Large model-based legal text generation proofreading method and system

Abstract

This application provides a method and system for generating and proofreading legal text based on a large-scale model, relating to the field of computer information processing technology. The method first acquires the input case text and matches the corresponding target legal document template based on a pre-set legal document template library. Next, according to format requirements, a pre-trained large-scale language model is invoked to generate an initial version of the legal document. Then, based on a pre-stored legal knowledge graph, the consistency of the internal logic and legal basis of the initial version of the legal document is checked, yielding a consistency check result. Finally, based on the consistency check result, conflicts and errors in the initial version of the legal document are identified to generate a proofreading report containing targeted modification suggestions. The technical solution provided by this application can not only automatically generate legal documents with standardized formats but also intelligently detect logical and legal citation errors and provide specific modification suggestions, thereby improving document quality and generation efficiency.

Description

Technical Field

[0001] This application relates to the field of computer information processing technology, and in particular to a method and system for generating and proofreading legal text based on a large model. Background Technology

[0002] With the application of artificial intelligence technology in the field of legal technology, there is a demand for technology that can assist non-professionals or legal practitioners in quickly and accurately drafting normative legal documents. Such technology needs not only to automatically convert the case facts described by users into legally rigorous text, but also to ensure the correctness of the generated text in terms of legal logic and basis, so as to reduce the risk caused by improper expression or incorrect legal citation.

[0003] Currently, the technical solution adopted is an automatic generation method based on natural language processing and predefined templates. This method first extracts entities and actions from the case description input by the user through keyword matching or simple rules, and then fills these elements into the corresponding positions of pre-set fixed format templates corresponding to different types of legal documents, thereby assembling and generating a preliminary legal document.

[0004] However, this method still has obvious limitations. First, its understanding of users' natural language descriptions is limited to surface keyword matching, making it difficult to deeply understand the semantic relationships between complex case logic and legal elements, which may cause the filled content to deviate from the user's original intention or not conform to the legal context. Second, it lacks the ability to automatically verify the consistency of legal logic within the generated document and the validity of the cited legal provisions, so the legal rigor of the generated text cannot be guaranteed. Summary of the Invention

[0005] This application provides a legal text generation and proofreading method and system based on a large model to solve the problems of insufficient accuracy of automatically generated legal texts and lack of automated legal logic verification capabilities in the prior art.

[0006] Firstly, this application provides a legal text generation and proofreading method based on a large model, including:

[0007] The system obtains the case text described in natural language input by the user, and matches the corresponding target legal document template to the case text based on a preset legal document template library.

[0008] According to the format requirements of the target legal document template, a pre-trained large language model is invoked to identify and standardize the legal elements of the case text in order to generate an initial version of the legal document.

[0009] Based on the pre-stored legal knowledge graph, the consistency between the internal logic and legal basis of the initial version of the legal document is verified, and the consistency verification result is obtained.

[0010] Based on the consistency verification results, conflicts and errors in the initial version of the legal document are identified to generate a proofreading report containing targeted modification suggestions.

[0011] Optionally, the user inputs a natural language description of the case, and based on a preset legal document template library, matches the case text with a corresponding target legal document template, including:

[0012] The case text was analyzed to identify key legal features contained therein, including the legal identities of the parties involved, the facts of the conduct that caused the dispute, and the legal claims of the users.

[0013] Based on the key legal features, a matching instruction is constructed, which is used to search in the legal document template library;

[0014] Using the matching instruction, the legal document template library is traversed and queried. During the traversal and query process, the information contained in the matching instruction is compared with the legal tag set associated with each target legal document template to determine the degree of correlation between the information contained in the matching instruction and the legal tag set.

[0015] Based on the degree of relevance, the target legal document template is determined from the legal document template library.

[0016] Optionally, based on the format requirements of the target legal document template, a pre-trained large language model is invoked to identify and standardize the legal elements of the case text to generate a preliminary version of the legal document, including:

[0017] The target legal document template is analyzed to obtain multiple fixed chapters and the categories of legal elements required to be included in each fixed chapter;

[0018] Based on the multiple fixed chapters and the legal element categories, an element extraction instruction is generated, which instructs the large language model to locate and output legal information of the target category from the text.

[0019] The case text and the element extraction instructions are combined into a complete input and provided to the large language model for processing.

[0020] The large language model responds to the element extraction command, traverses the case text, finds the original text expression corresponding to the legal element category, and outputs it as an element content block.

[0021] For each element content block, the wording was adjusted, transforming the colloquial original expressions into standardized expressions that conform to the formal language style of legal documents;

[0022] According to the arrangement order of the multiple fixed chapters in the target legal document template, each standardized expression is filled into the corresponding fixed chapter in sequence to form the initial version of the legal document.

[0023] Optionally, based on the multiple fixed chapters and the legal element categories, an element extraction instruction is generated, including:

[0024] Read the chapter name of each fixed section of the target legal document template;

[0025] Extract the textual descriptions of the legal element categories corresponding to each fixed chapter;

[0026] Based on the textual descriptions of the chapter titles and the legal element categories, a question statement is constructed for each fixed chapter, and the question statement directly inquires about the specific content in the case text related to the legal element category;

[0027] Multiple question statements constructed for all fixed chapters are connected according to the arrangement order of the fixed chapters in the target legal document template to form a question sequence;

[0028] A general task description is added before the question sequence, which instructs the large language model to extract answers from the case text sequentially according to subsequent questions;

[0029] The general task description and the question sequence are combined to form the element extraction instruction.

[0030] Optionally, based on a pre-stored legal knowledge graph, the initial version of the legal document is subjected to a consistency check between its internal logic and legal basis to obtain a consistency check result, including:

[0031] Extract the legal entities mentioned, the relationships between the legal entities, and the legal article numbers cited from the initial draft of the legal document;

[0032] For each extracted legal subject, search the pre-stored legal knowledge graph to see if there is a corresponding record point. When a corresponding record point exists, obtain the set of permitted relationships and attribute information associated with the record point.

[0033] The association description between the legal subjects in the initial legal document is compared one by one with the set of allowed relationships corresponding to the record points in the legal knowledge graph to determine whether the association description is included in the set of allowed relationships, and the comparison result is obtained.

[0034] The status marker corresponding to the legal provision number is queried in the legal knowledge graph to determine whether the status marker is in a valid state, and the status judgment result is obtained.

[0035] The comparison results and the state judgment results are summarized to generate a consistency verification result.

[0036] Optionally, based on the consistency verification results, conflicts and errors in the initial version of the legal document are identified to generate a proofreading report containing targeted modification suggestions, including:

[0037] The consistency verification results are analyzed to identify items in the comparison results whose association descriptions are not included in the corresponding allowed relationship set, and legal article numbers whose status is marked as invalid in the status judgment results, as items to be corrected;

[0038] Based on the content of the item to be corrected, the target text position of the item to be corrected in the initial version of the legal document is located in reverse;

[0039] For items to be corrected that are not included in the corresponding set of allowed relationships, select the target relationship that best fits the logic of the current case from the set of allowed relationships associated with the record points corresponding to the legal subjects in the legal knowledge graph, and use it as the correct relationship description.

[0040] For legal provision numbers marked as invalid, the alternative legal provision number corresponding to the legal provision number is queried from the legal knowledge graph;

[0041] For each item to be corrected, a modification suggestion record is generated. Each modification suggestion record includes the target text location, the original text of the item to be corrected, and the corresponding suggested replacement content. The suggested replacement content is the correct association description or the alternative legal provision number.

[0042] Arrange all the suggested modifications in the order of their location in the target text of the initial draft legal document, and attach the full text of the initial draft legal document to generate a proofreading report.

[0043] Optionally, a modification suggestion record is generated for each item to be modified, including:

[0044] Read an item to be corrected, and obtain the target text position corresponding to the item to be corrected, the original text content of the item to be corrected, and the suggested replacement content;

[0045] The target text location is used as the core index for the modification suggestion record;

[0046] The original text of the item to be corrected is used as the problem description section of the modification suggestion record;

[0047] The suggested replacement content will be used as part of the solution for modifying the suggestion record;

[0048] Combine the problem description section and the solution section into a complete modification suggestion;

[0049] Each complete modification suggestion is assigned a unique identifier, and the unique identifier, the core index, the problem description section, and the solution section are recorded in a structured format according to the order of the target text positions to form a modification suggestion record.

[0050] Secondly, this application provides a legal text generation and proofreading system based on a large model, including:

[0051] The acquisition module is used to acquire the case text described in natural language input by the user, and match the corresponding target legal document template to the case text based on a preset legal document template library;

[0052] The transcription module is used to call a pre-trained large language model to identify and standardize the legal elements of the case text according to the format requirements of the target legal document template, so as to generate an initial version of the legal document.

[0053] The verification module is used to verify the consistency between the internal logic and legal basis of the initial version of the legal document based on the pre-stored legal knowledge graph, and obtain the consistency verification result.

[0054] The generation module is used to locate conflicts and errors in the initial version of the legal document based on the consistency verification results, so as to generate a proofreading report containing targeted modification suggestions.

[0055] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a legal text generation and proofreading method based on a large model as described in the first aspect above.

[0056] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a legal text generation and proofreading method based on a large model as described in the first aspect.

[0057] This application utilizes a large language model to identify and standardize the legal elements of case texts, enabling a deep understanding of the complex case logic and legal context within natural language descriptions. This results in the generation of standardized initial legal documents that more closely reflect the user's true intentions, improving the accuracy and semantic fit of document generation. Furthermore, based on a pre-stored legal knowledge graph, the application performs consistency checks on the internal logic and legal basis of the documents, achieving automated review of the compliance of legal entity relationships and the validity of legal provisions, thus overcoming the shortcomings of existing solutions that lack in-depth verification capabilities.

[0058] Furthermore, based on the consistency verification results, conflicts and errors are located and a proofreading report containing targeted modification suggestions is generated. By structurally recording the location of the target text, the original content, and the suggested replacement content, clear and locatable modification guidance can be provided to users. This structured report generation method enables legal logic or basis issues to be directly and clearly pointed out and given a correction plan, improving the practicality of the proofreading results and the efficiency of users' subsequent modifications.

[0059] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart of a legal text generation and proofreading method based on a large model provided in this application is shown;

[0062] Figure 2 This paper presents a schematic diagram of the structure of a legal text generation and proofreading system based on a large model, as provided in this application.

[0063] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0064] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0065] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0066] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0067] Figure 1 A flowchart is provided for a legal text generation and proofreading method based on a large model, as shown in this application. Figure 1 As shown, the method includes:

[0068] Step 101: Obtain the case text described in natural language input by the user, and match the corresponding target legal document template to the case text based on the preset legal document template library.

[0069] Optionally, step 101 may specifically include:

[0070] Step 1011: Analyze the case text to identify the key legal features contained in the case text. The key legal features include the legal identities of the parties involved, the facts of the conduct that caused the dispute, and the legal claims of the users.

[0071] Step 1012: Based on the key legal features, construct a matching instruction, which is used to search in the legal document template library.

[0072] Step 1013: Using the matching instruction, traverse and query the legal document template library. During the traversal and query process, compare the information contained in the matching instruction with the legal tag set associated with each target legal document template to determine the degree of correlation between the information contained in the matching instruction and the legal tag set.

[0073] Step 1014: Based on the degree of relevance, determine the target legal document template from the legal document template library.

[0074] In this step, the case text refers to a narrative text about a specific legal event or dispute entered by the user in everyday spoken or written language. It describes the basic facts of the case and is obtained by directly receiving the text entered by the user in the interactive interface.

[0075] A pre-built legal document template library refers to a collection of files that are pre-built and stored in the system, containing legal document examples in various standard formats to provide a writing framework for different cases.

[0076] The target legal document template refers to a specific legal document format file selected from the preset legal document template library that best matches the current user's case description, and is used to guide the standardized generation of subsequent document content.

[0077] Key legal features refer to the core elements extracted from the case description texts provided by users, which play a decisive role in selecting the correct type of legal document. They are used to characterize the legal nature of the case and are obtained through automated text analysis of the case description texts.

[0078] The legal identity of the participating parties is one of the key legal characteristics, describing the roles played by the various legal entities involved in the case, such as plaintiff, defendant, employee, employer, etc., and is used to distinguish the subjects of the legal relationship.

[0079] The facts of conduct that give rise to a dispute refer to one of the key legal characteristics, describing the core event or behavior that leads to the dispute between the parties, such as failure to deliver goods or personal injury, and are used to define the focus of the dispute in the case.

[0080] A user's legal claim is one of the key legal characteristics, describing the purpose the user hopes to achieve through legal proceedings, such as demanding compensation, terminating a contract, or confirming rights, and is used to clarify the objective of the document.

[0081] A matching command is a formatted query command automatically created by the system that contains the aforementioned key legal features. It is used to perform precise retrieval in the template library by combining the identified key legal features into a computer-understandable query statement.

[0082] A legal tag set refers to a pre-associated set of descriptive keywords or phrases with each template in a pre-defined legal document template library. These keywords or phrases summarize the case characteristics to which the template applies and are obtained by manually or automatically assigning attribute tags to each template.

[0083] In this step, the syntax parsing and semantic understanding module in natural language processing technology is first used to automatically analyze the case text submitted by the user, and scan the entire text to identify sentence structure and understand the meaning of words in legal context. It can automatically locate and separate the parts describing the characters, the parts describing the process of the event, and the parts expressing the user's purpose from the text, thereby extracting the key legal features of the three dimensions of the legal identity of the participants, the facts of the behavior that caused the dispute, and the user's legal claims.

[0084] Secondly, through an information integration and formatting module, multiple key legal features extracted are used as input. According to predefined rules, these scattered feature information are combined into a structured query condition, i.e. a matching instruction. For example, it concatenates or encodes feature words such as identity, facts, and claims into a computer-executable retrieval command.

[0085] Next, a semantic similarity calculation engine is used to take the generated matching command as the query input, traverse each template in the legal document template library, read the set of legal tags that are pre-associated with each template, and calculate the degree of similarity between the semantic information expressed by the matching command and the semantic information expressed by each set of legal tags. This process is not a simple keyword matching, but rather an evaluation of the degree of relevance between the two at the semantic level based on word vectors or deep learning models.

[0086] Finally, a sorting and selection logic module receives the relevance scores calculated for all templates, sorts all templates from highest to lowest according to their relevance scores, and automatically selects the template with the highest score. This selected template is then identified as the target legal document template that best matches the current case and is output for use in subsequent processes.

[0087] For example, User A inputs a case text into the system: "I worked at Company B for two years, but the company never signed a labor contract with me. I have now been dismissed and want the company to compensate me." First, the text is analyzed to identify the legal identities of the parties involved as the employee and the company, the facts of the dispute being the failure to sign a labor contract and the dismissal, and the user's legal claim being for compensation. Second, these features are combined into a matching instruction and searched in the legal document template library. Next, the legal document template library contains templates such as labor arbitration applications and civil complaints, each associated with a set of legal tags such as labor dispute, labor contract dispute, and wage compensation. Then, calculations show that the user's case features are closest to the tags of the labor arbitration application template in terms of labor dispute, labor contract, and compensation, i.e., the highest degree of relevance. Therefore, the labor arbitration application is identified as the target legal document template.

[0088] This step solves the problem of users choosing their own format by automatically analyzing the case text and intelligently matching the most suitable document template. It can quickly and accurately convert unstructured case descriptions into a standardized document format starting point, thereby improving the accuracy and efficiency of subsequent document generation.

[0089] Step 102: According to the format requirements of the target legal document template, call the pre-trained large language model to identify and standardize the legal elements of the case text in order to generate the initial version of the legal document.

[0090] Optionally, step 102 may specifically include:

[0091] Step 1021: Parse the target legal document template to obtain multiple fixed chapters and the categories of legal elements required to be included in each fixed chapter.

[0092] Step 1022: Based on the multiple fixed chapters and the legal element categories, generate an element extraction instruction, which instructs the large language model to locate and output legal information of the target category from the text.

[0093] Optionally, step 1022 may specifically include the following steps: reading the chapter name of each fixed chapter of the target legal document template; extracting the text description of the legal element category corresponding to each fixed chapter; constructing a question statement for each fixed chapter based on the chapter name and the text description of the legal element category, the question statement directly inquiring about the specific content related to the legal element category in the case text; connecting multiple question statements constructed for all fixed chapters according to the arrangement order of the fixed chapters in the target legal document template to form a question sequence; adding a general task description before the question sequence, the general task description being used to instruct the large language model to extract answers from the case text sequentially according to subsequent questions; combining the general task description with the question sequence to constitute an element extraction instruction.

[0094] Step 1023: Combine the case text and the element extraction instructions into a complete input and provide it to the large language model for processing.

[0095] Step 1024: Responding to the element extraction instruction through the large language model, traverse the case text, find the original text expression corresponding to the legal element category, and output it as an element content block.

[0096] Step 1025: Perform wording adjustments on each element content block, converting the colloquial original expressions into standardized expressions that conform to the formal language style of legal documents.

[0097] Step 1026: According to the arrangement order of the multiple fixed chapters in the target legal document template, fill in the corresponding fixed chapters with the standardized expressions in sequence to form the initial version of the legal document.

[0098] In this step, the pre-trained large language model refers to an artificial intelligence model that is trained on massive amounts of text data and is capable of understanding and generating natural language. In this method, it is specifically used to process legally related text information.

[0099] The initial draft of a legal document refers to a text draft that the system automatically generates for the first time based on the target template format and the user's case details. This draft has a basic legal document structure and content and serves as the basis for subsequent proofreading. It is obtained by filling in the corresponding chapters of the target legal document template with standardized expressions.

[0100] Multiple fixed chapters refer to several standardized components that constitute the template of the target legal document, such as party information, litigation claims, facts and reasons, etc., which are used to define the organizational structure and content sections of the legal document.

[0101] Legal element categories refer to the specific legal concept types to which the content to be filled in under each fixed chapter belongs, such as name, gender, and address under the party information chapter. These are used to guide the large language model in extracting specific information from the case text.

[0102] An element extraction instruction is a text instruction automatically generated by the system to guide a large language model to perform a specific information extraction task. It is used to explicitly tell the large language model what specific information needs to be found in the case text.

[0103] The target category of legal information refers to the specific information category that needs to be found from the case text as specified by the element extraction instruction. It corresponds one-to-one with the legal element category, such as the plaintiff's name, the time of the dispute, etc. It is obtained by parsing and executing the element extraction instruction through the large language model.

[0104] The chapter names of fixed chapters refer to the titles or names of multiple fixed chapters, such as plaintiff information and claims, which are used to identify and distinguish different chapters. They are obtained by reading the format marks or title text of the target legal document template.

[0105] The text description refers to a brief textual explanation of each legal element category, used to more clearly define the information content to be extracted. For example, the category of litigation claims can be described as the specific demands made by the plaintiff to the court, which can be obtained by parsing the preset annotations in the template or by inferring from the category name.

[0106] A question statement refers to a specific and direct question constructed for a fixed chapter and its legal element categories. It is used to guide the large language model to locate information in the form of questions, such as "What is the name of the plaintiff in this case?". It is obtained by converting the textual descriptions of the chapter name and legal element categories into question sentences.

[0107] The question sequence refers to an ordered list formed by arranging the question statements generated for all chapters according to the order in which the fixed chapters appear in the template. This is used to systematically guide the large language model to complete the information extraction of the entire text.

[0108] The overarching task description is a summary text placed before the question sequence, used to clarify the overall task objective to the large language model. For example, "Please answer the following questions in sequence based on the following case description," which is obtained by the system automatically generating standard prompts based on the task type.

[0109] The element content block refers to the original answer text fragment that the large language model finds and directly outputs from the case text for each question in the question sequence. For example, for the question "What is the plaintiff's name?", the output might be "Zhang San". This is obtained by reading and understanding the case text and answering the question through the large language model.

[0110] Standardized expression refers to the textual expression that conforms to the formal written language requirements of legal documents after rewriting colloquial and non-standard elements and content blocks. It is used to enhance the professionalism and rigor of documents and is obtained by transforming the wording and format of the elements and content blocks.

[0111] In this step, a template structure parser is first used to analyze the target legal document template file and identify the various parts of the template defined by headings, paragraph marks or specific formats, thereby obtaining all the fixed chapters that constitute the document and the specific legal element categories that need to be filled in under each chapter.

[0112] Secondly, an instruction construction module is used to generate element extraction instructions by parsing multiple fixed chapters and the legal element categories of each chapter. First, the chapter name of each fixed chapter is read, and detailed text descriptions of each legal element category under each chapter are extracted or generated. Then, natural language generation technology is used to construct a direct and clear question statement for each chapter-element category pair, such as converting the party information chapter and name element category into "What is the name of the plaintiff in this case?". Then, the module arranges all the generated question statements into an ordered question sequence according to the original order of the chapters in the template. Finally, a predefined general task description text is attached to the beginning of the question sequence. The task description and the question sequence are concatenated to form a complete element extraction instruction.

[0113] Next, an input assembly module merges the user's original case text with the generated element extraction instruction text to form a complete prompt message, which is then submitted to a pre-trained large language model for processing. Upon receiving the combined input information, the large language model's internal natural language understanding and reasoning module begins to work. It first understands the requirements of the overarching task description, and then reads and understands each question in the question sequence in turn. For each question, it traverses and scans the case text, using its semantic association capabilities acquired during large-scale text training to locate the sentence or fragment most relevant to the question from the text, and directly extracts that part of the original text as the answer to the corresponding question, i.e., outputting an independent element content block.

[0114] Then, a text normalization processing module processes each element content block output by the large language model. Based on a built-in legal document formal terminology lexicon and grammar rule library, the words and sentence structures in the content blocks are transformed, such as converting colloquial "no signed contract" into formal "no signed written labor contract", thereby obtaining a normalized expression that meets the requirements of legal document style.

[0115] Finally, an automatic document assembly engine is used to accurately fill the corresponding positions in the template with the standardized text of each chapter, according to the original arrangement of multiple fixed chapters in the target legal document template. All parts are combined into a legal document draft with complete format and preliminary content, thus generating the first version of the legal document.

[0116] For example, following the specific implementation of the previous step, the target legal document template has been determined to be a labor arbitration application. The template is then parsed to obtain fixed sections such as applicant information, respondent information, arbitration request, facts and reasons, as well as legal element categories such as name, gender, company name, requested matters, and the course of events. Next, based on this information, an element extraction instruction is generated, including a sequence of questions such as "What is the applicant's name?", "What are the specific arbitration requests?", and "What are the facts of the labor dispute?", along with a general explanation. Then, the case text input by user A, along with this instruction, is input into the large language model. The model answers the questions sequentially, outputting element content blocks such as A, B companies, requests for compensation, and dismissal after two years of work without a contract. These content blocks are then standardized; for example, the request for compensation is specified as a request for the respondent to pay double the wage difference for the period without a signed labor contract. Finally, these standardized expressions are filled in according to the template order to generate a standard-formatted initial version of the labor arbitration application, containing specific applicant information, clear arbitration requests, and detailed facts and reasons.

[0117] This step generates structured instructions by parsing the target template, guiding the large language model to accurately extract information from the case details. The extracted colloquial content is then automatically converted into formal legal terminology and finally assembled into a document according to the template format. This achieves the automated generation of a first draft legal document that goes from fragmented and unprofessional case descriptions to a rigorously formatted and standardized version, thus improving the efficiency and quality of document drafting.

[0118] Step 103: Based on the pre-stored legal knowledge graph, perform consistency verification on the internal logic and legal basis of the initial version of the legal document to obtain the consistency verification result.

[0119] Optionally, step 103 may specifically include:

[0120] Step 1031: Extract the legal entities mentioned, the relationship descriptions between the legal entities, and the legal article numbers cited from the initial draft of the legal document.

[0121] Step 1032: For each extracted legal subject, search in the pre-stored legal knowledge graph to see if there is a corresponding record point. When a corresponding record point exists, obtain the set of permitted relationships and attribute information associated with the record point.

[0122] Step 1033: Compare the description of the relationship between the legal subjects in the initial version of the legal document with the set of allowed relationships corresponding to the record points in the legal knowledge graph one by one, and determine whether the description of the relationship is included in the set of allowed relationships to obtain the comparison result.

[0123] Step 1034: Query the status marker corresponding to the legal provision number in the legal knowledge graph, determine whether the status marker is in a valid state, and obtain the status judgment result.

[0124] Step 1035: Summarize the comparison results and the state judgment results to generate a consistency verification result.

[0125] In this step, the pre-stored legal knowledge graph refers to a data network of legal knowledge organized in a graphical structure that is pre-built and stored in the system. It contains definitions of legal concepts, entities, legal provisions and their interrelationships, and serves as an authoritative reference library for verifying the correctness of legal logic and basis.

[0126] Consistency verification results refer to the conclusive data generated by the system after reviewing the logic and basis of the initial version of the legal document. These data are used to clearly point out logical conflicts or legal citation issues in the document.

[0127] Legal entities refer to individuals or organizations mentioned in the initial draft of legal documents who enjoy legal rights or bear legal obligations, such as plaintiffs, defendants, companies, and individuals, and are used to identify the parties involved in a legal relationship.

[0128] The legal provision number refers to the clause number of the specific laws and regulations cited in the initial version of the legal document, such as Article 82 of the Labor Contract Law. It is used to locate the specific legal basis and is obtained by matching the specific legal provision citation format from the document text.

[0129] A record point is a node in a pre-stored legal knowledge graph that represents a specific legal entity or legal concept. It is used to store and associate detailed information about the entity and is obtained by searching for nodes in the knowledge graph that match the extracted legal entity name.

[0130] The allowed relation set refers to a predefined list of legally permissible relation types between a given record point in the knowledge graph. It is used to determine whether the description of the relationship between legal entities is legal and reasonable, and is obtained by querying the relation edges defined for that record point in the knowledge graph.

[0131] Attribute information refers to the data stored in the knowledge graph by a record point that describes the basic characteristics of the legal entity, such as the gender of a natural person or the type of a legal person. It is used to supplement the description of the legal entity and is obtained by reading the attribute fields associated with the record point.

[0132] A related description refers to a section of text in the initial draft of a legal document that describes the relationship between two or more legal entities. For example, Company A hired employee B. This section is used to express the core legal facts of the case and is obtained by extracting sentences containing multiple legal entities from the factual statement section of the document.

[0133] The comparison result refers to the judgment conclusion drawn after comparing the association description with the set of allowed relationships. It is used to identify whether the association description is legally allowed and is obtained by performing a semantic inclusion judgment.

[0134] Status markers refer to identifiers associated with a specific legal provision number in a pre-stored legal knowledge graph, used to indicate whether the provision currently has legal effect, such as valid, repealed, or amended. They are used to verify the timeliness of legal basis and are obtained by querying the attribute fields of the legal provision nodes in the knowledge graph.

[0135] The status judgment result refers to the conclusion drawn from the status flag regarding whether the legal provision number is usable. It is used to identify whether the reference is valid and is obtained by reading and judging the value of the status flag.

[0136] In this step, the generated initial version of the legal document is first scanned and analyzed using named entity recognition and rule matching technologies. The named entity recognition technology automatically locates and marks all names representing individuals and organizations in the text, identifying them as legal entities. At the same time, the rule matching technology extracts all legal article numbers from the text based on the unique citation format of legal provisions, such as book title marks plus article numbers. Secondly, a knowledge graph query engine uses each extracted legal entity name as a search keyword to search in a pre-stored legal knowledge graph, attempting to find knowledge nodes, i.e., record points, with names that are completely matched or highly similar. If a corresponding record point is found, the query engine will further obtain a list of all relationship edge types connected to that node as a set of allowed relationships, and read the descriptive fields stored in that node as attribute information.

[0137] Next, a semantic relationship verification module compares the obtained set of permissible relationships with the associated descriptions extracted from the initial legal documents. This is a semantic inclusion judgment, which analyzes the core semantic relationship expressed by the associated descriptions and then determines whether such semantic relationship exists in the legal relationship types listed in the set of permissible relationships. For example, it determines whether the employment relationship belongs to the set of permissible labor relationships between employers and employees. Each comparison will produce a conclusion, namely the comparison result.

[0138] Then, the knowledge graph query engine is used again to search for the corresponding legal provision node in the knowledge graph, using each extracted legal provision number as a query condition. A field on that node that specifically indicates its legal validity, namely the status flag, is read. Based on preset rules, such as whether the flag is valid or current, the status judgment result of whether the provision is usable is obtained. Finally, a result summary module collects and organizes all the comparison results and status judgment results generated in the above steps. These results are organized according to the corresponding legal subject, association description or legal provision number to form a structured list or report, clearly indicating which associations are legal and which are illegal, and which legal provisions are valid and which are invalid, thereby generating the final consistency verification result.

[0139] For example, following the specific implementation of the previous step, firstly, based on the generated initial version of the labor arbitration application, the legal entities applicant A and respondent company B are extracted from the document, along with the cited legal provision number, Article 82 of the Labor Contract Law. Simultaneously, the related description that company B did not sign a labor contract with company A is extracted. Secondly, the two record points of employee and employer are searched in the pre-stored legal knowledge graph, and the set of permissible relationships between employees and employers is obtained, which includes establishing a labor relationship and the occurrence of a labor dispute, but does not include the specific behavior of not signing a labor contract. Next, a comparison is performed, determining that the specific behavior of not signing a labor contract is a possible factual manifestation under the condition of a labor dispute relationship; therefore, this related description is logically permissible, and the comparison result is compliant. Simultaneously, the status marker of Article 82 of the Labor Contract Law is checked and found to be valid, and the status judgment result is valid. Finally, these results are summarized to generate a consistency verification result, indicating that no fundamental logical conflict or invalid basis issues were found between the factual description and legal citation in this initial version of the document during this verification.

[0140] This step utilizes a pre-stored legal knowledge graph to automatically verify the logical compliance of the relationships between legal entities and the timeliness of cited legal provisions in the initial draft of the legal document. It can identify deeper issues such as factual descriptions that violate common legal sense or citations of expired legal provisions, thereby effectively improving the logical rigor and reliability of the legal basis of the generated document and adding an automated layer of assurance for document quality.

[0141] Step 104: Based on the consistency verification results, locate the conflicts and errors in the initial version of the legal document to generate a proofreading report containing targeted modification suggestions.

[0142] Optionally, step 104 may specifically include:

[0143] Step 1041: Analyze the consistency verification result, identify items in the comparison result whose association description is not included in the corresponding allowed relationship set, and legal article numbers whose status is marked as invalid in the status judgment result, as items to be corrected.

[0144] Step 1042: Based on the content of the item to be corrected, locate the target text position where the item to be corrected appears in the initial version of the legal document.

[0145] Step 1043: For the item to be corrected that is not included in the corresponding set of allowed relationships, select the target relationship that best fits the logic of the current case from the set of allowed relationships associated with the record point corresponding to the legal subject in the legal knowledge graph, and use it as the correct relationship description.

[0146] Step 1044: For legal provision numbers whose status is marked as invalid, query the legal knowledge graph for the corresponding alternative legal provision number.

[0147] Step 1045: Generate a modification suggestion record for each item to be corrected. Each modification suggestion record includes the target text location, the original text of the item to be corrected, and the corresponding suggested replacement content. The suggested replacement content is the correct association description or the alternative legal provision number.

[0148] Optionally, step 1045 may specifically include the following steps: reading an item to be corrected, obtaining the target text position corresponding to the item to be corrected, the original text content of the item to be corrected, and the suggested replacement content; using the target text position as the core index of the modification suggestion record; using the original text content of the item to be corrected as the problem description part of the modification suggestion record; using the suggested replacement content as the solution part of the modification suggestion record; combining the problem description part and the solution part into a complete modification suggestion; assigning a unique identifier to each complete modification suggestion, and recording the unique identifier, the core index, the problem description part, and the solution part in a structured format according to the order of the target text positions to form a modification suggestion record.

[0149] Step 1046: Arrange all the modification suggestions in the order of their positions in the target text of the initial version of the legal document, and attach the full text of the initial version of the legal document to generate a proofreading report.

[0150] In this step, the proofreading report refers to a final document generated by the system that summarizes all the problems found and suggested revisions in the initial draft of the legal document. It is used to guide users in revising and improving the draft document.

[0151] Items to be corrected refer to matters selected from the consistency check results that represent a specific problem in the document, including logically disallowed related descriptions and expired legal provision numbers, which are used as the direct objects for generating modification suggestions.

[0152] The target text location refers to the specific start and end character numbers or paragraph line numbers of an item to be corrected in the initial version of the full legal document. It is used to accurately locate the problem in the report and is obtained by reverse querying the index of the document text based on the content of the item to be corrected.

[0153] The correct association description refers to the best relationship expression found from the legal knowledge graph that is legally permissible and consistent with the current case when a non-compliant association description needs to be corrected. This is used as a suggested solution for the user to modify the original text.

[0154] The modification suggestion record refers to a complete data entry created for an item to be corrected, which includes the problem location, the original text display, and the suggested modification plan. It is the basic unit that constitutes the proofreading report and is obtained by recording the target text location, the original text content, and the suggested replacement content in a structured manner.

[0155] The suggested replacement content refers to the specific text provided to users in the modification suggestion record to replace the incorrect parts of the original text. This includes the correct association description or the replacement of the legal provision number, which is used to directly guide users on how to make the modifications.

[0156] The core index refers to the key field within the modification suggestion record used to uniquely identify and quickly locate the suggestion. It is usually obtained by directly using the value of the target text position, which is used as the primary key field of the record.

[0157] The problem description section refers to the part of the modification suggestion record used to display and describe the original error content, that is, the original content of the item to be corrected, so that users can clearly see what the problem is. It is obtained by reading the original content field of the item to be corrected.

[0158] The solution section refers to the part of the modification suggestion record that provides the correct modification solution, i.e., the suggested replacement content, which tells the user how to correct the issue. It is obtained by reading the suggested replacement content field of the item to be corrected.

[0159] In this step, the generated consistency verification results are first parsed by a rule engine. The data in the verification results is read and, according to preset rules, all related description items that are not included in the comparison results and all legal article numbers that are invalid in the status judgment results are filtered out. These filtered problem items are uniformly marked as items to be corrected. Secondly, the position of each item to be corrected in the original text is determined by text position mapping technology. By using the text offset information recorded synchronously when extracting information, or by quickly matching the original text content of the item to be corrected with the full text of the initial legal document, the specific start and end positions of the content in the document, i.e., the target text position, are found and recorded.

[0160] Next, for different types of items to be corrected, the correction scheme query is executed in parallel. For items to be corrected whose association description is not included in the set of allowed relations, semantic similarity calculation and context association analysis technology are used. With the current association description and the overall case as reference, the degree of matching between each allowed relation and the current case context is calculated in the set of allowed relations corresponding to the legal subject record point in the knowledge graph. The relation that is most semantically appropriate and logically consistent is selected as the target association relation and transformed into a fluent sentence as the correct association description. For legal article numbers marked as invalid, the knowledge graph query engine is used again to search for articles with the same or similar legal effect and valid status in the graph, using the invalid article number as the key. The number of the invalid article number is used as the substitute legal article number.

[0161] Then, a structured data assembly module generates a modification suggestion record for each item to be corrected. It reads an item to be corrected and its corresponding target text position, original text content, and suggested replacement content obtained in the previous step. The target text position is used as the core index of the record. At the same time, the original text content is used as the problem description part, and the suggested replacement content is used as the solution part. These two parts of text are combined into a complete modification suggestion. Finally, a unique identifier is assigned to this suggestion, and these fields are packaged and recorded in a fixed table or JSON structured format according to the core index, i.e., the order of the target text position, to form a complete modification suggestion record.

[0162] Finally, a report generator sorts all the generated modification suggestions according to their core index values ​​from smallest to largest, ensuring that the order in which the issues are presented in the report is consistent with the order in which they appear in the original text. The report generator then merges this ordered list of suggestions with the complete text of the initial legal document, usually by marking the suggestions with serial numbers in the original text and detailing them at the end, or by using a columnar comparison method, to generate a final proofreading report that is easy for users to read and operate.

[0163] For example, following the specific implementation of the previous step, firstly, assume that the verification finds that another related description A has privately taken the company's customer list, and the comparison result is not included because there is no relationship of taking away trade secrets in the set of permitted relationships between employees and employers, and the status judgment result of a certain clause of the Contract Law cited is that it has been repealed; secondly, this related description and this legal article number are identified as items to be corrected; then, they are located in lines 5 and 8 of the fact and reason section of the document, respectively, and for the related description problem, from the set of permitted relationships between employees and employers in the knowledge graph, the violation of confidentiality obligations or infringement of trade secrets is selected as the target related relationship, and the correct related description A is suspected of infringing the company's trade secrets is generated; then, for the invalid legal article, the alternative legal article number is found in the graph as Article 9 of the Anti-Unfair Competition Law; then, modification suggestion records are generated for these two items to be corrected, and the records contain the line number position, the original text error content, and the suggested replacement content; finally, all records are sorted by line number, and the full text of the document is attached to generate a proofreading report, which clearly instructs the user on how to modify lines 5 and 8.

[0164] This step transforms the issues discovered during consistency checks into specific modification suggestions. By accurately locating the errors and providing feasible replacement solutions, it automatically generates a structured proofreading report. This allows users to clearly and quickly understand where the errors are in the document and how to correct them, achieving a complete closed loop from automatic generation and intelligent verification to assisted correction, thus improving the efficiency and final quality of document polishing.

[0165] Figure 2 This application provides a schematic diagram of the structure of a legal text generation and proofreading system based on a large model, such as... Figure 2 As shown, the system includes:

[0166] The acquisition module 21 is used to acquire the case text described in natural language input by the user, and match the corresponding target legal document template for the case text based on a preset legal document template library;

[0167] The transcription module 22 is used to call a pre-trained large language model to identify and standardize the legal elements of the case text according to the format requirements of the target legal document template, so as to generate a first version of the legal document.

[0168] The verification module 23 is used to perform consistency verification on the internal logic and legal basis of the initial version of the legal document based on the pre-stored legal knowledge graph, and obtain the consistency verification result.

[0169] The generation module 24 is used to locate the conflicts and errors in the initial version of the legal document based on the consistency verification results, so as to generate a proofreading report containing targeted modification suggestions.

[0170] Figure 2 The aforementioned legal text generation and proofreading system based on a large model can perform... Figure 1 The implementation principle and technical effects of the legal text generation and proofreading method based on a large model, as described in the illustrated embodiment, will not be repeated here. The specific methods by which each module and unit of the legal text generation and proofreading system based on a large model performs its operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0171] In one possible design, Figure 2 The legal text generation and proofreading system based on a large model, as shown in the embodiment, can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0172] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0173] The processing component 32 is used for the above Figure 1 The embodiment describes a method for generating and proofreading legal text based on a large model.

[0174] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0175] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0176] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0177] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0178] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0179] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0180] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The illustrated embodiment presents a legal text generation and proofreading method based on a large model.

[0181] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A large model-based legal text generation proofreading method, characterized in that, include: The system obtains the case text described in natural language input by the user, and matches the corresponding target legal document template to the case text based on a preset legal document template library. According to the format requirements of the target legal document template, a pre-trained large language model is invoked to identify and standardize the legal elements of the case text in order to generate an initial version of the legal document. Based on a pre-stored legal knowledge graph, the initial version of the legal document undergoes a consistency check of its internal logic and legal basis to obtain a consistency check result. This includes: extracting the mentioned legal entities, the descriptions of relationships between these legal entities, and the numbers of the cited legal provisions from the initial version of the legal document; for each extracted legal entity, searching the pre-stored legal knowledge graph for a corresponding record point; if a corresponding record point exists, obtaining the set of permissible relationships and attribute information associated with that record point; comparing the descriptions of relationships between the legal entities in the initial version of the legal document with the set of permissible relationships corresponding to the record point in the legal knowledge graph one by one to determine whether the descriptions of relationships are included in the set of permissible relationships, obtaining a comparison result; querying the legal knowledge graph for the status marker corresponding to the legal provision number, determining whether the status marker is in a valid state, obtaining a status judgment result; and summarizing the comparison result and the status judgment result to generate a consistency check result. Based on the consistency verification results, conflicts and errors in the initial version of the legal document are located to generate a proofreading report containing targeted modification suggestions. This includes: parsing the consistency verification results, identifying items whose association descriptions are not included in the corresponding allowed relationship set, and legal article numbers marked as invalid in the status judgment results, as items to be corrected; locating the target text position of the items to be corrected in the initial version of the legal document based on their content; for items whose association descriptions are not included in the corresponding allowed relationship set, selecting the most consistent one from the allowed relationship set associated with the record point corresponding to the legal subject in the legal knowledge graph. The target association relationship of the current case logic is used as the correct association description; for legal article numbers marked as invalid, the corresponding alternative legal article number is queried from the legal knowledge graph; a modification suggestion record is generated for each item to be corrected, and each modification suggestion record includes the target text position, the original content of the item to be corrected, and the corresponding suggested replacement content, where the suggested replacement content is either the correct association description or the alternative legal article number; all the modification suggestion records are arranged in the order of the target text positions in the initial version of the legal document, and the full text of the initial version of the legal document is attached to generate a proofreading report; The step of generating a modification suggestion record for each item to be modified includes: Read an item to be corrected, and obtain the target text position corresponding to the item to be corrected, the original text content of the item to be corrected, and the suggested replacement content; The target text location is used as the core index for the modification suggestion record; The original text of the item to be corrected is used as the problem description section of the modification suggestion record; The suggested replacement content will be used as part of the solution for modifying the suggestion record; Combine the problem description section and the solution section into a complete modification suggestion; Each complete modification suggestion is assigned a unique identifier, and the unique identifier, the core index, the problem description section, and the solution section are recorded in a structured format according to the order of the target text positions to form a modification suggestion record.

2. The method of claim 1, wherein, Obtain the case description text input by the user in natural language, and match the case text with the corresponding target legal document template based on a preset legal document template library, including: The case text was analyzed to identify key legal features contained therein, including the legal identities of the parties involved, the facts of the conduct that caused the dispute, and the legal claims of the users. Based on the key legal features, a matching instruction is constructed, which is used to search in the legal document template library; Using the matching instruction, the legal document template library is traversed and queried. During the traversal and query process, the information contained in the matching instruction is compared with the legal tag set associated with each target legal document template to determine the degree of correlation between the information contained in the matching instruction and the legal tag set. Based on the degree of relevance, the target legal document template is determined from the legal document template library.

3. The method according to claim 1, characterized in that, Based on the format requirements of the target legal document template, a pre-trained large language model is invoked to identify and standardize the legal elements of the case text, generating an initial version of the legal document, including: The target legal document template is analyzed to obtain multiple fixed chapters and the categories of legal elements required to be included in each fixed chapter; Based on the multiple fixed chapters and the legal element categories, an element extraction instruction is generated, which instructs the large language model to locate and output legal information of the target category from the text. The case text and the element extraction instructions are combined into a complete input and provided to the large language model for processing. The large language model responds to the element extraction command, traverses the case text, finds the original text expression corresponding to the legal element category, and outputs it as an element content block. For each element content block, the wording was adjusted, transforming the colloquial original expressions into standardized expressions that conform to the formal language style of legal documents; According to the arrangement order of the multiple fixed chapters in the target legal document template, each standardized expression is filled into the corresponding fixed chapter in sequence to form the initial version of the legal document.

4. The method according to claim 3, characterized in that, Based on the aforementioned fixed chapters and legal element categories, an element extraction instruction is generated, including: Read the chapter name of each fixed section of the target legal document template; Extract the textual descriptions of the legal element categories corresponding to each fixed chapter; Based on the textual descriptions of the chapter titles and the legal element categories, a question statement is constructed for each fixed chapter, and the question statement directly inquires about the specific content in the case text related to the legal element category; Multiple question statements constructed for all fixed chapters are connected according to the arrangement order of the fixed chapters in the target legal document template to form a question sequence; A general task description is added before the question sequence, which instructs the large language model to extract answers from the case text sequentially according to subsequent questions; The general task description and the question sequence are combined to form the element extraction instruction.

5. A legal text generation and proofreading system based on a large model, applied to the legal text generation and proofreading method based on a large model according to any one of claims 1-4, characterized in that, include: The acquisition module is used to acquire the case text described in natural language input by the user, and match the corresponding target legal document template to the case text based on a preset legal document template library; The transcription module is used to call a pre-trained large language model to identify and standardize the legal elements of the case text according to the format requirements of the target legal document template, so as to generate an initial version of the legal document. The verification module is used to verify the consistency between the internal logic and legal basis of the initial version of the legal document based on the pre-stored legal knowledge graph, and obtain the consistency verification result. The generation module is used to locate conflicts and errors in the initial version of the legal document based on the consistency verification results, so as to generate a proofreading report containing targeted modification suggestions.

6. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a legal text generation and proofreading method based on a large model as described in any one of claims 1 to 4.

7. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a legal text generation and proofreading method based on a large model as described in any one of claims 1 to 4.

Images