Audit rectification report automatic generation method and system

Abstract

The invention discloses an audit rectification report automatic generation method. The method comprises the following steps: constructing an audit rectification report ftl template; data acquisition and auditing; data identification and analysis: problem semantic extraction, rectification measure quality evaluation and rectification state judgment; automatic report generation: template filling, format conversion, DOCX / PDF output, and log archiving; the invention further discloses an audit rectification report automatic generation system, the generation time of the rectification report is reduced from the hour level to the minute level, and it is ensured that the content is standard, quantifiable and traceable.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for automatically generating audit rectification reports. Background Technology

[0002] The National Audit Office and its internal audit institutions at all levels have included "urging audit rectification" as a core part of their daily supervision, requiring serious accountability for ineffective rectification and false rectification. Rectification results must be submitted in a standardized "Audit Rectification Report," which includes a list of issues, rectification measures, responsible units, completion deadlines, and lessons learned.

[0003] The core process of the traditional template-based automatic audit report generation method is as follows: A template file (*.ftl) with bookmarks is pre-created in Word; data in three columns—problem title, brief description, and rectification result—is collected via a web page; the backend calls Apache POI to write the data column by column into the template according to the bookmark name, generating a rectification report; and PDF conversion and download functions are provided. This method has partially achieved the automatic merging of "data + template" and is currently the most similar existing technology in terms of purpose and architecture to the "automatic generation of audit rectification reports" in this application, as described in the publicly available literature.

[0004] In addition, traditional methods include: 1. Purely manual editing: Auditors paste the issues and measures one by one into Word and manually adjust the format; 2. Semi-automated tools: After collecting data using Excel, the data is merged into Word bookmarks through VBA or simple scripts to achieve field replacement, but the text content (quality of measures, classification of issues) still needs to be judged and written manually.

[0005] Traditional methods still have the following drawbacks: ① Low efficiency: A rectification report with 20 issues takes an average of 4 hours; ② Inconsistent quality: Differences in personnel experience lead to vague descriptions of measures and insufficient quantification; ③ Difficult to verify: Without a unified semantic standard, higher authorities cannot quickly verify whether the rectification is genuine or merely a "textual rectification"; ④ Lack of intelligent assistance: It cannot automatically match historical high-quality rectification cases, nor can it perform semantic-level scoring of measures. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this invention is to provide a method and system for automatically generating audit rectification reports. This invention reduces the time for generating rectification reports from hours to minutes, while ensuring that the content is standardized, quantifiable, and traceable.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a method for automatically generating audit rectification reports, comprising the following steps:

[0008] Step 1: Create an audit rectification report FTL template;

[0009] Step 2: Data collection and review;

[0010] Step 3, Data Identification and Analysis: Semantic extraction of problems, quality assessment of rectification measures, and determination of rectification status;

[0011] Step 4: Automatic report generation: template filling, format conversion, output DOCX / PDF, and log archiving.

[0012] As a further improvement of the present invention, step 1 specifically includes the following steps:

[0013] Step 1.1: Use a Word editor to design a standard layout that includes a table of contents, a list of issues, the results of rectification, and a summary of lessons learned.

[0014] Step 1.2: Insert the positions that need to be dynamically filled into Freemarker variables;

[0015] Step 1.3: Use a loop label for multi-row tables;

[0016] Step 1.4: Save as an XML file and rename it to .ftl to create a template engine file that can be called by Java.

[0017] As a further improvement of the present invention, step 2 specifically includes the following steps:

[0018] Step 2.1: Construct a web form that allows responsible units to enter problem descriptions, problem characteristics, rectification measures, time limits, responsible persons, accountability status, and supporting materials online;

[0019] Step 2.2, Front-end and back-end dual validation: field integrity, date sequence logic, file type and size;

[0020] Step 2.3: Introduce a flowable multi-level review process. If the review fails, the application will be automatically rejected and the results will be logged.

[0021] As a further improvement of the present invention, step 3 specifically includes the following steps:

[0022] Step 3.1, Problem Semantic Extraction: a) Call the pre-trained language model (BERT / RoBERTa) to encode the problem description; b) Add a fully connected layer and Softmax to output the problem type label; c) Simultaneously extract keywords, risk level, and business domain involved and write them into structured fields;

[0023] Step 3.2, Quality Assessment of Rectification Measures: a) Sequential understanding of the rectification measures text is performed using Seq2Seq and Attention models; b) Four-dimensional scores are calculated: executability, relevance, completeness, and closure; c) When the score is less than the preset threshold, the system highlights a warning and pushes similar high-scoring cases from the knowledge base for the user's reference.

[0024] Step 3.3 Automatic determination of rectification status: The rule engine combines the three elements of time limit, supporting materials, and confirmation of responsible person to output completed / in progress / delayed / incomplete.

[0025] As a further improvement of the present invention, step 4 specifically includes the following steps:

[0026] Step 4.1: Load the structured data obtained in Step 3 into a HashMap;

[0027] Step 4.2: Use the Freemarker engine to merge .ftl templates and generate DOCX files instantly;

[0028] Step 4.3: Output PDF via LibreOffice Headless or OnlyOffice conversion service.

[0029] This invention also provides an automatic audit rectification report generation system, comprising:

[0030] The audit issue rectification data collection module is used for online data entry and verification.

[0031] The AI ​​analysis module is used to perform problem classification and semantic analysis.

[0032] Template engine, used for merging data;

[0033] The output module is used to generate official Word and PDF files.

[0034] The beneficial effects of this invention are:

[0035] 1. Efficiency: The processing time for a single report has been reduced from 4 hours to within 2 minutes;

[0036] 2. Quality: The percentage of semantic scores above 90 increased from 42% to 81% (internal experiment with 200 samples).

[0037] 3. Standardization: The format, font, and chapter numbering are 100% in compliance with the latest templates used by the auditing department;

[0038] 4. Traceability: Original data, model version, and review records can be retrieved from any paragraph;

[0039] 5. Economic benefits: Saves 70% on labor costs. Based on 500 reports per year, the savings in labor costs are approximately 1,000 person-days. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention;

[0041] Figure 2 This is a structural block diagram of the quality assessment model for corrective measures in Embodiment 1 of the present invention;

[0042] Figure 3 This is a system framework diagram of Embodiment 2 of the present invention. Detailed Implementation

[0043] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0044] Abbreviation / Code Explanation

[0045] BERT: Bidirectional Encoder Representations from Transformers

[0046] Freemarker: A Java template engine for dynamic text generation.

[0047] DOCX: Microsoft Office Open XML document format

[0048] PDF:Portable Document Format

[0049] Example 1

[0050] like Figure 1 As shown, a method for automatically generating audit rectification reports includes:

[0051] Step 1: Constructing the audit rectification report template:

[0052] 1.1 Use the Word editor to design a standard layout that includes a table of contents, a list of issues, the results of rectification, and a summary of lessons learned;

[0053] 1.2 Insert the locations that need to be dynamically populated into Freemarker variables, such as ${projectName}, ${questionDesc}, and ${measure};

[0054] 1.3 Use the `<#list>` tag to loop through multi-row tables;

[0055] 1.4 Save as an XML file and rename it to .ftl to create a template engine file that can be called by Java.

[0056] Step 2, Data Collection:

[0057] 2.1 Construct a web form that allows responsible units to enter information online, including "problem description, problem type, corrective measures, time limit, responsible person, accountability status, and supporting materials";

[0058] 2.2 Front-end and back-end dual validation: field integrity, date sequence logic, file type and size; 2.3 Introduce a flowable multi-level review process (submission → review → approval), and automatically return applications that fail the review and record logs.

[0059] Step 3: Data Identification and Analysis (AI Core)

[0060] 3.1 Problem semantic extraction: a) Call the pre-trained language model (BERT / RoBERTa) to encode the "problem description"; b) Add a fully connected layer + Softmax to output the "problem type" label (18 categories such as economic responsibility audit, special audit, anti-money laundering, etc.); c) Simultaneously extract keywords, risk level, and business areas involved and write them into structured fields.

[0061] 3.2 Quality assessment of rectification measures: a) Sequential understanding of the “rectification measures” text is performed using the Seq2Seq+Attention model; b) A four-dimensional score is calculated: executability, relevance, completeness, and closure (0-100 points); c) If the score is <60 points, the system will highlight a warning and push similar high-scoring cases from the knowledge base for user reference.

[0062] 3.3 Automatic determination of rectification status: The rule engine combines the three elements of time limit, supporting materials, and confirmation of responsible person to output "completed / in progress / delayed / not completed".

[0063] Step 4: Automatic Report Generation and Output:

[0064] 4.1 Load the structured data obtained in step 3 into a HashMap;

[0065] 4.2 The Freemarker engine is used to merge .ftl templates and generate DOCX files instantly.

[0066] 4.3 Output PDF via LibreOffice Headless or OnlyOffice conversion services;

[0067] Specifically, such as Figure 2 As shown, a four-dimensional quantitative score can be given for each rectification measure, and case recommendations are automatically triggered when the score is below the threshold.

[0068] Example 2

[0069] like Figure 3 As shown, an automatic audit rectification report generation system includes: 1. An audit issue rectification data collection module responsible for online filling and verification; 2. An AI analysis module to complete issue classification and semantic analysis; 3. A template engine to merge data; 4. An output module to generate formal Word and PDF files.

[0070] Example 3

[0071] The audit rectification report includes:

[0072] 1. Data Collection: The audited entity entered 130 rectification measures for identified issues online;

[0073] 2. AI Analysis: The model automatically categorized 104 items related to "economic responsibility" and 26 items related to special audits;

[0074] 3. Quality Assessment: The system indicates that the scores of 3 measures are less than 60. After the system pushes the reference case, the user will supplement the quantitative indicators.

[0075] 4. Generation: After clicking "One-click Export", a 20-page DOCX + PDF will be generated in 2 minutes. It has been reviewed by the auditing department and has not been returned.

[0076] Experimental data:

[0077] Sample: 35 reports from 6 batches from January to June 2025;

[0078] Average generation time: 2 min 23 s;

[0079] The first-time pass rate for audit department reviews was 96.4% (compared to 74% for manual reviews during the same period).

[0080] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for automatically generating audit rectification reports, characterized in that, Includes the following steps: Step 1: Create an audit rectification report FTL template; Step 2: Data collection and review; Step 3, Data Identification and Analysis: Semantic extraction of problems, quality assessment of rectification measures, and determination of rectification status; Step 4: Automatic report generation: template filling, format conversion, output DOCX / PDF, and log archiving.

2. The method for automatically generating audit rectification reports according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1.1: Use a Word editor to design a standard layout that includes a table of contents, a list of issues, the results of rectification, and a summary of lessons learned. Step 1.2: Insert the positions that need to be dynamically filled into Freemarker variables; Step 1.3: Use a loop label for multi-row tables; Step 1.4: Save as an XML file and rename it to .ftl to create a template engine file that can be called by Java.

3. The method for automatically generating audit rectification reports according to claim 2, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Construct a web form that allows responsible units to enter problem descriptions, problem characteristics, rectification measures, time limits, responsible persons, accountability status, and supporting materials online; Step 2.2, Front-end and back-end dual validation: field integrity, date sequence logic, file type and size; Step 2.3: Introduce a flowable multi-level review process. If the review fails, the application will be automatically rejected and the results will be logged.

4. The method for automatically generating audit rectification reports according to claim 3, characterized in that, Step 3 specifically includes the following steps: Step 3.1, Problem Semantic Extraction: a) Call the pre-trained language model (BERT / RoBERTa) to encode the problem description; b) Add a fully connected layer and Softmax to output the problem type label; c) Simultaneously extract keywords, risk level, and business domain involved and write them into structured fields; Step 3.2, Quality Assessment of Rectification Measures: a) Sequential understanding of the rectification measures text is performed using Seq2Seq and Attention models; b) Four-dimensional scores are calculated: executability, relevance, completeness, and closure; c) When the score is less than the preset threshold, the system highlights a warning and pushes similar high-scoring cases from the knowledge base for the user's reference. Step 3.3 Automatic determination of rectification status: The rule engine combines the three elements of time limit, supporting materials, and confirmation of responsible person to output completed / in progress / delayed / incomplete.

5. The method for automatically generating audit rectification reports according to claim 4, characterized in that, Step 4 specifically includes the following steps: Step 4.1: Load the structured data obtained in Step 3 into a HashMap; Step 4.2: Use the Freemarker engine to merge .ftl templates and generate DOCX files instantly; Step 4.3: Output PDF via LibreOffice Headless or OnlyOffice conversion service.

6. An automatic audit rectification report generation system, characterized in that, include: The audit issue rectification data collection module is used for online data entry and verification. The AI ​​analysis module is used to perform problem classification and semantic analysis. Template engine, used for merging data; The output module is used to generate official Word and PDF files.

Images