Intelligent brief report merging method and system based on large language model

Abstract

The invention discloses an intelligent brief report merging method and system based on a large language model. Aiming at the problems of content conflicts, data calculation missing and non-uniform formats in multi-source brief report merging, a man-machine cooperative processing architecture guided by a tree-shaped directory is provided. The method comprises the following steps: structured preprocessing: analyzing to-be-merged brief reports to generate a standard directory tree, and extracting main body unit tags in a character string matching manner; content grading fusion: text type chapter: adding a main body unit identifier to the same-level content of each brief report, and integrating the same-level content into a coherent paragraph by a large language model; according to the data type chapter, entity data is extracted through a large language model, calculation is executed through a Python script, and an analysis text is generated through the large language model; and automatic output: automatically arranging the head, the tail, the font and the page number of the official document by utilizing an automatic generation engine, and generating a downloadable document. Compared with a traditional manual merging mode, the method has the advantages that the efficiency is obviously improved by more than 50%; on the premise that entity extraction is correct, the data calculation accuracy is gt; 99%; and the format compliance of the output document reaches 100% and strictly meets the standard.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of natural language processing and intelligent document processing, in particular to an intelligent report merging method and system based on a large language model. BACKGROUND

[0002] In government work reporting, enterprise management, engineering project monitoring and other scenarios, it is often necessary to merge periodic reports (such as daily reports, weekly reports, and special reports) submitted by multiple departments or units into a comprehensive document. The traditional merging method mainly relies on manual operation, which has the following significant defects:

[0003] Low content integration efficiency: manual reading of dozens or even hundreds of reports, extracting key information and reorganizing language, is time-consuming and labor-intensive. For reports containing numerical data, manual data extraction, calculation and analysis are required, which may introduce calculation errors or interpretation biases; structural consistency is difficult to guarantee: the structure of reports submitted by different units varies greatly (such as inconsistent chapters and non-standard title naming), and manual alignment to a unified directory is required, resulting in information merging confusion or semantic fragmentation. Existing automated tools only support mechanical merging of fixed format documents and cannot understand semantic associations, let alone handle the implicit logic in unstructured text; lack of professionalism and standardization: manually written merged reports are easily influenced by subjective factors, with inconsistent expression styles, and difficulty in meeting industry terminology standards or official writing standards. Existing document generation tools (such as report software) lack domain knowledge and cannot automatically generate deep analysis conclusions containing data interpretation.

[0004] Existing technologies are mostly rule-based content extraction systems: through predefined regular expression matching of titles and numerical values, but the generalization ability is poor, making it difficult to adapt to the expression variants and nested structures of report texts; traditional NLP summary models: can generate single-document summaries, but cannot realize the structured alignment and numerical fusion calculation of multi-source information, and the generated text lacks official standardization; directly prompting a large language model to merge multiple reports may result in content omission, hallucination or neglect of data calculation requirements, and cannot solve the problem of output format standardization.

[0005] Therefore, there is an urgent need for an intelligent merging solution that integrates semantic understanding, structured calculation and format standardization to realize the full-process automation of "content deep integration-data analysis-official document generation". SUMMARY

[0006] The present application aims at the deficiencies of the prior art, and provides an intelligent briefing merging method and system based on a large language model. The method realizes intelligent generation of a directory of multi-source briefings, structured analysis of content, numerical calculation and semantic fusion through a large language model, and automatically outputs official documents conforming to administrative norms, and is suitable for government agencies, enterprise groups and other scenarios requiring high-frequency aggregation of briefings of multiple departments and branch organizations.

[0007] The technical solution for achieving the object of the present application is as follows: on the one hand, an intelligent briefing merging method based on a large language model is provided, which comprises the following steps:

[0008] Step 1: Directory structure generation: the semantic topics and hierarchical logic of multiple briefings are analyzed by using a large language model, an initial directory tree structure is generated, and the generated directory tree structure is compared and corrected with a preset rule library, and finally a standardized directory tree structure is generated;

[0009] Step 2: Content structured extraction: the original text of each briefing is parsed, the text segments corresponding to the directory node in each briefing are recognized and extracted according to the generated directory tree node information, and the subject unit identifier is extracted to establish a structured mapping relationship;

[0010] Step 3: Multi-source content merging: for each node in the directory tree, differential merging processing is performed according to the content type of the node;

[0011] Step 4: Official document automatic generation: the structured content after merging is automatically added with standard header and tail information, a unified document style is set, and page numbers are inserted, so that a final official document that can be directly used is generated.

[0012] Further, the correction in step 1 specifically includes: calculating the type or non-calculating attribute of the directory tree node, and converting the corrected directory into a tree-shaped data structure in memory, wherein the root node is the document topic, the child node is the chapter title, and the node object record includes the title text, the hierarchical depth, and the expected content type.

[0013] Further, step 2 specifically includes:

[0014] A predefined unit name regular pattern list is used for string matching to extract the subject unit from the briefing as a content identification tag;

[0015] According to the directory tree node information, an accurate string matching based on the directory node title text is applied, and a rule inference strategy is combined with the hierarchical depth of the title and the relative position information in the document to cut the text content, so as to establish a mapping relationship of "directory node ID-main unit identifier-text content", and store it in a structured JSON object.

[0016] Further, step 3 specifically includes:

[0017] For non-computational nodes, a unit identifier prefix is added to each source text using the mapping relationship to ensure that the subject unit identifier is always associated with the corresponding text segment, and then a large language model is used to generate a standard narrative text.

[0018] For computational nodes, the following are sequentially performed: structured entity extraction based on a predefined prompt word template, calling an independent script to perform numerical calculation, inputting the original extracted entity data and the script calculated result data into a large language model, and generating a comprehensive analysis and fusion text containing data interpretation by the large language model.

[0019] Further, for the processing of computational nodes, it further includes:

[0020] The large language model entity extraction forcibly uses a predefined and configurable prompt word template, and the extracted structured entity data needs to be associated with the subject unit identifier of its source;

[0021] The calculation logic of the script execution includes basic summation, average value, and also includes weighted average, standard deviation, cumulative value, completion rate, and index comparison complex operations.

[0022] Further, when the large language model generates a comprehensive analysis and fusion text containing data interpretation, it needs to specify the specifications of the model generated content and ensure that the interpretation reflects the relationship between the data and the source subject unit, so that the generated content meets the requirements of industry background knowledge and analysis dimensions, and ensures that the interpretation is based on the provided original data and calculation results.

[0023] Further, in step 4, the automatic generation of the document specifically includes:

[0024] By reading the pre-set document template file or configuration parameters, automatically insert the government standard document header;

[0025] Call the style setting interface of the adaptive text extraction tool, set the body, headings, line spacing, and page margins uniformly according to the document standards;

[0026] Use the document sectioning and footer editing function to add right-aligned page numbers in the footer and ensure continuous page numbering.

[0027] On the other hand, an intelligent briefing merging system for implementing the above-mentioned large language model-based intelligent briefing merging method is provided, and the system includes:

[0028] Catalog analysis module: used to implement calling a selected large language model to generate a catalog tree, and integrating a rule engine to post-process the model output.

[0029] Document parsing engine: integrated adaptive text extraction tool for realizing parsing of each briefing original text, identifying and extracting the text segment corresponding to the directory node in each briefing according to the generated directory tree node information, and extracting the subject unit identifier to establish a structured mapping relationship;

[0030] Content merging processor: as a core processing unit, used for performing the shunting processing of computing and non-computing content, scheduling large language models for content generation and entity extraction, integrating a preset computing script library and an execution engine, calling and managing the execution environment of the computing script, and implementing the conversion of structured data to analysis text;

[0031] Document formatting output module: used for generating a final document in accordance with administrative specifications based on a template engine, and providing a download interface.

[0032] On the other hand, a computer device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the large language model-based briefing merging method when executing the computer program.

[0033] On the other hand, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the large language model-based briefing merging method.

[0034] Compared with the prior art, the present application has the following advantages:

[0035] (1) The method automatically generates a standard directory structure and accurately extracts multi-source briefing content through the deep semantic understanding ability of the large language model, solving the problems of low efficiency and poor consistency of traditional manual merging;

[0036] (2) Combined with an intelligent differentiated processing mechanism (narrative fusion of non-computing nodes and data extraction-computation-analysis integrated process of computing nodes), the accuracy and logicality of multi-source information integration are significantly improved.

[0037] (3) Through the automatic generation engine of the document template, it is ensured that the output document strictly conforms to the standard specification, realizing end-to-end automation from content aggregation to format layout, greatly reducing the cost of manual intervention, and improving the efficiency and standardization level of government and enterprise report generation.

[0038] The present application will be described in further detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a flowchart of the large language model-based intelligent briefing merging method in one embodiment. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the present application clearer, further detailed description will be given below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and not to limit the present application.

[0041] The present application innovatively provides a technical logic for constructing an intelligent merging system for a briefing by using a large language model. Specifically, a standardized directory tree is first generated by combining templates with semantic analysis, and then each briefing content is cut into a structured JSON segment with a main unit identifier by combining a document parsing tool. The core merging stage is divided into two categories: for descriptive nodes (such as work progress), the text is input into the large language model after adding a prefix of the main unit to generate a standardized narrative; for computational nodes (such as data statistics), a three-step process of "entity extraction-Python script numerical calculation-large language model fusion analysis" is adopted to output a comprehensive text containing data interpretation combined with industry templates. Finally, the content is automatically assembled by Python-docx, and the format elements such as standardized page header and footer, Songti font and fixed line spacing are strictly inserted according to the national standard to generate a compliant official document that can be directly used.

[0042] In one embodiment, in combination Figure 1 The briefing merging method based on the large language model specifically includes:

[0043] S1, directory structure generation: receiving a plurality of briefing files to be merged, first drafting an initial directory, and then using a large language model with strong instruction following ability and strong text understanding and structuring ability to analyze the semantics, theme distribution and hierarchical relationship of the content of each briefing, and correcting the initial directory to generate a standardized directory tree structure. This process combines the semantic inference ability of the large language model and the preset rule templates. And use Python function (not limited to this function) to convert the directory into a tree data structure, where the root node is the document title and the child node is the chapter title.

[0044] S2, content structuring extraction: calling a document parsing tool to extract the original text content of each briefing. According to the directory tree structure generated in S1, the text content of each briefing is accurately divided into structured JSON data segments corresponding to each node of the directory tree by using a combination of string matching (accurately matching chapter title text) and rule-based position inference (such as title level and subsequent content starting identifier). In the JSON data, the main unit information (such as department name and company name) extracted from the first-level title (or identified by a specific rule) is specially marked as the attribution identifier of the briefing content segment. Finally, a mapping relationship of "directory node ID-main unit identifier-text content" is established.

[0045] S3, multi-source content merging, perform merging operation for each sub-node of the directory tree:

[0046] If the node is marked or identified as a non-computational node (such as "overview", "work progress", "existing problems", "next step plan" and other descriptive content), add the subject unit identification to which the node belongs before the corresponding text content segment in each briefing: as a source prefix. Input the text set with source identification into the large language model, combine with the preset narrative template (for example, "the situation of each unit is as follows: unit A: ; unit B: ") that conforms to the target document or report style, to generate coherent and standardized merged narrative text.

[0047] If the node is marked or identified as a computational node, perform the following "entity extraction-value calculation-fusion analysis" sub-steps to generate comprehensive analysis text containing data presentation, key finding interpretation and preliminary conclusion.

[0048] Here, the core innovation of this step is to use a three-level data processing chain of "entity extraction-independent calculation-result interpretation" to effectively avoid the inherent defects of large language models in numerical calculation:

[0049] In the output data structure of the entity extraction step, the subject unit identification field needs to be included. In the input of the fusion analysis step, include the original data and calculation results associated by subject units, so that the large language model can reflect the source attribution when generating analysis text (for example: "Among them, unit A has a completion rate of 80%, which is significantly higher than units B (65%) and C (55%)").

[0050] S4, automatic generation of documents: input all the contents organized by the directory tree after S3 merging into the Python-docx library. Automatically insert the standard document header (including file title, document number, secret level, urgency, issuing unit, issuing date, etc.) and document footer (such as issuing authority, carbon copy, issuing date, contact information) that conforms to the target agency's specifications. Strictly follow national standards or enterprise internal document specifications to set the overall style of the document: including using Fongsong_GB2312 font for the main text, using Fangzhengxiaobiaosong Simplified Chinese or Songti (specific font size bold), fixed line spacing (such as 28.9 pounds) and page margins (such as 3.7 cm on the top, 3.5 cm on the bottom, 2.8 cm on the left, and 2.6 cm on the right). Automatically add page numbers in the footer position, right-aligned, and in the format "Page X of Y". Finally, generate a merged document that is format-compliant and ready for use, and provide a download link or direct push through the API response.

[0051] Further, in one of the embodiments, S1 includes the selection of the large language model and the use of scripts to generate a directory expressed in a tree data structure, specifically including:

[0052] S11, based on the tasks that the large language model needs to undertake in the system, the large language model is selected. A large language model with text understanding and structuring capabilities is selected, and is implemented through API calling or local deployment.

[0053] S12, the present application adopts but is not limited to a Python script to generate a tree data structure table expression directory.

[0054] Further, in one of the embodiments, in S2, each briefing original text is parsed, according to the generated directory tree node information, the text segment corresponding to the directory node in each briefing is recognized and extracted, and the subject unit identifier is extracted, and a structured mapping relationship is established with the directory node ID as the key, containing the subject unit identifier and the corresponding text content segment. Specifically, the following steps are included:

[0055] S21) Title level recognition. A regular expression pattern is constructed using the title text of the directory tree node, the title position and content are matched by scanning the briefing text, and the level depth is recorded, and the original document level is restored.

[0056] S22) Main body unit identifier extraction. From the document header metadata area, a predefined list of unit name regular patterns is applied to sequentially match and extract the first valid result as the main body unit identifier.

[0057] S23) Construction of node ID-unit-content mapping. The node ID extracted in S12, the main body unit extracted in S22, and the text chapter content extracted in S21 are mapped.

[0058] Further, in one of the embodiments, S3 performs multi-source content merging, for each node in the directory tree, differential merging processing is performed according to its content type (non-computing type, computing type), specifically including the following steps:

[0059] S31) For non-computing type nodes, add unit identifier prefix to each source text, and use large language model to generate standard narrative text.

[0060] S32) For computing type nodes:

[0061] (a) Entity extraction: using a large language model to extract structured information from the text content corresponding to each node in the briefing, according to the pre-defined, domain-specific entity description, and associating the subject unit identifier of its source.

[0062] (b) Numerical calculation: input the structured entity data extracted in step (a) into the pre-set Python calculation script corresponding to the calculation requirement. The script performs specified numerical operation logic (such as sum, average, weighted average, etc.), and includes data validity check (such as null value, outlier detection).

[0063] (c) Fusion analysis: the original structured entity data extracted in step (a) (associated by subject unit), the result data calculated in step (b), and the required industry background knowledge and analysis dimension requirements are input into a large language model. The large language model combines these inputs to generate a comprehensive analysis text containing data presentation, key finding interpretation, and preliminary conclusion, and reflects the relationship between the data and the source subject unit.

[0064] Further, in one embodiment, in the automatic generation of official documents in S4, the merged structured content is automatically generated into a final official document that can be directly used according to the target official document specification. Specifically, the following steps are included:

[0065] S41) Structured content assembly: using a preset official document template, the data output from S3 is filled into the corresponding placeholder position in the template to automatically generate a document segment containing the main body of the text.

[0066] S42) Insert standard document header: a pre-set standard official document header template file (unit name, document number, issuer, classification, urgency, copy, date, etc. Fixed format document segment or style) is used to replace the placeholders for dynamic information (document number, issuing unit, issuing date, etc.) required for generating the official document header through user input.

[0067] S43) Add page number footer: using OxmlElement and qn libraries of docx, the required page number is automatically added to the center position at the bottom of each page after the document is generated.

[0068] S44) Generate final document and provide url download link: the complete document content (including header, body, and page number) processed through S41-S43 is exported as a standard word file, saved to the temporary storage area of the server, a time-sensitive unique access URL is generated, and the URL is returned to the user interface for download; At the same time, task automatic cleaning of expired files can be set.

[0069] In summary, the present application proposes: tree directory to achieve cross-document chapter-level precise alignment; "entity extraction-independent calculation-result interpretation" three-level data processing chain to avoid large language model calculation defects; subject unit label system to solve the problem of multi-source content attribution.

[0070] In one embodiment, a briefing merging system based on a large language model is provided, which includes:

[0071] Directory analysis module: used to call a selected large language model to generate a directory tree, and integrate a rule engine to post-process the model output.

[0072] Document parsing engine: integrated adaptive text extraction tool for parsing each original text of the brief, identifying and extracting the text segment corresponding to the directory node in each brief according to the generated directory tree node information, extracting the subject unit identifier, and establishing a structured mapping relationship;

[0073] Content merging processor: as the core processing unit, used for performing the shunting processing of computing and non-computing content, scheduling large language model for content generation and entity extraction, integrating pre-set computing script library and execution engine, calling and managing the execution environment of computing script, and implementing the conversion of structured data to analysis text;

[0074] Official document formatting output module: used for generating the final document conforming to the administrative specification based on the template engine, and providing a download interface.

[0075] The specific limitations of the brief merging system based on the large language model can be referred to the limitations of the brief merging method based on the large language model in the above, which will not be repeated here. Each module in the high-fidelity image reconstruction system of the super-resolution structured light illumination microscope can be realized by software, hardware and their combination. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations of the above modules.

[0076] In one embodiment, a computer device is provided, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the following when executing the computer program:

[0077] Step 1, directory structure generation: using a large language model to analyze the semantic theme and hierarchical logic of multiple briefs, generating an initial directory tree structure, and comparing and correcting it with a pre-set rule library, finally generating a standardized directory tree structure;

[0078] Step 2, content structured extraction: parsing each original text of the brief, identifying and extracting the text segment corresponding to the directory node in each brief according to the generated directory tree node information, extracting the subject unit identifier, and establishing a structured mapping relationship;

[0079] Step 3, multi-source content merging: for each node in the directory tree, differential merging processing is performed according to its content type;

[0080] Step 4, automatic generation of official documents: adding standard header and tail information, setting uniform document style, inserting page numbers to the merged structured content according to the target official document specification, and generating the final official document that can be directly used.

[0081] The specific definition of each step can refer to the definition of the brief report merging method based on the large language model in the foregoing, which will not be repeated here.

[0082] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement:

[0083] Step 1, directory structure generation: using a large language model to analyze the semantic topics and hierarchical logic of multiple brief reports, generating an initial directory tree structure, and comparing and correcting with a preset rule base, finally generating a standardized directory tree structure;

[0084] Step 2, content structured extraction: parsing the original text of each brief report, identifying and extracting the text segments corresponding to the directory node in each brief report according to the generated directory tree node information, and extracting the subject unit identifier to establish a structured mapping relationship;

[0085] Step 3, multi-source content merging: for each node in the directory tree, differentially merging the content according to its content type;

[0086] Step 4, automatic generation of official documents: adding standard header and tail information, setting uniform document style, inserting page numbers to the merged structured content according to the target official document specification, and generating the final official document that can be directly used.

[0087] The specific definition of each step can refer to the definition of the brief report merging method based on the large language model in the foregoing, which will not be repeated here.

[0088] In the present application, the large language model dynamically selects the processing path according to the generated directory tree node type (descriptive or computational). For descriptive nodes, the large language model is directly called to integrate the text with source identification to generate a standard description; for computational nodes, the professional process of "large language model entity extraction Python script numerical calculation large language model fusion analysis" is chain called, and the structured parameters are strictly passed. Finally, the Python-docx tool is called to automatically generate a format compliant official document according to the national standard parameters, realizing the efficient cooperation of the intelligence of the large language model and the professional tool capability. The scheme of the present application significantly improves the efficiency and accuracy of merging multiple source brief reports into standard official documents.

[0089] Through experimental verification, compared with the traditional manual merging method, the present method significantly improves the efficiency by more than 50%; under the premise of correct entity extraction, the data calculation accuracy is > 99%; the output document format compliance reaches 100%, strictly complying with the national standard.

[0090] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A large language model-based intelligent briefing merging method, characterized in that, The method comprises the following steps: Step 1, directory structure generation: using a large language model to analyze the semantic topics and hierarchical logic of multiple briefing papers, generating an initial directory tree structure, and comparing and correcting it with a preset rule library, finally generating a standardized directory tree structure; Step 2, content structured extraction: parsing the original text of each briefing paper, identifying and extracting the text segments corresponding to the directory node information, and extracting the subject unit identifier to establish a structured mapping relationship; Step 3, multi-source content merging: for each node in the directory tree, differentially merge the content according to its type; Step 4, automatic generation of official documents: after merging the structured content, automatically adding standard header and footer information, setting a unified document style, inserting page numbers, and generating a final official document that can be directly used.

2. The large language model-based intelligent briefing merging method according to claim 1, characterized in that, In step 1, the correction specifically includes: marking the directory tree node as a calculation type or a non-calculation type attribute, and converting the corrected directory into a tree-shaped data structure in memory, where the root node is the document topic, the child node is the chapter title, and the node object record includes attributes such as title text, hierarchical depth, and expected content type. 3.The method of claim 1, wherein, Step 2 specifically includes: Use the pre-defined unit name regular pattern list to match strings and extract the subject unit from the briefing as a content identification tag; According to the directory tree node information, apply accurate string matching based on the directory node title text, and combine the title's hierarchical depth and the document's relative position information to infer the cutting strategy, establish the "directory node ID-main unit identifier-text content" mapping relationship, and store it in a structured JSON object.

4. The large language model-based intelligent briefing merging method according to claim 1, characterized in that, Step 3 specifically includes: For non-calculation type nodes, use the mapping relationship to add unit identifier prefixes to each source text to ensure that the subject unit identifier is always associated with the corresponding text segment, and then use a large language model to generate a standard narrative text; For calculation type nodes, perform the following in sequence: structured entity extraction based on pre-defined prompt word templates, call independent scripts to perform numerical calculations, input the original extracted entity data and script calculated result data into a large language model, and generate a comprehensive analysis fusion text containing data interpretation by the large language model.

5. The large language model-based intelligent briefing merging method according to claim 4, characterized in that, For the processing of calculation type nodes, it also includes: The large language model entity extraction uses pre-defined, configurable prompt word templates, and the extracted structured entity data needs to be associated with the source subject unit identifier; The calculation logic of the script execution includes basic summation, average value, and complex operations such as weighted average, standard deviation, cumulative value, completion rate, and index comparison.

6. The large language model-based intelligent briefing merging method according to claim 4, characterized in that, When the large language model generates a comprehensive analysis fusion text containing data interpretation, it needs to clearly specify the specifications of the model-generated content and ensure that the interpretation reflects the relationship between the data and the source subject unit, so that the generated content meets the requirements of industry background knowledge and analysis dimensions, and ensures that the interpretation is based on the provided original data and calculation results.

7. The large language model-based intelligent briefing merging method according to claim 1, characterized in that, In step 4, the automatic generation of official documents specifically includes: By reading the pre-configured official document template file or configuration parameters, automatically inserting the government standard official document header; The style setting interface of the adaptive text extraction tool is called to set the body, titles of different levels, line spacing and page margins according to the official document standard; The document segmentation and footer editing functions are used to add right-aligned page numbers in the footer and ensure continuous page counting.

8. An intelligent brief combining system for implementing the method according to any one of claims 1 to 7, characterized in that, The system comprises: A directory analysis module for calling a selected large language model to generate a directory tree and integrating a rule engine to post-process the model output. A document parsing engine integrated with an adaptive text extraction tool for parsing original texts of each briefing, identifying and extracting text segments corresponding to the directory node in each briefing according to the generated directory tree node information, extracting the subject unit identifier, and establishing a structured mapping relationship; A content merging processor serving as a core processing unit for performing the shunting processing of computational and non-computational contents, scheduling a large language model to generate content and extract entities, integrating a preset computing script library and an execution engine, calling and managing the execution environment of the computing script, and implementing the conversion of structured data to analysis text; An official document formatting output module for generating a final document conforming to administrative regulations based on a template engine and providing a download interface.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

Images