Automatic report document generation method fusing large language model and visual component

Abstract

According to the automatic report document generation method fusing the large language model and the visual component, a conversion scheme from data to a report is briefly and clearly constructed through three steps of data definition, component definition and template management, and intelligent data acquisition, text retouching, intelligent data analysis, intelligent style adjustment and intelligent document summarization are realized by fusing the large language model. According to the method, a user is helped to construct an efficient and available report template, template contents and styles are freely combined through a visual editing interface, and a high-availability report document is output. Compared with the prior art, the method has the following characteristics: (1) the intelligent summarization analysis capability of data driving is realized; (2) a high-degree-of-freedom and visual template editing interface is supported; and (3) the modularization architecture realizes'one-time definition and multi-report multiplexing '. And (4) the RAG enhanced natural language to SQL capability is fused, and the data access threshold is reduced. And (5) supporting AI-driven style optimization and full-text summarization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, further refers to the visual report document output application of data analysis, in particular to an automatic report document generation method combining large language model and visualization components. BACKGROUND

[0002] In the related work of data-driven and decision-making, such as communication operator network operation work, by analyzing customer network operation data, output customer network service report, can help users to arrange and make decisions related work.

[0003] The analyzed data includes various detailed data (such as alarm details, performance index details, work order details, etc.) and statistical result data (service compliance indicators, business size indicators, fault condition indicators, etc.). If using traditional manual method, it needs to collect data, formulate analysis target, disassemble analysis task, formulate analysis method, conduct statistical analysis, obtain analysis conclusion, output analysis content, and make analysis report. The analysis cycle is long and the labor cost is high. According to the existing conventional system application scheme, it usually provides fixed analysis target, analysis template and analysis data content, and realizes fixed report output through system. However, this needs to be redeveloped, cannot meet the rapid custom adjustment of report content, and the report conclusion is pre-set, which is too rigid to dynamically output. Although this saves a part of manual operation time, it increases a lot of development cost. Therefore, it is necessary to provide a tool with intelligent automatic analysis and data dynamic adaptability and rapid template adjustment.

[0004] Some existing automatic analysis report related tools and methods have the following problems:

[0005] 1. Cannot intelligently summarize the analysis of data. Summarized analysis is a descriptive analysis conclusion of data, such as whether the data has outstanding items, the degree of year-on-year and month-on-month change, and processing suggestions. Intelligence refers to giving only a general analysis target, and the large language model intelligently summarizes the data. For example, patent document CN119886099A "Automatic report system based on data-driven", "the text part is written in Markdown format, used to control the text content in the generated analysis report template and the structure of the analysis report template", which only has fixed template structure of text content, cannot realize the requirements;

[0006] 2. The lack of a highly usable and flexible visual template editing interface, including rich text editing, chart definition, and template editing, fails to meet the ever-changing reporting needs. For example, the template in patent document CN119886099A, a data-driven automated reporting system (medical, fixed template, emphasizing statistical methods), is written in R language and cannot be visually modified by the user. Other patent documents, such as CN119961330A, an industry report automatic generation system based on a large language model (finance, fixed template), CN117851481A, a method, system, medium, and equipment for automatic production of data analysis reports (fixed template), CN110263076A, a method for automatically generating data analysis reports (fixed template), CN108735275A, an automatic report generation system and report generation method (fixed template), and CN104903891A, an automatic report generation method (fixed template), all lack a visual interface for template editing, resulting in poor adaptability and practicality. Summary of the Invention

[0007] This invention addresses the shortcomings or defects in existing technologies by providing an automated report document generation method that integrates large language models and visualization components. The method involves constructing the dataset required for the report through data definition, referencing the data and creating a component set through component definition (such as text or charts), and finally using a visualization template arrangement module to modularly combine the components to form a report template, thereby achieving automated generation of Word / PPT format report documents.

[0008] The technical solution of the present invention is as follows:

[0009] An automated reporting document generation method integrating large language models and visualization components is characterized by the following steps:

[0010] Step 1: Construct the dataset required for the report through data definition. The data definition divides the data into detailed data and statistical data. The detailed data is used for subsequent detailed list output and as the analysis content of the large language model. The statistical data is used for subsequent chart creation and structured report output.

[0011] Step 2: Transform the dataset into presentable and composable report units through component definitions. The component definitions include structured components built by referencing statistical data, and unstructured components built by referencing detailed data and statistical data. The structured components are used for visualization or structured display, and the unstructured components are used for text generation or detailed list display.

[0012] Step 3, create a diversified report template and generate a report document through the template management combination component, the template management is the final integration layer of the entire AI automated report system, the template itself does not contain fixed data, but exists in the form of skeleton + dynamic content, so as to automatically or manually trigger data refresh and document output at specified time granularity, and ensure that the report content is always consistent with the latest business data, the report document includes Word document and PPT document.

[0013] The detailed data in step 1 includes work order data, conversation record data and log data, and the statistical data is a data type defined for charts and structured analysis, including three mandatory items of statistical dimension, statistical granularity and statistical index.

[0014] The data source mode of data definition in step 1 includes pre-defined SQL query mode and RAG enhanced natural language to SQL mode, through large language model to understand and analyze the database in RAG vector space, and then convert it into SQL language for query statistics, the result of data definition is output data list, including data name, data granularity, data source code and statistical range.

[0015] The structured component in step 2 relies on the statistical dimension, statistical index and statistical granularity defined in the statistical data to automatically map to the horizontal axis, vertical axis, legend or numerical label of the chart, realizing the automatic binding of data to view, the unstructured component includes AI analysis class component and detailed list class component, the AI analysis class component takes the detailed data as the context input to the large language model, and automatically generates natural language description, problem diagnosis or trend judgment and analysis conclusion suggestion content, the detailed list class component directly displays the data content in the form of table.

[0016] For the detailed list class component, turn on the AI analysis switch to automatically call the large language model to summarize, compare or root cause the detailed content, and generate natural language paragraphs.

[0017] The result of component definition in step 2 is to form a component list containing component name, type, reference data code, presentation configuration and AI strategy meta information, the component list is used as the basic material library for report template construction, which supports on-demand drag and combination in the subsequent report definition stage, realizing the flexible architecture of one-time definition and multi-report reuse.

[0018] In step 3, intelligent interpretation is enabled for statistical data to provide textual descriptions for index changes or outliers; AI style optimization is enabled for report styles, so that a large language model can intelligently recommend style parameters according to the content of the entire report, without the need for users to manually adjust the tedious details of various component styles; and an AI summary component is created for the entire report, so that a large language model can analyze and summarize the entire report and output summary content.

[0019] In the Word document in step 3, components are embedded in the form of rich text areas, which supports mixed arrangement with ordinary text to form a coherent analysis document; in the PPT document, multiple components can be freely arranged on each page, and auxiliary content such as rectangles, ellipses, lines, and / or pictures can be inserted, and the level, size, background, and border can be adjusted to build a rich presentation page.

[0020] The technical effects of the present application are as follows: The automatic reporting document generation method of the present application fuses a large language model and a visual component, and through three steps of data definition, component definition, and template management, a conversion scheme from data to report is briefly and clearly constructed, and a large language model is further fused to realize intelligent data acquisition, text polishing, data intelligent analysis, style intelligent adjustment, and document intelligent summary, helping users to construct efficient and usable report templates, and through a visual editing interface, template content and styles can be freely combined to output high-usable report documents.

[0021] Compared with the prior art, the present application has the following characteristics:

[0022] (1) Intelligent summary and analysis capabilities driven by data are realized.

[0023] (2) A high-degree and visual template editing interface is supported.

[0024] (3) A component-based architecture realizes "one definition, multiple report reuse".

[0025] (4) The natural language to SQL conversion capability of RAG is enhanced to reduce the data access threshold.

[0026] (5) AI-driven style optimization and full-text summary are supported. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a process schematic diagram of the automatic reporting document generation method of the present application fusing a large language model and a visual component. Figure 1 Step 1, data definition; step 2, component definition; and step 3, template management and report generation. DETAILED DESCRIPTION

[0028] The present application will be described below in conjunction with the accompanying drawings ( Figure 1 ) and examples.

[0029] Figure 1 is the flowchart of the fusion of large language model and visualization component of the automatic report document generation method of the present application. Referring to Figure 1 , the fusion of large language model and visualization component of the automatic report document generation method, comprising the following steps: step 1, constructing the required data set of the report through data definition, the data definition divides the data into detailed data and statistical data, the detailed data is used for subsequent detailed list output and as the analysis content of the large language model, and the statistical data is used for subsequent chart creation and structured report output; step 2, converting the data set into presentable and combinable report units through component definition, the component definition includes structured components constructed by referencing statistical data, and unstructured components constructed by referencing detailed data and statistical data, the structured components are used for visualization or structured display, and the unstructured components are used for text generation or detailed list display; step 3, combining components through template management, creating diversified report templates, and generating report documents, the template management is the final integration layer of the entire AI automatic report system, the template itself does not contain fixed data, but exists in the form of skeleton + dynamic content, so as to automatically or manually trigger data refresh and document output at a specified time granularity, and ensure that the report content is always consistent with the latest business data, and the report document includes Word document and PPT document.

[0030] The detailed data in step 1 includes work order data, dialogue record data and log data, the statistical data is a data type defined for charts and structured analysis, including three mandatory items of statistical dimension, statistical granularity and statistical index. The data source mode of data definition in step 1 includes pre-defined SQL query mode, and also includes RAG enhanced natural language to SQL mode, the database in RAG vector space is understood and parsed through the large language model, and then converted into SQL language for query statistics, the result of data definition is output data list, including data name, data granularity, data source CODE and statistical range.

[0031] The structured components in step 2 rely on the statistical dimension, statistical index and statistical granularity defined in the statistical data, and are automatically mapped to the horizontal axis, vertical axis, legend or numerical label of the chart, realizing the automatic binding of data to view, the unstructured components include AI analysis components and detailed list components, the AI analysis components input the detailed data as context to the large language model, and the large language model automatically generates natural language description, problem diagnosis or trend judgment and analysis conclusion suggestion content, the detailed list components directly display the data content in the form of table. For the detailed list components, the AI analysis switch is turned on to automatically call the large language model to summarize, compare or root cause the detailed content, and generate natural language paragraphs.

[0032] The result of the component definition in step 2 is to form a component list containing component name, type, reference data CODE, presentation configuration and AI strategy meta information, which serves as a basic material library for report template construction, supports on-demand drag and combination in the subsequent report definition stage, and realizes flexible architecture of one-time definition and multi-report reuse.

[0033] In step 3, for statistical data, intelligent interpretation is enabled to provide textual description of index changes or abnormal values; for report style, AI style optimization is enabled to allow a large language model to intelligently recommend style and style parameters according to the content of the entire report, without the need for users to manually adjust the tedious details of various component styles; for the entire report, an AI summary component is created separately to allow a large language model to analyze and summarize the entire report, and output summary description content. In the Word document in step 3, components are embedded in the form of rich text areas, supporting mixed arrangement with ordinary text to form a coherent analysis document; in the PPT document, each page can freely layout multiple components, and also supports inserting auxiliary content such as rectangles, ellipses, lines and / or pictures, and supports adjusting hierarchy, size, background and border to build rich demonstration pages.

[0034] The present application proposes an automatic report document generation method combining a large language model and visual components, the overall strategy of which is to build a data set required for a report through data definition, then reference data through component definition (such as text or chart) and create a component set, and finally use a visual template arrangement module to modularly combine components to form a report template, which can automatically generate Word / PPT format report documents according to report granularity (day, week, month, etc.) and report generation time.

[0035] In combination Figure 1 , the implementation steps of the present application are as follows:

[0036] In the first step S1, a data set required for a report is built through data definition:

[0037] Specifically, data definition divides data into two granularities, including detailed data and statistical data, the detailed data refers to various detailed data (such as work orders, dialogue records, logs, etc.), which can be used for subsequent detailed list output, or can be used as analysis content for a large model; the statistical data is a data type defined for charts and structured analysis, including three mandatory items of statistical dimension, statistical granularity and statistical index (such as work order situation statistics, business size statistics, etc.), which is used for subsequent chart creation and structured report output.

[0038] Data definition of data sources, including traditional pre-defined SQL query mode (SQL, Structured Query Language, structured query language), and RAG enhanced natural language to SQL mode (RAG, Retrieval-Augmented Generation, retrieval-augmented generation), the former is to develop some data queries or statistical SQL used for reporting in advance, and the latter does not need to be developed in advance, only needs to save the database structure into the RAG vector space, and then converts it into SQL language for query and statistics through a large language model to understand and analyze the database. Finally, the above two SQLs are packaged into CODE for identification (CODE, code).

[0039] The content of data definition includes the data granularity as described above, and also includes the range of data, which is divided into spatial range and time range. The spatial range further refers to the conditions of data query or statistics, such as filtering data of a certain province or a certain customer. The spatial range can specify the data query content according to user demand; the time range is to define the period (daily / weekly / monthly, etc.) and the range of the period (such as 7 days, 3 weeks, 6 months, etc.) of data query. The time range can correspond to the appeal of different report granularity (daily report, weekly report or monthly report, etc.).

[0040] The result of data definition is the output data list, including data name, data granularity, data source CODE, and statistical range, which is used for subsequent component definition.

[0041] The second step S2 is to convert the data set into a report unit that can be presented and combined through component definition:

[0042] Specifically, component definition is an intermediate bridge connecting "data definition" and final "report output". Its core goal is to convert the defined data set (from S1) into a report component unit with a specific presentation form and semantic intention. Each component represents an independent content module in the report, such as a chart, an analysis text, a detailed table, etc. Multiple components can be freely combined to form a complete report.

[0043] Component definition is divided into two types: structured component and unstructured component, according to the granularity type of the referenced data.

[0044] Structured component: reference statistical data construction, used for visualization or structured display. Typical forms include column chart, line chart, pie chart, data card, etc. This type of component relies on the statistical dimensions, statistical indicators and statistical granularity defined in the statistical data to automatically map the horizontal axis, vertical axis, legend or numerical label of the chart, realizing the automatic binding of "data as view".

[0045] Unstructured components: reference detailed data and statistical data construction for text generation or detailed list display. Typical forms include AI analysis paragraphs (AI, Artificial Intelligence), work order summary lists, dialogue log segments, etc. Among them, AI analysis components input detailed data as context into large language models, and automatically generate natural language descriptions, problem diagnoses or trend judgments, and analysis conclusions and suggestions; and detailed list components directly display data content in table form.

[0046] Whether structured or unstructured components, the system will retain the original page style code to meet the various style requirements required by users when generating reports.

[0047] Component definition process, including data reference: select a defined data item from the "data list" output from S1, and the component will completely inherit the data's granularity, source CODE, spatial range, and time granularity attributes. Presentation type: automatically recommend available component types based on data granularity (such as statistical data charts, detailed data text or list), support further configuration of chart styles, sorting rules, field display, etc.

[0048] AI enhancement: for detailed data components, the "AI analysis" switch can be turned on, and the system will automatically call large models to summarize, compare, or root cause the detailed content and generate natural language paragraphs; for statistical data, "intelligent interpretation" can also be enabled, which allows the model to explain index changes or outliers in words; for report style, "AI style optimization" can also be enabled, allowing the large model to intelligently recommend style and style parameters based on the content of the entire report, eliminating the need for users to manually adjust the tedious details of various component styles; for the full text of the report, an AI summary component can also be created separately, allowing the large model to analyze and summarize the full text of the report and output summary content.

[0049] The result of component definition is to form a component list containing component name, type, reference data CODE, presentation configuration, and AI strategy. This list serves as a basic material library for report template construction, supporting drag-and-drop combination as needed during subsequent report definition phase, achieving a flexible architecture of "one definition, multiple report reuse".

[0050] Step 3 S3, combine components through template management, create diversified report templates, and generate report documents:

[0051] Specifically, template management is the final integration layer of the entire AI automated reporting system, and its core goal is to organize various components (such as charts, AI analysis paragraphs, detailed tables, etc.) defined in S2 into a structured reporting template in a flexible and configurable manner, and support automatic or manual triggering of data refresh and document output at a specified time granularity. The template itself does not contain fixed data, but exists in the form of "skeleton + dynamic content", ensuring that the report content always keeps up with the latest business data.

[0052] The report document type supports both Word and PPT, and provides a visual editing interface to combine the defined components. In the Word template, components are embedded in the form of rich text areas, supporting mixed arrangement with ordinary text to form a coherent analysis document; in the PPT template, multiple components can be freely laid out on each page, and auxiliary content such as rectangles, ellipses, lines, and pictures can also be inserted, with support for adjusting levels, sizes, backgrounds, and borders to build rich presentation pages.

[0053] For the settings of the document template, including selecting the corresponding components by setting the report granularity, obtaining the data of the corresponding granularity, and switching the report content of different time points by setting the reference time. By configuring a timing task (such as every month at 9:00 am), the system can automatically generate and save reports according to the latest period, facilitating regular pushing or archiving.

[0054] The saving and generation of the document template preserves the original configuration of each inserted component in the template (such as the referenced data CODE, AI prompt words, chart styles, content positions, etc.), and re-executes data query and content rendering according to the current reference time during document generation, and outputs the final Word and PPT documents.

[0055] The present application is an automatic report document generation method and system that fuses a large language model and visual components, which briefly and clearly constructs a data-to-report conversion scheme through three steps of data definition, component definition, and template management, and then fuses a large language model to realize data intelligent acquisition, text polishing, data intelligent analysis, style intelligent adjustment, and document intelligent summary, helping users to construct efficient and usable report templates, and freely combining template content and styles through a visual editing interface to output high-usable report documents.

[0056] In the three steps of data definition, component definition, and template management, data definition constructs the data set required for the report, then component definition (such as text or chart) references the data and creates a component set, and finally the visual template arrangement module is used to modularly combine the components to form a report template.

[0057] Regarding the fusion of large language models: the large language model mentioned in the invention assists in design, including RAG enhanced natural language to SQL, which only needs to save the database structure to the RAG vector space, understand and analyze the database through the large language model, and then convert it into SQL language for query statistics. And AI component enhancement, for detailed data components, the system will automatically call the large model to summarize, compare or root cause the detailed content, and generate natural language paragraphs; for statistical data, the model will provide a textual explanation of the index changes or outliers; for report style, let the large model intelligently recommend style and style parameters according to the content of the entire report; for the full text of the report, an AI summary component can also be created separately, allowing the large model to analyze and summarize the full text of the report.

[0058] Visual editing interface: supports both Word and PPT document formats, and users can easily configure document template content through the interface, and supports free combination and various style detail settings.

[0059] Compared with the prior art, the technical advantages of the invention are as follows:

[0060] (1) Regarding the realization of data-driven intelligent summary and analysis capabilities:

[0061] The prior art (such as CN119886099A) only supports static text filling based on fixed templates, and cannot dynamically and intelligently understand and summarize data. The invention combines large language models (LLM) to automatically generate natural language analysis conclusions, trend judgments, anomaly diagnoses, and processing suggestions based on detailed data or statistical data, truly realizing "data → insight → text" end-to-end intelligent analysis, and significantly improving the depth and practicality of report content.

[0062] (2) Regarding the support of high degree of freedom and visual template editing interface:

[0063] Existing solutions rely on code (such as R language) or pre-defined structures to define report templates, and users cannot intuitively modify or flexibly adjust them. The invention provides a visual template editor for Word and PPT, supporting the drag-and-drop combination and style configuration of rich text, charts, and detailed lists, allowing users to quickly build or adjust report structures without programming, greatly improving the customizability and efficiency of templates.

[0064] (3) Regarding the component-based architecture to realize "one definition, multiple report reuse":

[0065] The application abstracts the report content into reusable "components" (including structured chart components and unstructured AI text components), each of which is independently bound with data definition and presentation logic. Compared with the solidification mode of the prior art in which report content and data logic are strongly coupled, the present scheme supports flexible reuse of components in different templates and different time granularities (day / week / month), reduces maintenance cost and improves system scalability.

[0066] (4) Regarding the fusion of RAG enhanced natural language to SQL capability, the data access threshold is reduced:

[0067] The traditional system needs to develop SQL query statements in advance, which has high development cost and poor flexibility. The application introduces RAG (Retrieval-Augmented Generation) technology, embeds the database structure into the vector space, so that the large language model can understand the user's natural language instructions and automatically generate accurate SQL, realizes "zero code" data definition, and significantly improves the data usage ability of non-technical personnel.

[0068] (5) Regarding supporting AI-driven style optimization and full-text summary:

[0069] The prior art completely relies on manual setting of report style and conclusion. The application further expands the application boundary of the large language model, not only generates content, but also intelligently recommends style parameters such as color matching, layout, font, etc. according to the overall semantics of the report, and can automatically generate a full-text abstract, realizing the full-link intelligentization from "content generation" to "overall presentation".

[0070] The contents not described in detail in the specification of the application belong to the prior art known to those skilled in the art. It is indicated here that the above description is helpful for those skilled in the art to understand the application, but does not limit the protection scope of the application. Any implementation of equivalent replacement, modification, improvement and / or deletion of the above description without departing from the essential content of the application falls within the protection scope of the application.

Claims

1. An automated report document generation method fusing a large language model and a visualization component, characterized in that, Comprising the following steps: Step 1, build the required data set for the report through data definition, which divides the data into detailed data and statistical data, the detailed data is used for subsequent detailed list output and as the analysis content of the large language model, the statistical data is used for subsequent chart creation and structured report output; Step 2, convert the data set into presentable and combinable report units through component definition, the component definition includes structured components constructed by referencing statistical data, and unstructured components constructed by referencing detailed data and statistical data, the structured components are used for visualization or structured display, and the unstructured components are used for text generation or detailed list display; Step 3, combine components through template management, create diversified report templates, and generate report documents, the template management is the final integration layer of the entire AI automated report system, the template itself does not contain fixed data, but exists in the form of skeleton + dynamic content, so as to automatically or manually trigger data refresh and document output at specified time granularity, and ensure that the report content is always consistent with the latest business data, the report document includes Word document and PPT document.

2. The method of claim 1, wherein the method further comprises: The detailed data in step 1 includes work order data, conversation record data and log data, the statistical data is a data type defined for charts and structured analysis, including three mandatory items of statistical dimension, statistical granularity and statistical indicator. 3.The method of claim 1, wherein the method further comprises: The data source mode of data definition in step 1 includes pre-defined SQL query mode, and RAG enhanced natural language to SQL mode, through the large language model to understand and analyze the database in the RAG vector space, and then convert it into SQL language for query statistics, the result of data definition is output data list, including data name, data granularity, data source CODE and statistical range. 4.The method of claim 1, wherein, The structured components in step 2 rely on the statistical dimension, statistical indicator and statistical granularity defined in the statistical data, which are automatically mapped to the horizontal axis, vertical axis, legend or numerical label of the chart, realizing the automatic binding of data to view, the unstructured components include AI analysis components and detailed list components, the AI analysis components input the detailed data as context to the large language model, and the large language model automatically generates natural language description, problem diagnosis or trend judgment and analysis conclusion suggestion content, the detailed list components directly display the data content in the form of table.

5. The method of claim 4, wherein the method further comprises: For detailed list components, turn on the AI analysis switch to automatically call the large language model to summarize, compare or root cause the detailed content, and generate natural language paragraphs.

6. The method of claim 1, wherein the method further comprises: The result of component definition in step 2 is to form a component list containing component name, type, reference data CODE, presentation configuration and AI strategy meta information, the component list is used as the basic material library for report template construction, supporting on-demand drag and combination in the subsequent report definition stage, realizing the flexible architecture of one definition and multiple report reuse.

7. The method of claim 1, wherein the method further comprises: In step 3, for statistical data, intelligent interpretation is enabled to provide textual descriptions for index changes or outliers; for report style, AI style optimization is enabled, allowing the large language model to intelligently recommend style and style parameters based on the content of the entire report, eliminating the need for users to manually adjust the tedious details of various component styles; for the entire report, an AI summary component is created separately, allowing the large language model to analyze and summarize the entire report and output summary content. 8.The method of claim 1, wherein, In the Word document in step 3, components are embedded in the form of rich text areas, supporting mixed arrangement with ordinary text to form a coherent analysis document; in the PPT document, multiple components can be freely arranged on each page, and auxiliary content such as rectangles, ellipses, lines, and / or pictures can be inserted, allowing adjustment of hierarchy, size, background, and border to build rich demonstration pages.

Images