Custom-supporting official document rule examination method

Abstract

The invention relates to the related technical field of text auditing, in particular to an official document rule auditing method supporting custom. The method comprises the following steps: acquiring a rule text input by a user; performing intention recognition and problem property judgment on the rule text, and generating a user-defined rule based on content meeting preset intention requirements and problem properties in the rule text; obtaining to-be-examined contents; reviewing the to-be-reviewed content based on the custom rule to obtain a first reviewing result; reviewing the to-be-reviewed content based on a preset text rule to obtain a second reviewing result; reviewing the to-be-reviewed content based on a preset knowledge rule to obtain a third reviewing result; and summarizing the first review result, the second review result and the third review result to obtain a target review result.

Description

Technical Field

[0001] The present application relates to the technical field related to text review, and specifically to a method for reviewing official documents that supports customized rules. Background Art

[0002] In today's wave of digital office work, electronic documents have become an indispensable part of the daily operations of government departments, enterprises, and institutions at all levels. The widespread use of electronic documents has greatly improved the efficiency of document processing. However, this has also brought with it higher requirements for document standardization and accuracy, and the importance of the document review process has also become more prominent. Traditional manual document review methods, due to their inefficiency and susceptibility to human error and numerous errors and omissions, are gradually failing to meet the needs of the times.

[0003] Existing document review methods primarily rely on predefined rule libraries to automate review. These rule libraries often include regular expressions, keyword libraries, and grammatical rule libraries. They perform pattern matching and semantic analysis on documents to complete the review task. While this approach has improved document review efficiency to a certain extent, it still has many drawbacks.

[0004] Its flexibility is severely limited. Predefined rule bases struggle to cover all possible review scenarios, often leaving it helpless when faced with the personalized review needs of different units, departments, and document types. For example, some units may have unique writing practices or internal regulations, but the existing system is unable to identify and review based on these specific rules. This results in inaccurate or missing information, significantly impacting the quality and reliability of document review. Summary of the Invention

[0005] In view of this, the embodiments of the present application are dedicated to providing a method for reviewing official documents that supports customization of rules, so as to conduct text review in a more comprehensive, effective and flexible manner.

[0006] This application provides a document review method that supports customization, including: Get the rule text entered by the user; Performing intent recognition and problem nature judgment on the rule text, and generating custom rules based on the content in the rule text that meets the preset intent requirements and problem nature; Obtain content to be reviewed; Reviewing the content to be reviewed based on the custom rules to obtain a first review result; Reviewing the content to be reviewed based on preset writing rules to obtain a second review result; Reviewing the content to be reviewed based on preset knowledge rules to obtain a third review result; The first review results, the second review results, and the third review results are summarized to obtain the target review results.

[0007] In some embodiments, the line performs intent recognition and question nature judgment on the rule text, and generates a custom rule based on the content in the rule text that meets the preset intent requirements and question nature, including: performing intent recognition on the rule text; If the intention of the rule text is related to the format and content of the official document to be reviewed, then the nature of the problem with the rule text will be determined; If the nature of the question in the rule text is a judgment question, a custom rule is generated based on the rule text.

[0008] In some embodiments, reviewing the content to be reviewed based on the custom rule to obtain a first review result includes: The content to be reviewed is input into the large language model. The large language model analyzes the content to be reviewed, obtains a judgment result based on the custom rule, and obtains a first judgment result.

[0009] In some embodiments, it further includes: Get the selection instruction; Based on the selection instruction, a writing rule is selected as a preset writing rule.

[0010] In some embodiments, it further includes: Obtaining knowledge rule selection instructions; An instruction is selected based on the knowledge rule, and the knowledge rule is selected as a preset writing rule.

[0011] In some embodiments, it further includes: Based on a large language model, writing rules corresponding to the type of the content to be reviewed are matched as preset writing rules, and knowledge rules corresponding to the type corresponding to the content to be reviewed are matched as preset knowledge rules.

[0012] In some embodiments, it further includes: Collect a preset number of samples of different types of official documents; Analyze official document samples through a large language model to extract format, structure, tone, and word features from the documents; Based on the extracted features and combined with expert experience and knowledge, corresponding writing rules are constructed for different types of documents.

[0013] In some embodiments, the review of the content to be reviewed based on preset writing rules to obtain a second review result includes: Input the content to be reviewed into the large language model, Based on the large language model, determine whether the content to be reviewed complies with the preset writing rules, and output the review result to obtain a second review result.

[0014] In some embodiments, it further includes: Collect a preset number of samples of official documents from different fields; By analyzing official document samples through a large language model, knowledge rules corresponding to different fields and general knowledge rules are extracted.

[0015] In some embodiments, the review of the content to be reviewed based on the preset knowledge rules to obtain the third review result includes: Input the content to be reviewed into the large language model; The large language model analyzes each word in the content to be reviewed based on its internal word knowledge and semantic understanding capabilities. It also verifies the terms in official documents using knowledge rule technology to determine whether there are any incorrect terms in the content to be reviewed. If there is an incorrect entry, an indication of the incorrect content and the corresponding correct expression will be output as the third review result.

[0016] The present application provides a method for reviewing official document rules that supports customization. First, the method obtains a rule text input by a user; performs intent recognition and problem nature judgment on the rule text, and generates a customized rule based on the content in the rule text that meets the preset intent requirements and problem nature; obtains the content to be reviewed; reviews the content to be reviewed based on the customized rule to obtain a first review result; reviews the content to be reviewed based on the preset writing rules to obtain a second review result; reviews the content to be reviewed based on the preset knowledge rules to obtain a third review result; and summarizes the first review result, the second review result, and the third review result to obtain a target review result. In this way, by obtaining the rule text input by the user and performing intent recognition and problem nature judgment to generate customized rules, it can meet the personalized review needs of different users and different business scenarios, avoiding the limitation of traditional fixed rule bases that are difficult to adapt to diverse needs. The customized rules, preset writing rules, and knowledge rules are comprehensively used to review official documents respectively, and the review results are finally summarized, making the review more comprehensive and detailed, effectively discovering various problems in the official documents, and improving the accuracy of the review. Users can directly participate in the formulation of rules and input rule text based on their own experience and actual needs. This enhances users' sense of control over the review process and their trust in the review results, which is conducive to promoting positive interaction between users and the review system and promoting continuous improvement in review quality. Allowing users to customize rules reduces the reliance on professional technicians to frequently modify code-level rules, reduces the manpower and time costs of system maintenance, and can also respond to rule changes more promptly, maintaining the timeliness and effectiveness of the review system. Customized rules are combined with preset writing rules and knowledge rules to give full play to their respective advantages and achieve complementary advantages. It not only takes into account personalized needs, but also ensures the standardization and scientific nature of the review, providing a more reliable solution for official document review. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 This is a flowchart of a method for reviewing official documents that supports customized rules, provided in one embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] This application provides a document review method that supports customizable rules, aiming to improve the flexibility, accuracy, and user engagement of document review, reduce maintenance costs, and integrate multiple advantages to improve the overall quality and efficiency of document review. The specific points are as follows: By acquiring user-entered rule text, identifying its intent and determining the nature of the problem, we generate custom rules based on content that meets pre-defined requirements. This process enables personalized review rules to meet the specific needs of different users and business scenarios, overcoming the limitations of traditional fixed rule bases.

[0021] Official documents are reviewed sequentially based on custom rules, preset writing rules, and knowledge rules, yielding first, second, and third review results. This multi-dimensional review approach ensures a comprehensive and detailed review, identifying potential issues in documents from different perspectives and effectively improving review accuracy.

[0022] By aggregating the review results at three different levels, we arrive at a comprehensive target review result. This step integrates multi-source information, fully leveraging the strengths of custom rules, text rules, and knowledge rules, forming a review mechanism that complements each other and provides users with professional review conclusions.

[0023] By allowing users to customize review rules based on their needs and experience, the system can quickly adapt to different review scenarios and changing requirements. This not only increases the system's flexibility, but also enhances its ability to adapt to diverse document types and complex review requirements.

[0024] Users deeply participate in the rule-making process, which enhances their trust and satisfaction with the review system. This interactivity promotes positive interaction between users and the system. Reducing reliance on frequent code modifications by professional technicians, users can manage rules themselves, reducing the manpower and time costs of system maintenance while ensuring the timeliness and effectiveness of the review system.

[0025] The solution of this application demonstrates significant advantages in meeting personalized review needs, improving review quality, enhancing user participation, and reducing maintenance costs. It provides an innovative, efficient, and reliable solution for the field of official document review, and has important application value and broad applicability prospects.

[0026] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart of a method for reviewing official documents that supports user-defined rules provided by an embodiment of the present application. Figure 1 As shown, the method includes the following contents.

[0028] Step S110, obtaining the rule text input by the user; First, the system receives user-entered rule text through a user-friendly interface or dedicated input port. The rule text details the user-defined review criteria, for example, specifying whether the last sentence of a document conforms to a specific reporting format. To ensure the validity of the rule, the system performs a preliminary format and grammar check upon receipt, annotating and recording key information in the rule text for subsequent processing.

[0029] Step S120: performing intent recognition and problem nature judgment on the rule text, and generating a custom rule based on the content in the rule text that meets the preset intent requirements and problem nature; The system uses advanced natural language processing technology to conduct in-depth analysis of rule texts. During the intent recognition stage, the system accurately determines the user's intention in entering the rule text to ensure that it is relevant to the official document review, such as determining whether the rule is specific to the format or content of the official document. During the problem nature judgment stage, the system further analyzes the nature of the rule text to ensure that it is a judgmental problem that can be clearly judged, such as whether the sentence is complete and whether the wording is appropriate. The system only considers the rule text content that meets the preset intent and nature requirements as the valid part, and through a specific rule generation algorithm, converts it into a custom rule that the system can recognize and execute, providing a personalized basis for subsequent official document review.

[0030] Step S130, obtaining the content to be reviewed; The system flexibly acquires documents for review from a variety of sources, including but not limited to local file systems, network transmissions, and integrated interfaces with other office systems. Upon acquisition, the system registers and pre-processes the documents, recording basic information such as document type, source, and date. It also performs preliminary formatting checks and organizes the documents to ensure their integrity and usability, laying the foundation for subsequent review steps.

[0031] Step S140: review the content to be reviewed based on the custom rule to obtain a first review result; The system precisely matches the content of documents under review against user-generated custom rules. A large language model deeply analyzes the document content and, in conjunction with the custom rules, conducts a detailed review. For example, if a custom rule requires a specific word to appear in a specific paragraph of a document, the model will accurately locate and verify the presence of that word. After the review is complete, the system summarizes all issues and judgments discovered based on the custom rules and generates a detailed first review result. This report covers key information such as the nature, location, and severity of the issue, providing users with more detailed review details.

[0032] Step S150, reviewing the content to be reviewed based on preset writing rules to obtain a second review result; Based on preset writing rules, the system conducts a regulatory review of official documents. These rules cover standard requirements for format, structure, tone, and wording. Leveraging the powerful analytical capabilities of a large language model, the system conducts in-depth analysis of document content. The model compares the document's actual content with the preset writing rule template to determine whether it meets requirements, such as consistency in font, size, paragraph format, and completeness of the document structure. It then generates a detailed secondary review, providing crucial support for ensuring the document's format is standardized and its structure is rigorous.

[0033] Step S160, reviewing the content to be reviewed based on preset knowledge rules to obtain a third review result; The system reviews the accuracy and reliability of official documents based on pre-set knowledge rules. These knowledge rules include specialized knowledge, factual information, and common knowledge within the document's domain. The large language model leverages its internal vocabulary knowledge and semantic understanding capabilities to meticulously analyze every word and sentence in the document. The model identifies and corrects typos and grammatical errors, verifies the accuracy of quotations, and ensures that the document's content meets the requirements of the relevant domain. After the review, the system generates comprehensive third-party review results, documenting any knowledge errors found and their suggested corrections, providing strong support for improving the quality of document content.

[0034] Step S170: Summarize the first review result, the second review result, and the third review result to obtain a target review result.

[0035] The system integrates the results of the first, second, and third reviews, performing a comprehensive analysis of these multi-source results through an intelligent fusion algorithm. This fusion algorithm weights the review results based on the characteristics and importance of each review dimension. The system generates comprehensive, targeted review results, including a detailed list of issues related to custom rules, writing standards, and knowledge accuracy, along with severity assessments and suggestions for improvement. This provides the target user with a comprehensive and accurate review report, providing clear guidance for subsequent document revisions and improvements, ensuring high quality and compliance.

[0036] Specifically, the line performs intent recognition and problem nature judgment on the rule text, and generates a custom rule based on the content in the rule text that meets the preset intent requirements and problem nature, including: The intention of the rule text is identified; if the intention of the rule text is related to the official document format and content of the content to be reviewed, the nature of the problem in the rule text is judged; if the nature of the problem in the rule text is a judgmental problem, a custom rule is generated based on the rule text.

[0037] After receiving the rule text entered by the user, the system first analyzes it using intent recognition methods from natural language processing technology. The purpose of intent recognition is to determine whether the rule text entered by the user is relevant to official document review, that is, whether the intention is to define a specific document review rule. For example, the rule text may be "Require official document titles to use size 2 small-font Song font." The system needs to recognize that this rule is a review requirement regarding the format of official documents. During implementation, the system trains the intent recognition model using a large amount of labeled training data. This data covers various common intent types of official document review rules, such as format specifications and content requirements. When new rule text is entered, the model analyzes the text's keywords, semantic structure, and other features, matching it with known intent categories to determine the intent category of the rule text and output corresponding intent labels, such as "format specification intent" and "content integrity intent."

[0038] After completing intent recognition, the system needs to further determine whether the identified rule text intent is relevant to the official document format and content to be reviewed. This is because there may be certain deviations or misunderstandings in the user's input, or the rule text may belong to other areas not related to official document review. For example, the rule text may be "requires that the attached pictures have high resolution." Although this may be related to the document, it is not a typical requirement for official document review. The system establishes a semantic library of intents related to official document review, which contains all common intent categories related to official document format and content review and their corresponding semantic features. The identified rule text intent is compared with the entries in the semantic library to calculate their semantic similarity. If the similarity exceeds the preset threshold, it is determined that the rule text intent is related to official document review and subsequent processing can continue; otherwise, the system will prompt the user that the rule text may not meet the requirements of official document review or requires further clarification and modification.

[0039] Once the intent of the rule text is determined to be relevant to official document review, the next step is to determine the nature of the question in the rule text. This step aims to determine whether the content expressed in the rule text is a judgmental question that can be clearly determined. Judgmental questions are those that have clear standards of right and wrong, or can be determined based on specific conditions and logic. For example, "Does the official document use standard honorific titles?" is a judgmental question because it can reach a clear conclusion by examining whether the specific terms of address conform to the standard. The system determines the nature of the rule text by analyzing its grammatical structure, keywords, and semantic logic. If the rule text contains clear conditional statements such as "whether," "whether it conforms to," or "exists," or its semantics point to a specific check or verification with clear standards, it is considered a judgmental question. In addition, the system can utilize defined judgmental question templates or pattern matching rules to compare the rule text with these templates to quickly and accurately determine its nature.

[0040] Once the rule text passes intent recognition and problem nature assessment, the system generates a custom rule based on it. This process first involves further parsing and normalizing the rule text to extract key elements, such as the review object, review conditions, and expected outcomes. For example, for the rule text "Require all data references in official documents to cite the source," the system parses it as follows: the review object is "data references in official documents," the review condition is "referenced data," and the expected outcome is "sources must be cited." The system then reorganizes and encodes these key elements according to a predefined rule template or format, forming a structured custom rule. This custom rule will contain clear syntax and semantics, enabling it to be recognized and executed by the system's review engine. During the custom rule generation process, the system also verifies its validity to ensure that the rule logic is clear and unambiguous, and that it complies with the basic standards and requirements for official document review. If potential logical errors or incompleteness are detected in the rule, the system will automatically correct it or prompt the user to modify it to ensure the quality and usability of the custom rule.

[0041] In some embodiments, the review of the content to be reviewed based on the custom rules to obtain a first review result includes: inputting the content to be reviewed into a large language model, and the large language model analyzes the content to be reviewed to obtain a judgment result for the custom rules, thereby obtaining a first judgment result.

[0042] Specifically, when reviewing official documents, the large language model is primarily used for analysis and judgment. Specifically, the document content is input into the large language model, which then conducts a comprehensive analysis based on its internal algorithms and logic. During this analysis, the model focuses on content related to custom rules, examining whether the document complies with these requirements. Ultimately, the model outputs a judgment result, the primary judgment result, indicating the document's review of the custom rules. This result may include information such as whether there are any non-compliances, the specific content of the non-compliance, and the severity of the non-compliance, providing a reference for subsequent review and revision.

[0043] Furthermore, the solution provided by the present application further includes: obtaining a selection instruction; and selecting a writing rule as a preset writing rule based on the selection instruction; and obtaining a knowledge rule selection instruction; and selecting a knowledge rule as a preset writing rule based on the knowledge rule selection instruction.

[0044] The system provides a user interface or API, allowing users to directly participate in the rule selection process. Users can manually select specific writing rules or knowledge rules based on their actual needs and review priorities. Users browse the rule library within the system interface, select a rule that meets their needs, and initiate the selection process by clicking the "Select" button or similar action. Upon receiving the selection instruction, the system verifies user permissions to ensure that the user is eligible to select the rule. Once verification is successful, the system confirms the selection and prepares to apply the selected rule to the review process.

[0045] In some embodiments, based on a large language model, writing rules corresponding to the type of the content to be reviewed are matched as preset writing rules, and knowledge rules corresponding to the type corresponding to the content to be reviewed are matched as preset knowledge rules.

[0046] In this step, the system uses the analysis capabilities of the Large Language Model (LLM) to automatically match and apply text rules and knowledge rules. The following is a detailed implementation process: Matching text rules based on large language models Receiving content awaiting review: The system receives the content of official documents awaiting review uploaded by the user.

[0047] Analyzing document types: The system inputs the document content into a large language model. The model identifies the document type (such as government documents, internal corporate reports, academic papers, etc.) based on its training data and semantic analysis capabilities.

[0048] Matching writing rules: The system searches the preset writing rule library for writing rules that match the identified document type.

[0049] Confirm the selected operation: The system confirms the selected operation and uses the matched writing rule as the preset writing rule.

[0050] Apply to the review process: The system is ready to apply the selected writing rules to the current document review process and begin to review the format, structure, tone, wording, etc. of the document.

[0051] Matching knowledge rules based on large language models Analyze the content of official documents: The system inputs the content of official documents into the large language model again, and the model further analyzes key information such as the subject and field of the document.

[0052] Matching knowledge rules: The system searches for knowledge rules that match the subject and field of the document in the preset knowledge rule library.

[0053] Confirm the selection operation: The system confirms the selection operation and uses the matched knowledge rules as the preset knowledge rules.

[0054] Apply to the review process: The system is ready to apply the selected knowledge rules to the current document review process, reviewing the professional knowledge, factual information, etc. in the document to ensure the accuracy and reliability of the content.

[0055] Through the above steps, the system can automatically and intelligently match appropriate writing rules and knowledge rules for different types of official documents without the need for manual selection by users, greatly improving the efficiency and accuracy of review, while also reducing manual intervention and realizing the automation and intelligence of the review process.

[0056] In some embodiments, it specifically includes: collecting a preset number of different types of official document samples; analyzing the official document samples through a large language model to extract the format, structure, tone, and word usage features in the official documents; based on the extracted features, combined with expert experience and knowledge, constructing corresponding writing rules for different types of documents.

[0057] In some embodiments, the process of constructing a writing rule specifically includes the following steps: Collect a preset number of samples of different types of official documents; The system collects a large number of different document samples from multiple channels, including but not limited to official documents issued by government agencies, enterprises, institutions, and academic institutions. These document samples cover a wide range of common document types, such as notices, reports, requests for instructions, replies, meeting minutes, and academic papers, ensuring diversity and representativeness. The number of samples collected meets the preset requirements to fully cover the characteristics of documents in different fields and types.

[0058] The system analyzes document samples using a large language model. The system inputs the collected document samples into the large language model, leveraging its powerful natural language processing capabilities to conduct in-depth analysis of the document samples. The large language model can automatically identify and extract multiple features from documents, including: Format features: such as font, font size, line spacing, page margins, heading level, list format, etc.

[0059] Structural features: such as the organization of the beginning, main body, and ending of the document, paragraph distribution, chapter division, etc.

[0060] Tone characteristics: such as formality, politeness, intensity of tone, etc.

[0061] Vocabulary characteristics: such as the frequency of use of professional terms, the selection of common words, word collocation habits, etc.

[0062] Through the analysis of large language models, the system can efficiently extract features from a large number of official document samples and form structured feature data.

[0063] Based on the extracted features and expert knowledge, corresponding writing rules are constructed for different document types. The system combines the extracted feature data with the expert knowledge, and the experts review and adjust the feature data. Based on their in-depth understanding of official document standards and practical experience, the experts screen and classify the extracted features, determining which features are key and representative and can reflect the standard writing requirements of different types of official documents.

[0064] For each document type, the system incorporates expert experience and knowledge to develop specific writing rules. For example, for government documents, titles must use a specific font and size, the structure of the body must meet specific hierarchical requirements, and the wording must be consistent with the context of a formal document. These rules are organized into structured rule sets and stored in the system's rule library for subsequent document review.

[0065] Through the above steps, the system can construct a set of comprehensive, accurate, and practical writing rules, providing a solid foundation for document review. This not only improves the efficiency and accuracy of the review, but also ensures the standardization and professionalism of the documents.

[0066] In some embodiments, the review of the content to be reviewed based on preset writing rules to obtain a second review result includes: Input the content to be reviewed into the large language model, Based on the large language model, determine whether the content to be reviewed complies with the preset writing rules, and output the review result to obtain a second review result.

[0067] In some embodiments, the process of reviewing the content to be reviewed based on preset writing rules specifically includes the following steps: Input the content to be reviewed into the large language model Receiving content awaiting review: The system receives official documents uploaded by users for review. These contents can be documents of various types and formats, such as official documents from government agencies, internal reports from companies, papers from academic institutions, etc.

[0068] Preprocessing: Perform necessary preprocessing operations on the official document content, including but not limited to text cleaning (removing extra spaces and special characters) and formatting (such as unified encoding format and line break specifications), to ensure that the text input into the large language model is of high quality and standardized, thereby improving the accuracy and efficiency of model analysis.

[0069] Input to the Big Language Model: The pre-processed document content is passed to the Big Language Model as input. The Big Language Model serves as the core engine for text analysis and understanding, leveraging its powerful natural language processing capabilities and experience learning from large amounts of text data to provide foundational support for subsequent review of text rules.

[0070] Determine whether the content to be reviewed complies with the preset writing rules based on the large language model and output the review results Analyzing the content to be reviewed: The large language model conducts a comprehensive and in-depth analysis of the content of the official document under review. It first extracts the document's formatting features, such as font size, line spacing, margins, heading levels, and list format. It then analyzes structural features, including the organization of the document's opening, body, and conclusion, paragraph distribution, and chapter divisions. It also focuses on tone features, such as formality, politeness, and intensity, as well as vocabulary features, such as the frequency of use of professional terminology, the selection of common vocabulary, and word collocation habits.

[0071] Comparing with preset writing rules: The system compares the document features extracted by the large language model with preset writing rules. These rules cover aspects such as document formatting (e.g., government documents require titles to use size 2 Song font, and the body to use size 3 Fang Song font), structural standards (e.g., reports should include basic elements such as a title, body, and signature, with the body clearly structured and logically coherent), and tone (e.g., formal documents should use a solemn and rigorous tone, avoiding colloquial or casual expressions). The large language model leverages its inherent semantic understanding and logical reasoning capabilities to determine whether the document's content matches these preset rules.

[0072] Generate audit results: Based on the comparison, the large language model generates detailed audit results. For each writing rule, it clearly indicates whether the content of the official document meets the requirements. If non-compliance with the rules is found, the location of the specific problem, the form of manifestation, and the degree of non-compliance are accurately recorded. For example, "In the third paragraph of the official document, the non-standard colloquial word 'we' was found to be used, which does not meet the tone standard requirements of formal documents." Finally, the system summarizes all these issues to form a complete second review result, and outputs it for users to review and subsequent processing, providing clear guidance for the modification and improvement of official documents, ensuring that the official documents comply with the standard in terms of writing, thereby improving the overall quality and professionalism of the official documents.

[0073] In some embodiments, it also includes: collecting a preset number of official document samples from different fields; analyzing the official document samples through a large language model to extract knowledge rules corresponding to different fields and general knowledge rules.

[0074] The system collects a large number of official document samples from multiple channels across various fields, including government agencies, enterprises, institutions, academic institutions, and the legal profession. The number of samples collected meets the preset requirements to ensure that the characteristics and knowledge rules of official documents in different fields are covered.

[0075] Analyze official document samples using a large language model The system feeds collected official document samples from various fields into a large language model, leveraging its natural language processing capabilities to conduct in-depth analysis of the document samples. The model extracts the language structure, semantic information, and domain-specific knowledge features from the documents.

[0076] Extract knowledge rules corresponding to different fields and general knowledge rules The system combines the extracted features with expert knowledge, which is then reviewed and adjusted by experts. Based on their understanding of official document standards and practical experience, experts screen and categorize features, identify key features, and construct corresponding knowledge rules. This mechanism provides comprehensive and accurate knowledge rules for document review, improving efficiency and accuracy and ensuring the professionalism and reliability of document content.

[0077] In some embodiments, the review of the content to be reviewed based on the preset knowledge rules to obtain the third review result includes: The content to be reviewed is input into the large language model; the large language model analyzes each word in the content to be reviewed based on its internal word knowledge and semantic understanding ability, and verifies the entries in the official document in combination with knowledge rule technology to determine whether there are erroneous entries in the content to be reviewed; if there are erroneous entries, the error content indication and the corresponding correct expression are output as the third review result.

[0078] In some embodiments, the process of reviewing the content to be reviewed based on preset knowledge rules is as follows: Input the content to be reviewed into the large language model Receiving content awaiting review: The system receives official documents uploaded by users awaiting review, which may cover official documents of various types and fields.

[0079] Preprocessing: Perform necessary preprocessing on the content to be reviewed, including text cleaning, formatting, and other operations, to improve the accuracy and efficiency of subsequent analysis.

[0080] Input into the large language model: Input the preprocessed document content into the large language model to provide basic data for the model's analysis.

[0081] Analyze the content to be reviewed and verify the terms Word analysis and semantic understanding: The large language model uses its internal word knowledge and semantic understanding capabilities to carefully analyze each word in the content to be reviewed, understanding the meaning, usage, and semantic role of each word in the context.

[0082] Verification technology combined with knowledge rules: The model verifies terms in official documents based on pre-set knowledge rules. This includes checking vocabulary for professionalism, accuracy, typos, and adherence to field-specific conventions. For example, in legal documents, the model verifies the use of correct legal terminology; in academic papers, the model checks the correct citation of relevant concepts and theories.

[0083] Error Term Identification: Through the aforementioned verification process, the large language model determines whether there are any error terms in the content to be reviewed. Error terms may include typos, inappropriate word usage, and incorrect use of professional terminology.

[0084] Output error content indication and correct expression Generate error content indication: For each incorrect term found, the large language model generates a detailed error content indication, clearly indicating the location of the incorrect term in the document, the specific word, and the type of error (such as typos, inappropriate word usage, etc.).

[0085] Providing correct expressions: Based on its rich knowledge base and semantic understanding capabilities, the model provides a corresponding correct expression for each incorrect word. This may be one or more correct word replacements or a more accurate expression.

[0086] Generating a third-level review result: All incorrect content indications and their corresponding correct expressions are summarized to form a complete third-level review result. This result is presented to the user in an easy-to-understand format, helping them quickly identify the problem and make corrections.

[0087] Through the above steps, the system can effectively utilize the knowledge and capabilities of the large language model to conduct in-depth knowledge rule review of official document content to ensure the accuracy, professionalism and standardization of the document content.

[0088] After completing the multi-dimensional review of the official document, the system enters the summary stage to generate comprehensive target review results. The detailed process is as follows: The system collects the results of the first review (based on custom rules), the second review (based on text rules), and the third review (based on knowledge rules) from different modules. These results are stored in a standardized format for subsequent processing.

[0089] The collected review results are cross-checked to remove duplicate problem records. For example, if multiple review dimensions report the same typo, the system will consolidate them into a single issue to avoid duplication and ensure clear results.

[0090] Consolidate issues found across different review dimensions into a unified issue list. Each issue record includes the issue type (e.g., formatting error, knowledge error), location information (page number, paragraph, etc.), severity (e.g., minor, moderate, severe), and related suggestions (e.g., formatting correction, terminology correction).

[0091] The consolidated list of issues is compiled into a comprehensive target review report. This report uses tables, charts, and other visual formats to display the distribution and types of issues, summarize and assess the overall quality and compliance of the document, and provide improvement recommendations based on the review results to guide users in revising and improving the document content.

[0092] Through the above steps, the target review results generated by the system provide users with practical and reliable feedback on the quality of official documents, helping users to improve the professionalism and standardization of official documents.

[0093] In addition to the above method, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method for supporting customized official document rule review according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0094] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0095] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enables the processor to execute the steps of the method for reviewing official documents supporting customization according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0096] The computer-readable storage medium may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0097] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A document review method supporting customized rules, characterized in that: include: Get the rule text entered by the user; Performing intent recognition and problem nature judgment on the rule text, and generating custom rules based on the content in the rule text that meets the preset intent requirements and problem nature; Obtain content to be reviewed; Reviewing the content to be reviewed based on the custom rules to obtain a first review result; Reviewing the content to be reviewed based on preset writing rules to obtain a second review result; Reviewing the content to be reviewed based on preset knowledge rules to obtain a third review result; The first review results, the second review results, and the third review results are summarized to obtain the target review results.

2. The method for reviewing official documents supporting user-defined rules according to claim 1, characterized in that: The line performs intent recognition and problem nature judgment on the rule text, and generates a custom rule based on the content in the rule text that meets the preset intent requirements and problem nature, including: performing intent recognition on the rule text; If the intention of the rule text is related to the format and content of the official document to be reviewed, then the nature of the problem with the rule text will be determined; If the nature of the question in the rule text is a judgment question, a custom rule is generated based on the rule text.

3. The method for reviewing official documents supporting user-defined rules according to claim 2, characterized in that: The step of reviewing the content to be reviewed based on the custom rule to obtain a first review result includes: The content to be reviewed is input into the large language model. The large language model analyzes the content to be reviewed, obtains a judgment result based on the custom rule, and obtains a first judgment result.

4. The method for reviewing official documents supporting user-defined rules according to claim 1, characterized in that: Also includes: Get the selection instruction; Based on the selection instruction, a writing rule is selected as a preset writing rule.

5. The method for reviewing official documents supporting user-defined rules according to claim 1, characterized in that: Also includes: Obtaining knowledge rule selection instructions; An instruction is selected based on the knowledge rule, and the knowledge rule is selected as a preset writing rule.

6. The document rule review method supporting customization according to claim 1, characterized in that: Also includes: Based on a large language model, writing rules corresponding to the type of the content to be reviewed are matched as preset writing rules, and knowledge rules corresponding to the type corresponding to the content to be reviewed are matched as preset knowledge rules.

7. The method for reviewing official document rules supporting customization according to claim 1, characterized in that: Also includes: Collect a preset number of samples of different types of official documents; Analyze official document samples through a large language model to extract format, structure, tone, and word features from the documents; Based on the extracted features and combined with expert experience and knowledge, corresponding writing rules are constructed for different types of documents.

8. The document review method supporting user-defined rules according to claim 1, characterized in that: The content to be reviewed is reviewed based on the preset writing rules to obtain a second review result, including: Input the content to be reviewed into the large language model, Based on the large language model, determine whether the content to be reviewed complies with the preset writing rules, and output the review result to obtain a second review result.

9. The method for reviewing official document rules supporting customization according to claim 1, characterized in that: Also includes: Collect a preset number of samples of official documents from different fields; By analyzing official document samples through a large language model, knowledge rules corresponding to different fields and general knowledge rules are extracted.

10. The document review method supporting user-defined rules according to claim 1, characterized in that: The content to be reviewed is reviewed based on the preset knowledge rules to obtain a third review result, including: Input the content to be reviewed into the large language model; The large language model analyzes each word in the content to be reviewed based on its internal word knowledge and semantic understanding capabilities. It also verifies the terms in official documents using knowledge rule technology to determine whether there are any incorrect terms in the content to be reviewed. If there is an incorrect entry, an indication of the incorrect content and the corresponding correct expression will be output as the third review result.

Images