers according to the added tags, obtains the final audit results of the digital copies of the vouchers and outputs them to the user. The user manages and adjusts the audit process according to the output results;

[0096] The knowledge base learning module stores the classification and audit results into the knowledge base for continuous large language model training and optimization based on user feedback on the output results, and automatically adjusts the classification and audit rules.

[0097] Embodiment three:

[0098] This embodiment provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, an intelligent credential auditing method based on a large language model as described in any embodiment of the present invention is implemented.

[0099] Embodiment 4:

[0100] This embodiment provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement an intelligent credential auditing method based on a large language model as described in any embodiment of the present invention.

[0101] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0102] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0103] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0104] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0105] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent voucher auditing method based on a large language model, characterized in that: The following steps are involved: Step S1, collecting digital copies of user uploaded credentials and adding tags to the uploaded digital copies, and tracking and sorting the processing flow through the added tags; Step S2, using optical character recognition (OCR) technology to extract structured text information from the uploaded digital copy; Step S3, using a large language model to perform semantic analysis on the text information extracted by optical character recognition (OCR); Step S4, classifying the digital copy of the voucher into a preset category based on the result of the semantic analysis; Step S5, using preset rules and deep semantic understanding of the large language model, auditing the classified voucher digital copies with reference to the collected original bill structured data; Step S6, the classification and audit results of the voucher digital copies are summarized according to the added tags, and the final audit results of the voucher digital copies are obtained and output to the user, and the user manages and adjusts the audit process according to the output results; Step S7, based on the user's feedback on the output results, the classification and audit results are stored in the knowledge base to continuously train and optimize the large language model, and automatically adjust the classification and audit rules.

2. According to claim 1, a smart voucher audit method based on a large language model is characterized in that: The method of extracting structured text information of the uploaded digital copy using optical character recognition (OCR) technology includes the following steps: removing background noise present in the digital copy of the user uploaded credentials, converting the digital copy image to black and white, and adjusting the brightness and contrast of the digital copy image; Locating text regions in the digital copy using a text localization algorithm; Perform character segmentation on each text area, use the optical character recognition (OCR) engine to recognize each character, and convert the characters in the image into text data; The recognition results are post-processed and text data is output; the post-processing operation includes character connection, error correction, and removal of extra spaces and special characters.

3. According to claim 1, a smart voucher audit method based on a large language model is characterized in that: Based on the results of semantic analysis, the digital copies of the vouchers are classified into preset categories, where the first-level classification includes: scientific research travel expense reimbursement business, general reimbursement business, conference and training reimbursement business, labor fees, equipment procurement, and human cost reimbursement business; several second-level classifications are set under each first-level classification.

4. According to claim 1, the intelligent voucher audit method based on a large language model is characterized in that: The classified digital copies of the vouchers are audited with reference to the collected structured data of the original bills; wherein, the project names of the digital copies of the vouchers are audited, the audited project names are vectorized by the large-scale Chinese text vectorization model text2vec-large-chinese, and the similarity with the original bill names is calculated using cosine similarity. If the calculated similarity is greater than a preset threshold, the voucher project names that are highly similar and semantically identical are retained.

5. An intelligent voucher audit system based on a large language model, characterized in that: include: The credential collection module collects digital copies of credentials uploaded by users and adds tags to the uploaded digital copies, and tracks and sorts the processing flow through the added tags; Optical character recognition (OCR) processing module, using optical character recognition (OCR) technology to extract structured text information from uploaded digital copies; The semantic analysis module uses a large language model to perform semantic analysis on text information extracted by optical character recognition (OCR); A classification module, which classifies the digital copy of the voucher into a preset category based on the result of semantic analysis; The intelligent audit module uses preset rules and deep semantic understanding of large language models to audit classified digital copies of vouchers with reference to the collected original bill structured data; The result output and feedback adjustment module summarizes the classification and audit results of the digital copies of the vouchers according to the added tags, obtains the final audit results of the digital copies of the vouchers and outputs them to the user. The user manages and adjusts the audit process according to the output results; The knowledge base learning module stores the classification and audit results into the knowledge base for continuous large language model training and optimization based on user feedback on the output results, and automatically adjusts the classification and audit rules.

6. According to claim 5, the intelligent voucher audit system based on a large language model is characterized in that: The optical character recognition (OCR) processing module uses optical character recognition (OCR) technology to extract structured text information of the uploaded digital copy, including the following steps: removing background noise present in the digital copy of the user uploaded credentials, converting the digital copy image to black and white, and adjusting the brightness and contrast of the digital copy image; Locating text regions in the digital copy using a text localization algorithm; Perform character segmentation on each text area, use the optical character recognition (OCR) engine to recognize each character, and convert the characters in the image into text data; The recognition results are post-processed and text data is output; the post-processing operation includes character connection, error correction, and removal of extra spaces and special characters.

7. The intelligent voucher auditing system based on a large language model according to claim 5 is characterized in that: Based on the results of semantic analysis, the classification module classifies the digital copies of the vouchers into preset categories, where the first-level classification includes: scientific research travel expense reimbursement business, general reimbursement business, conference and training reimbursement business, labor fees, equipment procurement, and human cost reimbursement business; and several second-level classifications are set under each first-level classification.

8. The intelligent voucher auditing system based on a large language model according to claim 5 is characterized in that: The classification module audits the classified digital copies of the vouchers with reference to the collected original invoice structured data; wherein, the project names of the digital copies of the vouchers are audited, the audited project names are vectorized by the large-scale Chinese text vectorization model text2vec-large-chinese, and the cosine similarity is calculated to determine the similarity with the original invoice names. If the calculated similarity is greater than a preset threshold, the voucher project names that are highly similar and semantically identical are retained.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements an intelligent credential auditing method based on a large language model as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements an intelligent credential auditing method based on a large language model as described in any one of claims 1 to 4.

Citation Information

Patent Citations

  • Financial expense auditing method and device based on bill data automatic extraction

    CN110544161A

  • Automatic intelligent reimbursement and accreditation system and method thereof

    CN117876130A

  • Engineering settlement auditing method and system based on LLM large model

    CN118261560A

  • System, Method and Apparatus for Automatic Categorization and Assessment of Billing Narratives

    US20170293951A1

Cited By

  • Data auditing method and device, electronic equipment and nonvolatile storage medium

    CN120541537A

  • Anti-money laundering risk rule generation method and system based on multi-modal large language model

    CN120598567A

  • Financial data auditing method, device and equipment

    CN121120280A