A standard intelligent audit and compliance inspection method, system, terminal and medium

Abstract

The present application relates to the field of intelligent audit data, and in particular to a standard intelligent audit and compliance inspection method, system, terminal and medium, which includes: obtaining a document to be audited; dividing the document to be audited into multiple document elements based on the document to be audited and preset classification rules; analyzing the document to be audited based on multiple document elements and the preset audit direction corresponding to each document element; generating an audit report and a review draft based on the audit results of each document element. The present application improves the efficiency and accuracy of document auditing.

Description

Technical Field

[0001] The present application relates to the field of intelligent audit data, and in particular to a standard intelligent audit and compliance checking method, system, terminal and medium. Background Art

[0002] In the modern production and service industry, standardization management has become a key factor in improving product quality, service level and market competitiveness. The standard system covers multi-level and multi-field specifications such as group standards, national standards, industry standards and local standards. While ensuring consistency and standardization, they also promote the healthy development of the industry. However, with the changes in the market environment and the frequent updates of standards, traditional standard audits and compliance inspection methods have gradually revealed many problems.

[0003] Existing standard audit methods usually rely on manual review and paper documents, which is not only time-consuming and labor-intensive, but also susceptible to human subjective factors. Errors and omissions in the manual audit process may lead to inaccurate compliance checks, thus affecting the effective implementation of standards. In addition, traditional methods are difficult to cope with massive amounts of data and complex standard requirements, and often require a lot of time and resources to complete audit tasks. Summary of the invention

[0004] On the first hand, in order to improve the efficiency and accuracy of standard document review, the present application provides a standard intelligent review and compliance checking method.

[0005] This application provides a standard intelligent audit and compliance inspection method, which adopts the following technical solutions:

[0006] A standard intelligent audit and compliance inspection method, system, terminal and medium, including:

[0007] Get documents for review;

[0008] Based on the documents to be reviewed and preset classification rules, the documents to be reviewed are divided into multiple document elements;

[0009] Analyze the documents to be reviewed based on multiple document elements and the preset review direction corresponding to each document element;

[0010] Generate an audit report and a review draft based on the audit results of each document element.

[0011] By adopting the above technical solution, after the system obtains the input document to be reviewed, it decomposes the document into multiple document elements and reviews them one by one, and finally generates a corresponding review report and a review draft. During the process, it ensures that each document element is checked in detail, which improves the comprehensiveness and accuracy of the review. In addition, each document element corresponds to a different preset review direction, which improves the accuracy of the document review. Finally, through system review, manual operations and review time are reduced, the review cycle is accelerated, and the efficiency and accuracy of document review are improved.

[0012] Preferably, the step of analyzing the document to be reviewed based on multiple document elements and the preset review direction corresponding to each document element includes:

[0013] The preset review direction corresponding to each document element includes a specific review direction and a general review direction, wherein the general review direction includes a typo detection direction, a content missing detection direction, and a duplicate review direction;

[0014] Based on the document elements, a corresponding specific review direction model is obtained, wherein the specific review direction model uses a pre-trained Chinese model, and the model is pre-trained in a standard field;

[0015] The specific review direction model automatically reviews document elements based on pre-trained domain knowledge;

[0016] Get general review direction based on document elements;

[0017] Perform text preprocessing on document elements to obtain preprocessed text and a pre-trained typo detection model;

[0018] Use a pre-trained typo detection model to identify common typos in pre-processed text;

[0019] Correct common typos in preprocessed text and generate preliminary processed text;

[0020] Acquire a contrastive vocabulary base based on preliminary processed text;

[0021] Compare the words in the document to be tested with the words in the standard dictionary, detect and mark the wrong words;

[0022] Modify the incorrect words in the preliminary processed text and generate the corrected text;

[0023] Compare the document generated by merging multiple error correction texts with the preset template document;

[0024] If they are different, a re-upload control will be displayed on the human-computer interaction page;

[0025] If they are the same, generate the text to be deduplicated;

[0026] Obtain document features and a literature retrieval library based on the text to be deduplicated;

[0027] Retrieve identical or similar documents from a document retrieval database based on document features of the text to be deduplicated;

[0028] Analyze the similarity between the text to be deduplicated and similar documents.

[0029] By adopting the above technical solutions, the step-by-step process ensures that each link has clear tasks and goals. From typo review to content missing check to repeated review, the entire process covers the main aspects of document review, improves the accuracy and efficiency of the review, and reduces the possibility of human errors and missed reviews.

[0030] Preferably, the method further comprises:

[0031] Obtaining review orders;

[0032] Obtain revised documents based on review instructions;

[0033] Based on the modified document, calculate the similarity between the original document and the modified document;

[0034] Execution of the preliminary judgment: Compare the similarity with the limit value;

[0035] If the similarity is higher than the limit, the modified document is retrieved again;

[0036] If the similarity is lower than the limit, the modified document will be analyzed and reviewed again.

[0037] By adopting the above technical solution, by calculating the similarity between the original and the modified document, it is possible to determine whether the modification is effective. If the similarity is high, it means that the modification has not effectively changed the document content, thereby avoiding duplication of work and reducing review omissions caused by insufficient modification.

[0038] Preferably, after the steps of generating an audit report based on the audit results of each document element and generating a review draft, the process further includes:

[0039] Calculate the recommended revision degree of each document element based on the review results of each document element;

[0040] Sorting and generating a modification order based on the suggested modification degree of each document element;

[0041] Display the modification order on the document to be reviewed on the human-computer interaction page based on the modification order;

[0042] Obtaining a preliminary modified text, wherein the preliminary modified text is a modified text of the to-be-reviewed document at the top of the modification order;

[0043] Adjust the modification order of the documents to be reviewed in the latter column based on the preliminary modified text and adjust the modification order.

[0044] By adopting the above technical solution, by calculating the recommended modification degree of each document element and sorting them, it is possible to identify which modifications are the most important or urgent, thereby optimizing the revision process and ensuring that the most critical modifications are processed first; processing document elements in the order of modification makes the revision process more systematic and orderly, reduces confusion and duplication of work in revision, and improves overall efficiency; by giving priority to document elements with a high recommended modification degree, the most critical problems can be solved more effectively, thereby improving the quality and compliance of the final document.

[0045] Preferably, the step of generating an audit report based on the audit results of each document element and generating a review draft includes:

[0046] Get spot check instructions;

[0047] Based on the spot check instruction, obtain the documents to be reviewed that have generated the review draft, and hide the review draft;

[0048] Obtain the manual review draft input by the spot checker based on the document to be reviewed for which the review draft has been generated;

[0049] Based on the review draft and the manual review draft of the document to be reviewed, similarity is calculated and generated;

[0050] If the similarity exceeds the set value, a review operation is performed: obtaining review information and generating a comparison draft based on the review information;

[0051] If the similarity does not exceed the set value, the sampling personnel and the system review are deemed qualified, and a sampling draft is generated.

[0052] By adopting the above technical solution, the sampling personnel hide the review draft that has been generated for the document to be reviewed during the sampling, and review the document to be reviewed and finally input the manual review draft. By comparing the similarity between the review draft and the manual review draft, the system can verify the accuracy of the sampling personnel's review and the system review, thereby determining whether the system review is reliable, whether the sampling personnel are lazy, and the level of the sampling personnel, thereby improving the credibility and consistency of the review.

[0053] Preferably, the step of calculating and generating similarity based on the review draft and the manual review draft of the document to be reviewed further includes:

[0054] Based on the review draft and the manual review draft of the document to be reviewed, obtaining annotation information of the review draft and the manual review draft, the annotation information including the number of annotations and the position of the annotations;

[0055] Perform the first similarity determination: compare the number of comments in the review draft with the number of comments in the manual review draft;

[0056] If the number of comments in the review draft is different from the number of comments in the manual review draft, obtain the difference between the number of comments in the review draft and the number of comments in the manual review draft, and obtain the number of non-overlapping positions of the comments in the review draft and the positions of the comments in the manual review draft;

[0057] Based on the quantity difference and the non-overlapping quantity, the similarity is calculated and generated;

[0058] If the number of comments in the review draft is the same as the number of comments in the manual review draft, obtain the number of differences between the comment positions in the review draft and the comment positions in the manual review draft;

[0059] The similarity is calculated based on the number of non-overlapping annotation positions in the review draft and the manually reviewed draft.

[0060] By adopting the above technical solution, by comparing the number and position of annotations, the consistency of the review draft and the manually reviewed draft can be evaluated more accurately. By comparing the number and position of annotations, detailed feedback on the specific differences between the review draft and the manually reviewed draft can be provided, helping to improve the document and review process.

[0061] Preferably, the method further comprises:

[0062] Obtain modification information of corresponding documents based on comparison draft and sampling draft;

[0063] Adjusting the document information of the to-be-reviewed document based on the modification information and obtaining a revised document;

[0064] Determine whether the document is the final version of the uploaded document;

[0065] If it exists, add a historical document reference area and enter the revised document into the historical parameter reference area.

[0066] By adopting the above technical solution, the system will enter the revised document generated by merging the comparison draft and the inspection draft into the historical parameter reference area based on the documents uploaded by the spot checkers. Since some documents have been delivered or completed due to customer requirements, it will be convenient for later tracing, or as an aid for subsequent reply corrections and deliberations, thereby improving efficiency.

[0067] Secondly, in order to improve the efficiency and accuracy of data auditing, this application provides a standard intelligent auditing and compliance inspection system, which adopts the following technical solutions:

[0068] A standard intelligent audit and compliance checking system, comprising:

[0069] Data entry module, used to obtain documents to be reviewed;

[0070] An element extraction module is used to divide the document to be reviewed into multiple document elements based on the document to be reviewed and preset classification rules;

[0071] An intelligent analysis module for analyzing the document to be reviewed based on multiple document elements and the preset review direction corresponding to each document element;

[0072] The result output module is used to generate an audit report and a review draft based on the audit results of each document element.

[0073] By adopting the above technical solution, the system can obtain the document to be reviewed through the data input module, and then divide the document to be reviewed into multiple document elements through the element extraction module, and then analyze the document to be reviewed through the intelligent analysis module, and finally generate an audit report and a review draft through the result output module, thereby improving the efficiency and accuracy of document review.

[0074] In a third aspect, the present application provides a smart terminal, which adopts the following technical solution:

[0075] An intelligent terminal includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the above-mentioned standard intelligent audit and compliance checking method.

[0076] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0077] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the above-mentioned standard intelligent audit and compliance checking method.

[0078] In summary, the present application includes at least one of the following beneficial technical effects:

[0079] 1. After the system obtains the input document to be reviewed, it decomposes the document into multiple document elements and reviews them one by one, and finally generates the corresponding review report and review draft. During this period, it ensures that each document element has been checked in detail, which improves the comprehensiveness and accuracy of the review. In addition, each document element corresponds to a different preset review direction, which improves the accuracy of the document review. Finally, through system review, manual operations and review time are reduced, the review cycle is accelerated, and the efficiency and accuracy of document review are improved;

[0080] 2. The step-by-step process ensures that each link has clear tasks and goals. From typo review to content missing check to repeated review, the entire process covers the main aspects of document review, improves the accuracy and efficiency of the review, and reduces the possibility of human errors and missed reviews;

[0081] 3. When spot checking, the sampling personnel hide the review draft that has been generated for the document to be reviewed, and review the document to be reviewed and finally input the manual review draft. By comparing the similarity between the review draft and the manual review draft, the system can verify the accuracy of the sampling personnel's review and the system review, thereby determining whether the system review is reliable, whether the sampling personnel are lazy, and the level of the sampling personnel, thereby improving the credibility and consistency of the review. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps S100-S400.

[0083] Figure 2 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SA1-SA5.

[0084] Figure 3 It is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SB1-SB6.

[0085] Figure 4 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SC1-SC3.

[0086] Figure 5 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SD1-SD6.

[0087] Figure 6 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SE1-SE6.

[0088] Figure 7 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SF1-SF4.

[0089] Figure 8 This is a partial method flow chart of a standard intelligent audit and compliance checking method of an embodiment of the present application, mainly showing steps SG1-SG4. DETAILED DESCRIPTION

[0090] The present application is further described in detail below in conjunction with all the accompanying drawings.

[0091] The present application embodiment discloses a standard intelligent audit and compliance inspection method. Figure 1 , a standard intelligent audit and compliance checking method includes:

[0092] S100: Obtain the document to be reviewed.

[0093] Specifically, a file upload box and a text input box can be displayed on the human-computer interaction page. The file upload box allows the user to drag files to upload word and pdf files of the document to be reviewed, and the text input box allows the user to enter text of the document to be reviewed.

[0094] S200: Divide the document to be reviewed into multiple document elements based on the document to be reviewed and preset classification rules.

[0095] Specifically, the document to be reviewed is divided into multiple document elements, including title and foreword, normative references, terms and definitions, requirements and test methods, implementation and appendix. The preset classification rules can be based on the title of each document element, and the content of each document element is the content after its title. NLP technology (such as word segmentation and entity recognition) can also be used to extract and identify different parts of the document, and the document content can be classified using a rule engine (such as Drools) or a machine learning model (such as a classifier).

[0096] S300: Analyze the document to be reviewed based on multiple document elements and a preset review direction corresponding to each document element.

[0097] Specifically, the document elements include title and foreword, normative references, terms and definitions, requirements and test methods, and implementation and appendix. The preset review directions corresponding to each document element include specific review directions and general review directions. The specific review direction is to review each document element according to the format rules of each document element and the specific requirements of each document element. The general review direction applies to all document elements, including: typo detection: identifying and correcting text and vocabulary errors in documents; content missing detection: determining whether the document is missing document elements; duplication detection: evaluating whether the document is original.

[0098] The specific steps of the review process include:

[0099] Step S1: based on the document elements, a corresponding specific review direction model is obtained, wherein the specific review direction model uses a pre-trained Chinese model, and the model is pre-trained in a standard field;

[0100] Step S2: the specific review direction model automatically reviews the document elements based on pre-trained domain knowledge;

[0101] Specifically, choose a pre-trained model: use a Chinese pre-trained model (such as BERT, RoBERTa) and fine-tune it on data from a standard domain. Domain fine-tuning: train the model in a specific standard domain so that it can understand and process document-related terms and requirements. Among them, the pre-training part of the specific audit direction model includes but is not limited to the documents of group standards, national standards, industry standards and local standards (these documents should include the complete text, samples, specifications and implementation details of various standards), namely ISO standards, national quality standards, local standards, etc.; professional terms and definitions related to group standards, national standards, industry standards and local standards, such as special terms, normative terms, technical terms, etc. in various standard documents; national and local laws and regulations, which may involve the formulation and implementation of standards, such as the Standardization Law of the People's Republic of China, local regulations, industry regulations, etc.; relevant standardization policies and guidance documents, such as policy documents and guidelines issued by the National Standardization Administration, such as the "Regulations on the Implementation of the Standardization Law" and "Documents of the National Standardization Administration". Standard documents at different levels usually have fixed structures and formats, which need to be understood and processed, for example: title, foreword, normative reference documents, terms and definitions, requirements and test methods, specific formats and specifications of implementation and appendix; document formatting requirements, such as fonts, typesetting, reference format, etc., for example: ISO format specifications, industry document format requirements, etc. The model needs to be trained to understand and process the specific content and clauses in specific standard documents, including the actual text and implementation requirements of different standards, as well as pre-training on the intersection and integration of standards at different levels (such as national standards and local standards) to identify and process the hierarchical relationships and cross-references that may exist in these documents. Finally, each document element is input into the specific audit direction model, and the model analyzes the document elements based on domain knowledge to check whether each element meets the requirements of each document element.

[0102] Step S3: obtaining a general review direction based on document elements;

[0103] Step S4: Perform text preprocessing on the document elements to obtain preprocessed text and a pretrained typo detection model;

[0104] Step S5: using a pre-trained typo detection model to identify common typos in the pre-processed text;

[0105] Step S6: modify common typos in the preprocessed text and generate a preliminary processed text;

[0106] Step S7: obtaining a comparison vocabulary based on the preliminary processed text;

[0107] Step S8: Compare the words in the document to be detected with the words in the standard dictionary, detect and mark the wrong words;

[0108] Step S9: modifying the wrong words in the preliminary processed text and generating a correction text;

[0109] Specifically, after a specific review of the document, typos are detected. First, the document text is cleaned up, such as removing unnecessary punctuation marks, converting to lowercase, etc. The text is decomposed into vocabulary and common stop words are removed to more efficiently process the text and generate preprocessed text. The pre-trained typo detection model is then used to identify spelling errors in the text, the detected errors are corrected, and a preliminarily processed text is generated. Finally, the vocabulary is corrected and a comparative vocabulary library is constructed: a standard dictionary is used as a comparison library, the vocabulary in the document to be detected is compared with the standard dictionary, typos are identified, typos are corrected, and the final corrected text is generated.

[0110] Step S10: comparing a document generated by merging multiple error correction texts with a preset template document;

[0111] Step S11: If they are different, then a re-upload control is displayed on the human-computer interaction page;

[0112] Step S12: If they are the same, generate the text to be deduplicated;

[0113] Specifically, after conducting a specific review and correcting typos, the corrected text is checked for missing content, and a machine learning model is used to automatically scan the document against a preset template document to identify missing parts. The document structure is reviewed first: the document is reviewed and compared with the document standard outline or preset structure list to ensure that all required parts (such as title, preface, definition of terms, requirements and test methods, etc.) are included; a checklist is used to ensure that each chapter or section contains the expected content. For example, whether the requirements section covers all technical standards and specifications; compare templates or standards: check whether some common parts are missing by comparing with predefined templates or standard documents. For example, whether normative references or related appendices are included; compare industry standards or specification documents to ensure that all required requirements and test methods are included; finally, text analysis tools or natural language processing (NLP) technology are used to identify and mark common terms or topics that are missing in the document. For example, automatically detect whether all key terms or topics are mentioned, and mark and annotate them in the corresponding positions. Finally, if all of the above reviews are passed, the text to be deduplicated will be generated. If one or more of them fail, a re-upload box will be displayed on the human-computer interaction page, and the user can drag the file to upload the word or pdf file of the document to be reviewed to supplement the missing content.

[0114] Step S13: obtaining document features and a document retrieval library based on the text to be deduplicated;

[0115] Step S14: Retrieve identical or similar similar documents from a document retrieval database based on the document features of the text to be deduplicated;

[0116] Step S15: Analyze the similarity between the text to be deduplicated and similar documents.

[0117] Specifically, first obtain the document retrieval database online or input it in advance, and then preprocess the deduplicated text, including text cleaning, format standardization and word segmentation processing to ensure data consistency, extract text features from the deduplicated text, including keywords, subject terms, phrases and paragraph structure to generate a document feature vector, and use a similarity calculation algorithm (such as cosine similarity, Euclidean distance or Manhattan distance) to compare the feature vector of the deduplicated text with the feature vector of the database document to evaluate the similarity between the documents, and based on the similarity calculation results, retrieve documents with high similarity to the document to be audited, and store them in order of similarity to facilitate the subsequent generation of audit reports.

[0118] S400: Generate an audit report and a review draft based on the audit results of each document element.

[0119] Specifically, the audit results of each document element are summarized to generate a detailed audit report. The audit report usually includes: Document Overview: Contains basic information of the document, such as title, version, review date, etc. Audit Findings: Summarizes the problems and non-conformities found during the audit; Suggestions and Improvements: Provides improvement suggestions and solutions; Conclusion: Gives the audit conclusion, such as whether it has passed the audit, needs further modification, etc. The review draft is the original manuscript with annotations and modification opinions.

[0120] Reference Figure 2 ,In addition, the marked documents to be reviewed can be displayed on the ,human-computer interaction page. Specific methods include:

[0121] Step SA1: Obtain the review draft;

[0122] Step SA2: Based on the review draft, obtain modification information of the corresponding document;

[0123] Step SA3: Based on the modification information of the corresponding document, a modification mark is displayed on the document to be reviewed on the human-computer interaction page;

[0124] Step SA4: Obtain a trigger instruction for modifying the mark;

[0125] Step SA5: Based on the trigger instruction of the modification mark, the modification suggestion is displayed on the human-computer interaction page.

[0126] Specifically, the review draft file or data is extracted from the system, the modification suggestions in the review draft are parsed and associated with the original document, the modification information in the review draft (such as modification location, suggested content) is specifically extracted, the modification information is mapped to the specific location of the original document and the location is displayed in red, so that the modification mark is displayed on the document to be reviewed on the human-computer interaction page. The writer or other personnel clicks the red marked location with a mouse or touch, which is the modification mark, and a comment or annotation box is displayed next to the original manuscript on the human-computer interaction interface.

[0127] Reference Figure 3 , the writer modifies the original document based on the review report and the review draft, and then re-uploads it to the system for review again. The specific methods include:

[0128] Step SB1: Obtain review instructions;

[0129] Step SB2: based on the review instruction, obtaining the modified document;

[0130] Step SB3: Based on the modified document, calculate the similarity between the original document and the modified document;

[0131] Step SB4: perform preliminary determination: compare the similarity with the limit value;

[0132] Step SB5: if the similarity is higher than the limit value, re-acquire the modified document;

[0133] Step SB6: If the similarity is lower than the limit, the modified document is analyzed and reviewed again and a review draft is generated.

[0134] Specifically, the review instruction is input by the team leader or manager. For example, a review control will be displayed on the human-computer interaction interface. When the team leader or manager triggers the review control through touch input or external input, a review instruction can be generated. After receiving the review instruction, the system displays a file upload box and a text input box on the human-computer interaction page. The file upload box allows users to drag files to upload word or pdf files for modifying documents. The text input box allows users to enter text for modifying documents and displays a confirmation control. After the user clicks the confirmation control, the system uses, for example, Cosine Similarity, JaccardIndex or model-based BERT cosine similarity to calculate the content similarity between the original and the modified document. Specific text preprocessing: remove stop words and punctuation marks, perform stem extraction or word form restoration; bag-of-words model: convert documents into term frequency vectors (Term Frequency - TF); TF-IDF: use TF-IDF to calculate the weighted value of words; calculate similarity: use the cosine similarity formula: to calculate the similarity. It can also be text preprocessing: input the document into the pre-trained model; obtain vector representation: extract the vector representation of the document from the model (for example, the representation of the [CLS] token); calculate similarity: use cosine similarity or other measurement methods. Set a limit value, such as 0.98. If the similarity is higher than the limit value, the document does not meet the requirements in red on the human-computer interaction page, and redisplay a file upload box and a text input box. If the similarity is lower than the limit value, the modified document is analyzed and reviewed again and a review draft is generated. The specific steps are as follows S100-S400.

[0135] Reference Figure 4 , and the modification order of all document elements of each document to be reviewed can also be obtained according to the review result of each document element. The specific steps include:

[0136] Step SC1: Calculate the suggested modification degree of each document element based on the review result of each document element;

[0137] Step SC2: sorting and generating a modification order based on the suggested modification degree of each document element;

[0138] Step SC3: displaying the modification order on the document to be reviewed on the human-computer interaction page based on the modification order;

[0139] Step SC4: obtaining a preliminary modified text, wherein the preliminary modified text is a modified text of the to-be-audited document at the top of the modification order;

[0140] Step SC5: Adjust the modification order of the documents to be reviewed in the latter column based on the preliminary modification text and adjust the modification order.

[0141] Specifically, each document element may have different results after review, such as: error type (formal error: typo; format problem; missing content), three levels of error severity (minor, medium, severe), number of improvement suggestions (specific modification suggestions), and develop a scoring system to quantify the severity of the error. It can be defined in the following ways: Weight assignment: assign a weight value to each error type and severity. For example, the weight of formal errors is 10, the weight of format problems is 5, and the weight of missing content is 20; Score calculation: For each document element, calculate its total recommended modification score. For example, if a document element has 3 errors, one of which is a severe error (weight 20) and two are medium errors (weight 10), then the total score of this element is 20 + 10 + 10 = 40, and the calculated modification score is used to rank the document elements. Elements with higher modification scores have more and more important modification suggestions, so they should be processed first. All document elements are sorted from high to low according to the modification scores to determine the modification priority. The modification order number is displayed next to each document element in the document to be reviewed on the human-computer interaction page. When the writer modifies the first-column elements in the document according to the generated modification order, the system saves the modified text as "preliminary modification text" as the basis for subsequent adjustments. The preliminary modification text is compared with the original document to be reviewed to analyze the specific content of the modification. This can be achieved using a text difference comparison tool, such as the modification of formal errors in the preliminary modification text, that is, technical feature modifications and similar label modifications. Based on these modifications, text replacement, supplementation and other methods can be used to synchronize these modifications in the document to be reviewed that is listed later in the modification order. Finally, re-evaluate the degree of modification: re-evaluate the degree of modification of the remaining document elements based on the preliminary modified text. The preliminary modification may affect the modification priority of other elements; update the score: re-score the remaining document elements based on the preliminary modified text. The modified content may introduce new errors or affect the severity of the original errors; calculate the new recommended degree of modification: update the recommended degree of modification of each document element based on the updated score; re-order: re-order the document elements based on the new recommended degree of modification to generate a new modification order; update the user interface: update the adjusted modification order on the user interface to reflect the new modification priority.

[0142] Reference Figure 5 Since there is a probability of errors in the system review documents, spot checkers are required to conduct regular spot checks on them. The specific steps include:

[0143] Step SD1: Get the spot check instruction:

[0144] Step SD2: based on the spot check instruction, obtain the document to be reviewed for which the review draft has been generated, and hide the review draft;

[0145] Step SD3: obtaining a manual review draft input by a spot checker based on the document to be reviewed for which a review draft has been generated;

[0146] Step SD4: Calculate and generate similarity based on the review draft and the manual review draft of the document to be reviewed;

[0147] Step SD5: If the similarity exceeds the set value, a review operation is performed: review information is obtained, and a comparison draft is generated based on the review information;

[0148] Step SD6: If the similarity does not exceed the set value, the sampling personnel and the system review are determined to be qualified, and a sampling draft is generated.

[0149] Specifically, the sampling instruction is input by the sampling personnel. For example, a sampling control will be displayed on the human-computer interaction interface. When the sampling personnel triggers the sampling control through touch input or external input, the sampling instruction can be generated. After receiving the sampling instruction, the system takes out one or more documents to be sampled from all the documents to be reviewed for which review drafts have been generated, hides their review drafts, and displays their original manuscripts on the human-computer interaction page. The sampling personnel approves them through the correction tool and displays the end control. After the sampling personnel clicks the end control, the system obtains the manual review draft entered by the sampling personnel, and calculates the content similarity between the review draft and the manual review draft through the calculation steps SE1-SE6 below. If the similarity exceeds the set value, the review operation is performed: the original manuscript is redisplayed on the human-computer interaction page, and the team leader or audit team re-approves it, and the end control is displayed. After the document team leader or audit team clicks the end control, the system obtains the review information and generates a comparison draft based on the review information. If the similarity does not exceed the set value, the qualified text will be displayed on the review draft of the human-computer interaction page, and a sampling draft will be generated. The sampling draft is the original manuscript with modification suggestions and modifications.

[0150] Reference Figure 6 , wherein the specific steps of calculating and generating similarity based on the review draft and the manual review draft of the document to be reviewed include:

[0151] Step SE1: Based on the review draft and the manual review draft of the document to be reviewed, obtain the annotation information of the review draft and the annotation information of the manual review draft, where the annotation information includes the number of annotations and the annotation position;

[0152] Step SE2: Execute the first similarity determination: compare the number of comments in the review draft with the number of comments in the manual review draft;

[0153] Step SE3: If the number of comments in the review draft is different from the number of comments in the manual review draft, obtain the difference between the number of comments in the review draft and the number of comments in the manual review draft, and obtain the number of non-overlapping positions of the comments in the review draft and the positions of the comments in the manual review draft;

[0154] Step SE4: Calculate and generate similarity based on the quantity difference and the number of non-overlapping.

[0155] Step SE5: If the number of comments in the review draft is the same as the number of comments in the manual review draft, obtain the number of differences between the comment positions in the review draft and the comment positions in the manual review draft;

[0156] Step SE6: Calculate and generate similarity based on the number of non-overlapping values ​​between the annotation positions of the review draft and the annotation positions of the manual review draft.

[0157] Specifically, annotation information is extracted from the review draft and the manual review draft of the document to be reviewed. The annotation information includes the number of annotations in each document and the specific position of each annotation (such as page number, paragraph number, line number, etc.), determine the difference in the number of annotations between the review draft and the manual review draft, identify the number of annotation positions in the review draft that are inconsistent with the annotation positions in the manual review draft, and calculate the similarity based on the difference in the number of annotations and the number of non-overlapping annotation positions. The similarity can be calculated by taking these differences as influencing factors and applying a certain weighted formula to measure the overall similarity. For example, the ratio of the total difference (the difference in quantity plus the number of non-overlapping) to the maximum possible difference can be calculated to obtain a similarity score. If the number of annotations in the review draft and the manual review draft is the same, then: Calculate the different number of annotation positions: compare the position of each annotation in the two and calculate the number of inconsistent positions. Calculate the similarity: Calculate the similarity based on the number of non-overlapping annotation positions. The similarity score can be determined by the ratio of the number of non-overlapping annotation positions to the number of all annotation positions. For example, the ratio of the number of non-overlapping annotation positions to the total number of annotation positions can be calculated, and the similarity value can be obtained from it. Calculate similarity (different quantity cases): Calculate the similarity using the quantity difference and the non-overlapping quantity.

[0158] Calculate similarity (when the number is the same): When the number of annotations is the same, the similarity is calculated based only on the number of non-overlapping annotation positions.

[0159] Reference Figure 7 Finally, since some documents have been delivered or completed due to customer requirements, the revised documents are entered into the historical parameter reference area. The specific steps include:

[0160] Step SF1: obtaining modification information of corresponding documents based on the comparison draft and the sampling draft;

[0161] Step SF2: adjusting the document information of the document to be reviewed based on the modification information and obtaining a revised document;

[0162] Step SF3: Determine whether the document is a final version that has been uploaded;

[0163] Step SF4: If it exists, add a historical document reference area and enter the revised document into the historical parameter reference area.

[0164] Specifically, obtain the modification records made to the comparison draft and the sampling draft, including the specific modification content and location. This may include added, deleted or modified content, as well as the specific location of the modification (such as chapters, paragraphs, line numbers, etc.), load the original version of the document to be reviewed, adjust the document to be reviewed one by one according to the modification information extracted from the comparison draft and the sampling draft, and after applying all the modification information, save the updated version of the document, that is, the revised document, and then the system checks whether the current document version has been marked as the final draft. If the document is the final draft, the team leader and management personnel will mark it as the finished draft. If there is a mark, it is determined to be the final draft of the document. If so, in the document management system, the revised document will be entered into the history record area. The history record area is an area that saves all document versions and their modification history. The relevant information of the revised document (such as version number, modification content, modification time, etc.) is added to the historical parameter reference area for future reference and review.

[0165] Reference Figure 8 In addition, after each audit, review and random inspection, common errors in the documents to be audited can be recorded and an error library can be established. The specific methods include:

[0166] Step SG1: Obtain the review draft, re-review draft, comparison draft and documents to be reviewed;

[0167] Step SG2: Obtain corresponding modification content based on the review draft, re-review draft, comparison draft and document to be reviewed;

[0168] Step SG3: Obtain error information based on the modified content;

[0169] Step SG4: Establish an error library based on the error information.

[0170] Specifically, after the review, re-review and random review process, use text comparison tools or algorithms to compare the review draft, re-review draft, comparison draft with the document to be reviewed, identify and extract the newly added, deleted or modified parts of the text to generate the modified content, and use a rule engine or machine learning model to classify the modified content corresponding to the review draft, re-review draft, comparison draft and the document to be reviewed, and record the identified errors in the error library, define the error type (such as typos, missing content errors, content format errors, etc.) according to the modified content, to facilitate the subsequent writing training of writers before writing.

[0171] The implementation principle of a standard intelligent audit and compliance checking method in an embodiment of the present application is as follows: after the system obtains the input document to be audited, it decomposes the document into multiple document elements and audits them one by one, and finally generates a corresponding audit report and a review draft. During the process, it ensures that each document element is subject to detailed inspection, thereby improving the comprehensiveness and accuracy of the audit, and each document element corresponds to a different preset audit direction, thereby improving the accuracy of the document audit. Finally, through system audit, manual operations and audit time are reduced, the audit cycle is accelerated, and the efficiency and accuracy of the document audit are improved.

[0172] The embodiment of the present application discloses a standard intelligent audit and compliance checking system.

[0173] A standard intelligent audit and compliance checking system, comprising:

[0174] Data entry module, used to obtain documents to be reviewed;

[0175] An element extraction module is used to divide the document to be reviewed into multiple document elements based on the document to be reviewed and preset classification rules;

[0176] An intelligent analysis module is used to analyze the document to be reviewed based on multiple document elements and the preset review direction corresponding to each document element. The preset review direction corresponding to each document element includes a specific review direction and a general review direction. The general review direction includes a typo detection direction, a content missing detection direction, and a duplicate review direction; based on the document element, a corresponding specific review direction model is obtained, wherein the specific review direction model uses a pre-trained Chinese model, and the model is pre-trained in a standard field; the specific review direction model performs automated review on the document element based on the pre-trained field knowledge; a general review direction is obtained based on the document element; text pre-processing is performed on the document element to obtain pre-processed text and a pre-trained typo detection model; and The pre-trained typo detection model identifies common typos in the pre-processed text; modifies common typos in the pre-processed text and generates a preliminary processed text; obtains a comparison vocabulary library based on the preliminary processed text; compares the vocabulary in the document to be detected with the vocabulary in the standard dictionary, detects and marks the typos; modifies the typos in the preliminary processed text and generates a correction text; compares the document generated by merging multiple correction texts with the preset template document; if they are different, displays a re-upload control on the human-computer interaction page; if they are the same, generates a text to be deduplicated; obtains document features and a literature retrieval library based on the text to be deduplicated; retrieves the same or similar similar documents from the literature retrieval library based on the document features of the text to be deduplicated; analyzes the similarity between the text to be deduplicated and similar documents;

[0177] The result output module is used to generate an audit report and a review draft based on the audit results of each document element;

[0178] The review module is used to obtain the review instruction; based on the review instruction, obtain the modified document; based on the modified document, calculate the similarity between the original document and the modified document; perform the preliminary judgment: compare the similarity with the limit value; if the similarity is higher than the limit value, re-acquire the modified document; if the similarity is lower than the limit value, analyze and review the modified document again;

[0179] A modification degree sorting module is used to calculate the recommended modification degree of each document element based on the review result of each document element; sort and generate a modification order based on the recommended modification degree of each document element; display the modification order on the document to be reviewed on the human-computer interaction page based on the modification order; obtain a preliminary modification text, which is the modified text of the document to be reviewed in the front column of the modification order; adjust the document to be reviewed in the back column of the modification order based on the preliminary modification text and adjust the modification order;

[0180] The sampling module is used to obtain the sampling instruction; based on the sampling instruction, obtain the document to be reviewed for which the review draft has been generated, and hide its review draft; based on the document to be reviewed for which the review draft has been generated, obtain the manual review draft input by the sampling personnel; based on the review draft of the document to be reviewed and the manual review draft, calculate and generate similarity; if the similarity exceeds the set value, perform the review operation: obtain the review information, and generate a comparison draft based on the review information; if the similarity does not exceed the set value, determine that the sampling personnel and the system have passed the review, and generate a sampling draft;

[0181] A similarity calculation module is used to obtain the annotation information of the review draft and the annotation information of the manual review draft based on the review draft and the manual review draft of the document to be reviewed, and the annotation information includes the number of annotations and the position of the annotations; perform a first similarity judgment: compare the number of annotations of the review draft with the number of annotations of the manual review draft; if the number of annotations of the review draft is different from the number of annotations of the manual review draft, obtain the difference in the number of annotations of the review draft and the number of annotations of the manual review draft, and obtain the number of non-overlapping values ​​of the annotation positions of the review draft and the annotation positions of the manual review draft; based on the difference in number and the number of non-overlapping values, calculate and generate similarity; if the number of annotations of the review draft is the same as the number of annotations of the manual review draft, obtain the different number of annotation positions of the review draft and the annotation positions of the manual review draft; calculate and generate similarity based on the non-overlapping value of the annotation positions of the review draft and the annotation positions of the manual review draft;

[0182] The historical parameter area module is used to obtain the modification information of the corresponding document based on the comparison draft and the sampling draft; adjust the document information of the document to be reviewed based on the modification information and obtain the revised document; determine whether the document is the final draft of the uploaded document; if so, add a historical document reference area and enter the revised document into the historical parameter reference area.

[0183] The implementation principle of a standard intelligent audit and compliance checking system is as follows: the system can obtain the document to be audited through the data input module, and then divide the document to be audited into multiple document elements through the element extraction module, and then analyze the document to be audited through the intelligent analysis module, and finally generate an audit report and a review draft through the result output module, thereby improving the efficiency and accuracy of document auditing.

[0184] The present application provides a smart terminal.

[0185] An intelligent terminal includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the above-mentioned standard intelligent audit and compliance checking method.

[0186] The present application provides a computer-readable storage medium,

[0187] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the above-mentioned standard intelligent audit and compliance checking method.

[0188] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A standard intelligent audit and compliance inspection method, characterized by: include: Get documents for review; Based on the documents to be reviewed and preset classification rules, the documents to be reviewed are divided into multiple document elements; Analyze the documents to be reviewed based on multiple document elements and the preset review direction corresponding to each document element; Generate audit reports and review drafts based on the audit results of each document element. The method further comprises: Obtaining review orders; Obtain revised documents based on review instructions; Based on the modified document, calculate the similarity between the original document and the modified document; Execution of the preliminary judgment: Compare the similarity with the limit value; If the similarity is higher than the limit, the modified document is retrieved again; If the similarity is lower than the limit, the modified document will be analyzed and reviewed again. After the steps of generating an audit report based on the audit results of each document element and generating a review draft, the following steps are also included: Calculate the recommended revision degree of each document element based on the review results of each document element; Sorting and generating a modification order based on the suggested modification degree of each document element; Display the modification order on the document to be reviewed on the human-computer interaction page based on the modification order; Obtaining a preliminary modified text, wherein the preliminary modified text is a modified text of the to-be-reviewed document at the top of the modification order; Adjust the revision order of the documents to be reviewed based on the preliminary revision text and adjust the revision order. The steps of generating an audit report based on the audit results of each document element and generating a review draft include: Get spot check instructions; Based on the spot check instruction, obtain the documents to be reviewed that have generated the review draft, and hide the review draft; Obtain the manual review draft input by the spot checker based on the document to be reviewed for which the review draft has been generated; Based on the review draft and the manual review draft of the document to be reviewed, similarity is calculated and generated; If the similarity exceeds the set value, a review operation is performed: obtaining review information and generating a comparison draft based on the review information; If the similarity does not exceed the set value, the sampling personnel and the system review are considered qualified, and a sampling draft is generated. The step of calculating and generating similarity based on the review draft and the manual review draft of the document to be reviewed also includes: Based on the review draft and the manual review draft of the document to be reviewed, obtaining annotation information of the review draft and the manual review draft, the annotation information including the number of annotations and the position of the annotations; Perform the first similarity determination: compare the number of comments in the review draft with the number of comments in the manual review draft; If the number of comments in the review draft is different from the number of comments in the manual review draft, obtain the difference between the number of comments in the review draft and the number of comments in the manual review draft, and obtain the number of non-overlapping positions of the comments in the review draft and the positions of the comments in the manual review draft; Based on the quantity difference and the number of non-overlapping values, the similarity is calculated and generated; If the number of comments in the review draft is the same as the number of comments in the manual review draft, obtain the number of differences between the comment positions in the review draft and the comment positions in the manual review draft; The similarity is calculated based on the number of non-overlapping annotation positions in the review draft and the manually reviewed draft.

2. A standard intelligent audit and compliance inspection method according to claim 1, characterized in that: The step of analyzing the document to be reviewed based on multiple document elements and the preset review direction corresponding to each document element includes: The preset review direction corresponding to each document element includes a specific review direction and a general review direction, wherein the general review direction includes a typo detection direction, a content missing detection direction, and a duplicate review direction; Based on the document elements, a corresponding specific review direction model is obtained, wherein the specific review direction model uses a pre-trained Chinese model, and the model is pre-trained in a standard field; The specific review direction model automatically reviews document elements based on pre-trained domain knowledge; Get general review direction based on document elements; Perform text preprocessing on document elements to obtain preprocessed text and a pre-trained typo detection model; Use a pre-trained typo detection model to identify common typos in pre-processed text; Correct common typos in preprocessed text and generate preliminary processed text; Obtaining a contrastive vocabulary base based on preliminary processed texts; Compare the words in the document to be tested with the words in the standard dictionary, detect and mark the wrong words; Modify the incorrect words in the preliminary processed text and generate the corrected text; Compare the document generated by merging multiple error correction texts with the preset template document; If they are different, a re-upload control will be displayed on the human-computer interaction page; If they are the same, generate the text to be deduplicated; Obtain document features and a literature retrieval library based on the text to be deduplicated; Retrieve identical or similar documents from a document retrieval database based on document features of the text to be deduplicated; Analyze the similarity between the text to be deduplicated and similar documents.

3. A standard intelligent audit and compliance inspection method according to claim 1, characterized in that: The method further comprises: Obtain modification information of corresponding documents based on comparison draft and sampling draft; Adjusting the document information of the to-be-reviewed document based on the modification information and obtaining a revised document; Determine whether the document is the final version of the uploaded document; If it exists, add a historical document reference area and enter the revised document into the historical parameter reference area.

4. A standard intelligent audit and compliance inspection system, characterized by: include, Data entry module, used to obtain documents to be reviewed; An element extraction module is used to divide the document to be reviewed into multiple document elements based on the document to be reviewed and preset classification rules; An intelligent analysis module for analyzing the document to be reviewed based on multiple document elements and the preset review direction corresponding to each document element; The result output module is used to generate an audit report and a review draft based on the audit results of each document element; A review module, used to obtain review instructions; Obtain revised documents based on review instructions; Based on the modified document, calculate the similarity between the original document and the modified document; perform preliminary judgment: compare the similarity with the limit value; If the similarity is higher than the limit, the modified document is retrieved again; if the similarity is lower than the limit, the modified document is analyzed and reviewed again; A modification degree sorting module is used to calculate the recommended modification degree of each document element based on the review result of each document element; sort the recommended modification degree of each document element and generate a modification order; Display the modification order on the document to be reviewed on the human-computer interaction page based on the modification order; Obtaining a preliminary modified text, wherein the preliminary modified text is a modified text of the to-be-reviewed document at the top of the modification order; Adjust the revision order of the documents to be reviewed based on the preliminary revision text and adjust the revision order; The sampling module is used to obtain the sampling instruction; based on the sampling instruction, obtain the document to be reviewed for which the review draft has been generated, and hide its review draft; based on the document to be reviewed for which the review draft has been generated, obtain the manual review draft input by the sampling personnel; based on the review draft of the document to be reviewed and the manual review draft, calculate and generate similarity; if the similarity exceeds the set value, perform the review operation: obtain the review information, and generate a comparison draft based on the review information; If the similarity does not exceed the set value, the sampling personnel and the system review are considered qualified, and a sampling draft is generated; A similarity calculation module is used to obtain the annotation information of the review draft and the annotation information of the manual review draft based on the review draft and the manual review draft of the document to be reviewed, the annotation information including the number of annotations and the position of the annotations; perform a first similarity determination: compare the number of annotations of the review draft with the number of annotations of the manual review draft; if the number of annotations of the review draft is different from the number of annotations of the manual review draft, obtain the difference between the number of annotations of the review draft and the number of annotations of the manual review draft, and obtain the number of non-overlapping values ​​of the annotation positions of the review draft and the annotation positions of the manual review draft; Based on the quantity difference and the number of non-overlapping values, the similarity is calculated and generated; If the number of annotations in the review draft is the same as the number of annotations in the manual review draft, obtain the number of differences between the annotation positions in the review draft and the manual review draft; calculate and generate the similarity based on the number of non-overlapping values ​​of the annotation positions in the review draft and the manual review draft.

5. An intelligent terminal, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program which can be loaded by the processor and executes the standard intelligent audit and compliance checking method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: The computer program is stored which can be loaded by a processor and execute the standard intelligent audit and compliance checking method according to any one of claims 1 to 3.

Images