Intelligent industrial report generation platform fusing an industrial large model and network data crawling

Abstract

The embodiment of the specification provides an intelligent industrial report generation platform fusing an industrial large model and network data crawling, wherein the intelligent industrial report generation platform fusing the industrial large model and the network data crawling crawls and pre-processes multi-source network data through a data acquisition module to provide structured input for subsequent analysis; an industrial large model analysis module performs deep analysis on the structured data based on a pre-trained language model optimized by industrial knowledge to generate analysis conclusions including trend identification, competition pattern and market prediction; and a report generation module automatically forms a complete report with pictures and texts according to the analysis conclusions. The application realizes full-process automation from data acquisition, intelligent analysis to report generation, and significantly improves the efficiency and timeliness of industrial report writing.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of industry analysis technology, and in particular to an intelligent industry report generation platform that integrates large-scale industry models with web data scraping. Background Technology

[0002] In the field of industry analysis, traditional report writing methods rely heavily on manual data collection and organization, which is not only time-consuming and inefficient but also fails to meet the rapidly changing market information demands. Existing automated tools typically generate content based on fixed templates, lacking the ability to perceive and integrate dynamic online data in real time. This results in reports with limited analytical dimensions and outdated data, making it difficult to produce professional analytical results with in-depth insights.

[0003] Therefore, a better solution is urgently needed. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide an intelligent industry report generation platform that integrates large-scale industry models and web data crawling to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, an intelligent industry report generation platform that integrates large-scale industry models and web data crawling is provided, comprising: The data acquisition module is used to capture raw data related to the industry from multiple network data sources and preprocess the raw data to obtain structured data. The Industry Big Data Model Analysis Module is used to analyze structured data based on a pre-trained language model fine-tuned with industry knowledge, and generate industry analysis conclusions. The report generation module is used to automatically generate industry reports containing text descriptions and visual charts based on industry analysis conclusions.

[0006] In one possible implementation, the data acquisition module includes a data crawling unit for periodically accessing a preset network data source via a distributed crawler to crawl raw web page data; a data cleaning unit for removing noise from the raw web page data to obtain cleaned text; an entity recognition unit for extracting key entity information from the cleaned text and storing the key entity information in association with the cleaned text; and a timeliness evaluation unit for scoring the timeliness and credibility of the cleaned text based on the data source, publication time, and frequency of citation, generating a timeliness weight.

[0007] In one possible implementation, the industry big data model analysis module includes an industry knowledge vector library, which stores text paragraphs extracted from high-quality industry documents and their vectorized representations; a trend identification unit, used to input structured data into a pre-trained language model, identify hot events and trends within the industry, and generate event summaries; a competitive landscape analysis unit, used to analyze the market share changes and competitive dynamics of major companies based on enterprise entity information in structured data, and generate competitive landscape analysis text; a market size prediction unit, used to extract historical statistical information from structured data, combine it with industry growth rates to infer the current market size, and generate future prediction descriptions; and a sentiment and risk assessment unit, used to perform sentiment scoring and risk level assessment on public opinion information, and generate risk analysis conclusions.

[0008] In one possible implementation, the trend identification unit, competitive landscape analysis unit, market size prediction unit, and sentiment and risk assessment unit call on the vectorized representation in the industry knowledge vector library for comparison and verification when generating analysis conclusions, so as to improve the consistency between the analysis conclusions and industry professional terminology.

[0009] In one possible implementation, the report generation module includes a report structure arrangement unit for automatically planning the report chapter structure based on the industry analysis conclusions and determining the logical order of paragraphs within each chapter; a text generation unit for expanding and refining the industry analysis conclusions to generate a report text with logical depth; a chart generation unit for extracting numerical sequences from the industry analysis conclusions, automatically selecting chart types, and generating visual charts with data labels and analytical annotations; and a text-image fusion unit for associating the report text with the visual charts and formatting them according to the structure planned by the report structure arrangement unit to generate a complete report document.

[0010] In one possible implementation, the chart generation unit automatically selects one of the chart types, including pie chart, line chart, bar chart, radar chart, or bubble chart, and adapts it according to the data type.

[0011] In one possible implementation, a user interaction module is also included. This module receives the industry analysis topic, time range, and key focus areas set by the user, and passes the user settings to the data acquisition module to adjust the data capture strategy, and to the report generation module to customize the report content.

[0012] In one possible implementation, the user interaction module is also used to provide a report preview interface, receive user instructions on modifying the report content, and record the modification behavior as feedback data.

[0013] In one possible implementation, the user interaction module supports exporting the generated report as PDF, HTML, or PPT format, and adapting the layout according to the selected format.

[0014] In one possible implementation, the user interaction module is also used to automatically save each generated report version and build an index to support historical report tracing and comparison.

[0015] This specification provides an intelligent industry report generation platform that integrates industry big data models and web data scraping. The platform uses a data acquisition module to scrape and preprocess multi-source web data, providing structured input for subsequent analysis. An industry big data model analysis module performs in-depth analysis of the structured data based on a pre-trained language model optimized with industry knowledge, generating analytical conclusions including trend identification, competitive landscape, and market forecasts. The report generation module automatically generates a complete, richly illustrated report based on these analytical conclusions. This invention automates the entire process from data acquisition and intelligent analysis to report generation, significantly improving the efficiency and timeliness of industry report writing. Simultaneously, the deep reasoning capabilities of the big data model enhance the professionalism and forward-looking nature of the analytical conclusions, effectively overcoming the shortcomings of superficial report content and data lag in existing technologies, providing users with high-quality, customized industry analysis results. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an intelligent industry report generation platform that integrates a large-scale industry model with web data crawling, provided in one embodiment of this specification. Detailed Implementation

[0017] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0018] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0019] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0020] This specification provides an intelligent industry report generation platform that integrates large-scale industry models with web data scraping, which will be described in detail in the following embodiments.

[0021] See Figure 1 , Figure 1 This diagram illustrates a system schematic of an intelligent industry report generation platform that integrates an industry big data model and web data scraping, according to an embodiment of this specification. Specifically, it includes a data acquisition module for scraping raw industry-related data from multiple web data sources and preprocessing the raw data to obtain structured data; an industry big data model analysis module for analyzing the structured data based on a pre-trained language model fine-tuned with industry knowledge to generate industry analysis conclusions; and a report generation module for automatically generating an industry report containing textual descriptions and visual charts based on the industry analysis conclusions.

[0022] The data acquisition module can refer to the functional unit responsible for obtaining raw information from the internet and performing preliminary processing. For example, it might use a distributed web crawler to periodically access various industry websites and extract text, tables, and other data, then perform deduplication and format standardization to create standardized data suitable for subsequent analysis. Raw data can refer to unprocessed information directly scraped from the web, such as news articles, company announcements, government statistical bulletins, and social media discussions; this data often contains a lot of noise. Structured data refers to data sets that have been cleaned, classified, and labeled to have a certain format, such as database records organized by time, entity, and numerical fields, making them easy for the analysis module to use directly. The industry large-scale model analysis module can refer to an intelligent analysis unit formed by optimizing a large-scale language model based on deep learning for a specific industry sector. For example, a model using the Transformer architecture and fine-tuned using industry reports and patent documents can understand industry terminology and generate professional analytical conclusions. Industry analysis conclusions can refer to the model's inferences and summaries about the current state, trends, and competitive relationships of the industry, such as descriptions of market size growth, ranking of major companies' market shares, and predictions of technological development directions. The report generation module can refer to an automated component that transforms analysis results into human-readable documents. For example, it can combine text paragraphs with automatically generated charts such as bar charts and pie charts based on preset templates to create a neatly formatted report file.

[0023] The present invention will be further described below through a detailed embodiment. In one embodiment of the present invention, an intelligent industry report generation platform is used to generate monthly analysis reports for the new energy vehicle industry. The data acquisition module accesses the official website of the Ministry of Industry and Information Technology, the news release pages of major car companies, the battery industry portal website, and the stock exchange bulletin board daily according to pre-configured crawling rules to crawl news, production and sales data, policy documents, and financial reports related to new energy vehicles. The crawled original web pages are parsed using HTML, the main text is extracted, and advertisements are filtered to obtain plain text; then, key information such as company name, model name, sales volume, and release time are extracted from them using regular expressions and named entity recognition models, and this information is stored in the database along with the text blocks. At the same time, the timeliness evaluation subroutine assigns timeliness weights to each piece of data according to the authority of the data source and the release time. For example, official announcements have a higher weight than self-media articles, and the latest data has a higher weight than outdated data. The industry big model analysis module loads a GPT-type large language model that has been fine-tuned with corpus in the field of new energy vehicles, and is equipped with a vector knowledge base containing summaries of industry white papers and mainstream research reports over the years. For the latest text block in the database, the model first performs event clustering and hotspot identification. For example, after detecting "a car company releases a timetable for solid-state battery mass production," it categorizes it as a technological breakthrough event. Subsequently, the model combines background knowledge from the vector library to analyze the potential impact of this event on the competitive landscape, generating the analytical statement "Solid-state battery technology is expected to reshape the ranking of battery companies in the next three years." Simultaneously, the model extracts sales sequences for each brand from the captured monthly sales data, automatically infers market share trends, and provides a predicted description of market concentration for the next quarter. After receiving the analysis conclusions, the report generation module automatically arranges the chapter structure according to the report type (monthly report), including a summary, market dynamics, technological frontiers, and key company tracking. The text generation engine expands the short sentences output by the model into coherent paragraphs. For example, "a certain brand's market share increased" is expanded to "a certain brand benefited from the strong sales of its new model, and its market share increased by two percentage points month-on-month this month, further consolidating its leading position in the segment." The chart generator extracts brand comparison values ​​from sales data, automatically selects clustered bar charts to display sales for each brand, and adds analytical text such as "Brand X has maintained growth for three consecutive months" to the chart captions. Finally, the text and image fusion engine inserts the charts near the relevant paragraphs, uniformly adjusts the font and line spacing, and generates a formatted PDF report. Users can preview the report through the interactive interface and export it as HTML or PPT format as needed.

[0024] The beneficial effects of this embodiment include at least the following: by modularly integrating data collection, large-scale model analysis, and report generation, the entire process from web information scraping to professional report output is automated, significantly shortening the writing cycle of industry analysis reports; at the same time, by utilizing large-scale models finely tuned with industry knowledge and enhanced vector libraries, the analysis conclusions are made more in-depth and timely, overcoming the shortcomings of traditional automation tools that result in superficial content; in addition, the automatic generation function of graphics and text ensures the readability and professionalism of the report, providing users with high-quality decision support information.

[0025] In one possible implementation, the data acquisition module includes a data crawling unit for periodically accessing a preset network data source via a distributed crawler to crawl original web page data; a data cleaning unit for removing noise from the original web page data to obtain cleaned text; an entity recognition unit for extracting key entity information from the cleaned text and storing the key entity information in association with the cleaned text; and a timeliness evaluation unit for scoring the timeliness and credibility of the cleaned text based on the data source, publication time, and frequency of citation, and generating a timeliness weight.

[0026] The data crawling unit can refer to a functional component responsible for batch retrieving web page content from the internet. For example, a distributed crawler cluster built on the Scrapy framework can concurrently access multiple target sites and adhere to the robots.txt protocol. Pre-defined network data sources can refer to a list of websites configured by the platform that need continuous monitoring, such as industry association websites, major company news pages, and financial data websites. Raw web page data can refer to unprocessed HTML documents downloaded by the crawler. The data cleaning unit can refer to a structured processing component used to remove irrelevant content from web pages. For example, it can use regular expressions or templates to filter and remove advertisements, navigation bars, footers, etc., and extract the main text and tables. Cleaned text can refer to plain text content with noise removed. The entity recognition unit can refer to a functional module that uses a named entity recognition model to extract predefined categories of entities from text. For example, it can use the BERT-BiLSTM-CRF model to identify key entities such as company names, personal names, place names, dates, and amounts. Key entity information can refer to the identified entity categories and corresponding text fragments, such as "Company: CATL" and "Date: March 2025". Association storage refers to mapping entity information with information such as the ID and paragraph position of the original text and storing it in a database. Timeliness assessment units can refer to modules that quantify the freshness and credibility of data. For example, this might involve calculating a time decay factor based on the number of days since the data's publication, assigning a basic credibility score based on the authority of the source domain (e.g., .gov, .org), and adjusting the score based on the number of times the content has been cited by other sites. Timeliness weight can refer to a comprehensive score used by subsequent analysis modules to indicate the importance they place on the data.

[0027] Based on the aforementioned new energy vehicle industry report generation platform, this embodiment provides a specific implementation of the data acquisition module. The data crawling unit employs a distributed crawler architecture, deployed in a cloud server cluster, and triggers crawling tasks daily at set times. Preset network data sources include the "Recommended Models Catalog for the Promotion and Application of New Energy Vehicles" page on the Ministry of Industry and Information Technology website, official news blogs of major automakers (such as BYD and Tesla), industry media such as BatteryNet, and announcement sections of the Shanghai Stock Exchange and Shenzhen Stock Exchange. After the crawler retrieves the original webpage, it sends the HTML source code and URL to the data cleaning unit. The data cleaning unit first extracts the main text area using a rule-based HTML parser, then uses a text classifier to filter out noise from comment sections, related readings, etc., ultimately obtaining cleaned text containing the main content. The entity recognition unit loads a named entity recognition model fine-tuned for the industry sector, capable of recognizing seven types of entities, including company names, product models, and financial values. The model infers from each cleaned text segment and outputs an entity list. For example, from the sentence "CATL released its 2024 annual report, with net profit increasing by 15% year-on-year," it extracts "Company: CATL," "Time: 2024," "Indicator: Net Profit," and "Value: 15% Year-on-Year Growth." This entity information is associated with the unique identifier of the text and stored in a structured database for easy retrieval later. A timeliness assessment unit is also activated, calculating a timeliness weight for each text: the weight decreases by a preset decay coefficient for each day the publication date is more than one day from the current date; the highest base weight is assigned to data from government websites or stock exchanges, followed by industry media, and the lowest to personal blogs; the number of times the text content is mentioned on other websites is also counted, with higher mentions resulting in higher credibility. The final timeliness weight value is appended to the text record for reference in larger model analyses; for example, data with higher weights will have a greater impact in trend analysis.

[0028] This embodiment achieves efficient capture and refined processing of massive amounts of network data by refining the internal components of the data acquisition module. The data acquisition unit adopts a distributed architecture to ensure the real-time performance and scalability of the acquisition; the cleaning and entity recognition unit transforms unstructured text into structured information, providing high-quality input for subsequent analysis; the timeliness evaluation unit uses multi-dimensional scoring to ensure that the analysis model prioritizes authoritative and up-to-date data, thereby improving the accuracy and reliability of industry analysis conclusions.

[0029] In one possible implementation, the industry big data model analysis module includes an industry knowledge vector library, which stores text paragraphs extracted from high-quality industry documents and their vectorized representations; a trend identification unit, used to input structured data into a pre-trained language model, identify hot events and trends within the industry, and generate event summaries; a competitive landscape analysis unit, used to analyze the market share changes and competitive dynamics of major companies based on enterprise entity information in structured data, and generate competitive landscape analysis text; a market size prediction unit, used to extract historical statistical information from structured data, combine it with industry growth rates to infer the current market size, and generate future prediction descriptions; and a sentiment and risk assessment unit, used to perform sentiment scoring and risk level assessment on public opinion information, and generate risk analysis conclusions.

[0030] The industry knowledge vector library can refer to a vector database that stores prior knowledge in a specific domain. For example, industry reports and technical standards are segmented into paragraphs, converted into high-dimensional vectors through an embedding model, and then stored for use in retrieval enhancement. High-quality industry documents can refer to professional literature published by authoritative institutions, such as brokerage research reports, government planning documents, and industry association yearbooks. Vectorized representation can refer to numerical vectors mapped from text through a neural network model, which can be used for semantic similarity calculation. The trend recognition unit can refer to a component responsible for capturing industry dynamics from the latest data. For example, it can analyze news clustering results through a large model to extract trends such as "policy subsidy reduction" and "solid-state battery breakthrough." The competitive landscape analysis unit can refer to a module that calculates market share and evaluates competitiveness for specific enterprises. For example, it can calculate the market share of each brand based on sales data and identify changes in ranking. The market size prediction unit can refer to an analysis module that extrapolates future market capacity based on historical data. For example, it can use a time series model combined with expert knowledge to predict sales revenue for the next year. The sentiment and risk assessment unit can refer to a module that identifies the sentiment and risk factors of public opinion content. For example, it can analyze social media discussions to determine whether the public's attitude towards a brand is positive or negative, and identify risk signals such as "overcapacity" and "trade barriers".

[0031] Building upon the aforementioned new energy vehicle industry report generation platform, this embodiment further describes the internal structure and workflow of the industry large-scale model analysis module. An industry knowledge vector library is pre-built: the system collects in-depth reports on the new energy vehicle industry published by mainstream securities firms over the past five years, industry development plans released by the Ministry of Industry and Information Technology, yearbooks from the China Association of Automobile Manufacturers, and financial reports from leading companies. These documents are segmented into paragraphs, and 768-dimensional vectors are generated using the Sentence-BERT model and stored in the Milvus vector database, while retaining the original text paragraph metadata (such as source and year). The trend identification unit receives structured data from the data acquisition module, including daily news texts and announcements. The unit calls the large-scale model to perform topic clustering on batches of text and generates summaries. For example, when multiple news articles mention "multiple provinces and cities have introduced charging pile construction subsidy policies," the model automatically categorizes it as an "infrastructure policy dividend" trend and generates a summary: "Recently, many local governments have released charging pile construction plans, which are expected to drive growth in demand for charging equipment." The competitive landscape analysis unit extracts enterprise entity and sales data from the structured data and uses the large-scale model to calculate the market share dynamics of major automakers. For example, comparing sales data from Q1 2025 and Q4 2024, the model found that a certain emerging brand's market share increased from 8% to 12%, attributing this to accelerated deliveries of new models and generating the text "This brand has achieved a leap in market share with its blockbuster models and is now challenging the top tier." The market size prediction unit aggregates historical monthly sales, import / export data, and industry average growth rates, combining this with the model's understanding of macroeconomic policies to infer the current year's market size and provide forecasts for the next three years. For instance, based on the growth rates of the previous two years and the recent impact of subsidy reductions, the model predicts that new energy vehicle sales will grow by 18% in 2026 and outputs the descriptive statement "The market size is expected to exceed 12 million vehicles." The sentiment and risk assessment unit focuses on public opinion analysis, capturing social media, forums, and news comments. The model assigns a sentiment score (positive, neutral, negative) to each piece of information and identifies risk keywords. For example, the model finds increased discussion about "battery recycling pollution" with a negative sentiment, generating a risk warning: "Increased environmental regulatory risks may affect the short-term valuation of battery recycling companies."

[0032] This embodiment constructs an industry knowledge vector library and multiple professional analysis units, enabling the large model to perform in-depth reasoning by combining domain knowledge, significantly improving the professionalism and accuracy of the analysis conclusions. The trend identification unit ensures the report responds promptly to industry dynamics; the competitive landscape analysis unit provides multi-dimensional enterprise comparisons; the market size forecasting unit gives the report a forward-looking perspective; and the sentiment and risk assessment unit adds risk warning functionality, enabling the generated report to not only describe the current situation but also assist decision-makers in predicting the future.

[0033] In one possible implementation, the trend identification unit, competitive landscape analysis unit, market size prediction unit, and sentiment and risk assessment unit, when generating analysis conclusions, call the vectorized representation in the industry knowledge vector library for comparison and verification, so as to improve the consistency between the analysis conclusions and industry professional terminology.

[0034] In this context, comparative validation refers to each analysis unit retrieving prior knowledge paragraphs from a vector library that are semantically similar to the current input before generating text, and providing them as context to the larger model. This ensures that the output terminology and logic are consistent with industry conventions. Industry-specific terminology can refer to expressions commonly used in benchmark reports, such as "penetration rate," "capacity utilization rate," and "LFP battery." Vector retrieval can guide the model to adopt these terms.

[0035] In the aforementioned platform implementation that includes an industry knowledge vector library, when generating a summary of events related to "solid-state batteries," the trend recognition unit first converts the current news text into a query vector and searches the industry knowledge vector library for the most similar past research report paragraphs. The retrieved paragraphs may contain professional expressions such as "solid-state battery mass production timeline" and "sulfide electrolyte technology route." The model uses this information as part of the prompts, thus correctly using terms such as "solid-state electrolyte" and "energy density" in the summary and avoiding confusion with "semi-solid." Similarly, when analyzing changes in a car company's market share, the competitive landscape analysis unit searches the vector library for past analyses of that car company, referencing terms such as "product matrix" and "channel penetration" to make the generated text more consistent with the language habits of industry analysts. When inferring future growth rates, the market size prediction unit compares the actual values ​​with the predicted values ​​in historical forecast reports in the vector library, adjusting the model's judgment of the current growth rate to improve prediction credibility. When identifying risks, the sentiment and risk assessment unit refers to classic descriptions of "policy risks" and "technology route risks" in the vector library to make risk warnings more targeted.

[0036] This embodiment effectively reduces the arbitrariness of model-generated content by introducing a vector library comparison and verification mechanism, making the conclusions output by each analysis unit closer to the level of industry experts in terms of terminology and logical structure, thus enhancing the professional authority of the report.

[0037] In one possible implementation, the report generation module includes a report structure arrangement unit for automatically planning the report chapter structure based on the industry analysis conclusions and determining the logical order of paragraphs within each chapter; a text generation unit for expanding and refining the industry analysis conclusions to generate a report text with logical depth; a chart generation unit for extracting numerical sequences from the industry analysis conclusions, automatically selecting chart types, and generating visual charts with data labels and analytical annotations; and a text-image fusion unit for associating the report text with the visual charts and formatting them according to the structure planned by the report structure arrangement unit to generate a complete report document.

[0038] The report structure and arrangement unit can refer to a component that automatically determines which chapters a report should include and their order based on the theme and scale of the analysis content. For example, it might automatically insert a "Market Overview" chapter when the analysis includes market size data, or an "Competitive Landscape" chapter when it includes company comparisons. The logical order of paragraphs within a chapter can refer to the arrangement of paragraphs according to a "general-specific-general" or "chronological" logic. The text generation unit can refer to a natural language generation module that expands brief analytical conclusions into complete sentences and paragraphs. For example, it might expand "a company's market share increased" into a paragraph including reasons, impacts, and prospects. The chart generation unit can refer to a component that automatically draws graphs based on numerical data. For example, it might select a line chart for time series data and a pie chart for percentage data, and then label the charts with specific values ​​and add brief analyses. Data labels can refer to the specific values ​​or percentages displayed on the charts. Analytical annotations can refer to concluding text added below or within the charts, such as "Sales reached a record high this quarter." The text-image integration unit can refer to a layout engine that combines text and charts into a coherent document, such as ensuring that charts follow the relevant text and that fonts, margins, and other formatting are consistent.

[0039] In the aforementioned implementation of the new energy vehicle industry report generation platform, the report generation module receives multiple conclusions from the analysis module, including total market sales, company market share, technology trend summaries, and risk warnings. The report structure arrangement unit first evaluates the content: due to the inclusion of sales time series and comparisons of market share among multiple companies, it automatically determines the chapter structure as follows: Summary, Overall Market Situation, Company Competitive Analysis, Technology Trends, and Risk Warnings. Within the "Competitive Analysis" chapter, the arrangement unit determines the paragraph order based on the importance of the companies (e.g., sorted by market share), describing leading companies first, followed by challengers. The text generation unit receives the output from the competitive landscape analysis unit, "A certain brand's market share increased from 8% to 12%," and expands it to: "In the first quarter of 2025, a certain emerging brand significantly increased its market share, from 8% in the previous quarter to 12%, thanks to the strong delivery performance of its latest model. This change allowed it to surpass traditional second-tier brands and approach the top three in the industry, reflecting that consumers' preference for intelligent vehicles is reshaping the market landscape." The chart generation unit extracts the sales data series of each brand for four consecutive quarters from the data, automatically determines that it is suitable to use a line chart to display the trend, and generates a line chart with the horizontal axis representing the quarter and the vertical axis representing sales. Each line represents a brand, and the specific value is labeled next to the data points. At the same time, an analytical note is added to the bottom of the chart: "After two consecutive quarters of decline, the sales of a certain brand stabilized and rebounded this quarter." The text and chart fusion unit combines this text and the line chart, ensuring that the chart is below the paragraph describing the trend, and adjusts the page layout so that the text and chart are not separated. Finally, the system outputs a complete PDF report including a cover, table of contents, and main text.

[0040] This embodiment achieves intelligent transformation from structured analysis results to the final document through the subdivided units of the report generation module. The report structure arrangement ensures logical clarity, text generation improves language quality, chart generation enhances data visualization, and the integration of text and graphics ensures aesthetically pleasing layout, allowing users to directly obtain ready-to-use reports and significantly reducing the workload of later editing.

[0041] In one possible implementation, the chart generation unit automatically selects a chart type, including a pie chart, line chart, bar chart, radar chart, or bubble chart, and adapts it according to the data type.

[0042] Chart type refers to different graphic styles used to visualize data. Pie charts are suitable for showing the proportion of components, line charts are suitable for showing time series trends, bar charts are suitable for comparing different categories of values, radar charts are suitable for displaying multi-dimensional indicators, and bubble charts are suitable for displaying three-dimensional data relationships. Data type matching refers to automatically matching the most suitable chart based on the dimensions, units, and analytical purpose of the data.

[0043] Building upon the aforementioned report generation module, the chart generation unit incorporates a rule-based decision engine. When the input data is the market share of each company at a single point in time (summing up to 100%), the engine determines it as proportional data and automatically selects a pie chart for display. When the input data is sales figures for multiple consecutive quarters, containing at least three time points, the engine selects a line chart to show the trend. When the input data is a comparison of different companies' values ​​on the same indicator (such as R&D investment), the engine selects a bar chart. When the input data contains multiple dimensions (such as brand ratings in performance, price, and reputation), the engine selects a radar chart to display overall capabilities. When the input data contains three dimensions (such as company market share, growth rate, and profit margin), the engine selects a bubble chart, using the X-axis, Y-axis, and bubble size to represent the three dimensions respectively. For example, when analyzing the competitiveness of battery companies, the system uses each company's market share, compound annual growth rate over the past three years, and gross profit margin as three-dimensional data to generate a bubble chart that visually displays each company's position, facilitating the identification of high-growth, high-profit star companies.

[0044] This embodiment automatically adapts to different chart types, ensuring that data is presented in the most intuitive way, improving the readability of reports and the efficiency of analysis, while avoiding the tedious steps of users manually selecting charts.

[0045] In one possible implementation, a user interaction module is also included, which receives the industry analysis topic, time range, and key focus areas set by the user, and passes the user settings to the data acquisition module to adjust the data capture strategy, and to the report generation module to customize the report content.

[0046] The user interaction module can refer to components that provide a graphical interface or API for users to input parameters and start tasks, such as a web front-end or command-line interface. The industry analysis topic can refer to a user-specified industry sector or specific topic, such as "new energy vehicle charging piles." The time range can refer to the start and end dates of the data coverage, such as "January 2024 to March 2025." The key focus area can refer to a specific link in the industry chain, such as "battery manufacturing" or "vehicle sales." Adjusting the data crawling strategy can refer to expanding or narrowing the list of seed URLs for the crawler based on the topic, or adjusting the crawling frequency.

[0047] Building upon the aforementioned platform, a user interaction module has been added. Users access the platform's homepage through a browser, enter the analysis topic "Solid-State Battery" in the form, select "Full Year 2024" for the time range, and choose "Technology R&D and Patents" as the key focus area. After the user clicks submit, the interaction module encapsulates the parameters into task instructions and sends them to the data acquisition module. The data acquisition module updates its crawler configuration accordingly, increasing its crawling of solid-state battery-related technical blogs, patent databases, and academic preprint websites, while also increasing the frequency of access to these source sites. When processing data, the analysis module also prioritizes content involving keywords such as "solid-state battery," "electrolyte," and "patent licensing." The report generation module, based on the key focus areas, strengthens the technology R&D section in the report while de-emphasizing the market sales section, making the final report more tailored to user needs.

[0048] This embodiment enables personalized customization of report generation through a user interaction module, allowing users to focus on specific areas or processes, avoiding information redundancy in generalized reports, and improving the platform's adaptability to different user scenarios.

[0049] In one possible implementation, the user interaction module is also used to provide a report preview interface, receive user instructions to modify the report content, and record the modification behavior as feedback data.

[0050] The report preview interface refers to the online viewing page displayed to users after the report is generated, where users can edit text, adjust chart positions, or delete paragraphs. Editing instructions refer to user operations on text, such as modifying the description of a data point or changing the chart type. Feedback data refers to the recorded user modification history, used for subsequent optimization of the model or template.

[0051] In the aforementioned platform that includes a user interaction module, after the report is generated, the system pushes an HTML preview version to the user. Users can directly click on paragraphs in their browser to edit them, such as changing a sentence to a more accurate expression or dragging and dropping charts to different locations. After making changes, the user clicks "save," and the interaction module records the differences between the modified content and the original version, storing it in the feedback database. This feedback data can be used to retrain the text generation model, making it more aligned with users' expression preferences; it can also be used to adjust report layout rules, for example, if multiple users move the position of a certain type of chart, the system may default to placing that type of chart at the beginning.

[0052] This embodiment introduces a user interaction and feedback mechanism, enabling the platform to continuously learn and adapt to user habits, constantly optimize the quality of generated reports, and form a virtuous cycle of human-machine collaboration.

[0053] In one possible implementation, the user interaction module supports exporting the generated report as PDF, HTML, or PPT format, and adapting the layout according to the selected format.

[0054] The export format refers to the file type of the final report obtained by the user. PDF is suitable for printing and distribution, HTML is suitable for web viewing, and PPT is suitable for presentations. Adaptive layout refers to adjusting the page layout according to the characteristics of different formats, such as fixed page breaks in PDF, one core point per page in PPT, and responsive design in HTML.

[0055] In the aforementioned user interaction module, an export button is located at the bottom of the report preview interface, allowing users to choose between PDF, HTML, or PPT. If PDF is selected, the system uses wkhtmltopdf to convert the HTML preview into a fixed-page PDF file, ensuring that font embeddings and hyperlinks are preserved. If HTML is selected, the system generates a single-file HTML document containing all text and images, using responsive CSS for smooth reading on mobile devices. If PPT is selected, the system breaks the report content down into multiple slides by chapter, each slide containing a title, a key text paragraph, and a chart, applying a unified theme template for easy use in meetings.

[0056] This embodiment meets users' needs in different scenarios by exporting in multiple formats and adapting the layout, thus improving the practicality and flexibility of the report.

[0057] In one possible implementation, the user interaction module is also used to automatically save each generated report version and build an index to support historical report tracing and comparison.

[0058] Automatic saving refers to the system storing the complete report each time a user generates or modifies it. Indexing refers to a search directory built according to dimensions such as time, topic, and user. Tracing and comparison allow users to view reports from a past point in time and compare them side-by-side with the current report to observe changes in data and analysis conclusions.

[0059] In the aforementioned user interaction module, each time a user generates or modifies a report and saves it, the system packages and stores the final version of the report, the generation time, the parameters used, and the user's modification history into the version database, generating a unique version number. Users can view all versions in chronological order on the history page, and click on any version to preview the report at that time. Simultaneously, the comparison function allows users to select two versions, and the system will display them side-by-side, highlighting the differences in text and changes in chart data, such as the increase or decrease in the height of the market share bar chart. This helps users track industry evolution or evaluate the effectiveness of improvements to analytical models.

[0060] This embodiment provides users with a tool for long-term tracking of industry changes through version management functions, while facilitating auditing and backtracking analysis processes, thereby enhancing the platform's reliability and interpretability.

[0061] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0062] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0063] Building upon the aforementioned embodiments, this platform further implements an end-to-end closed-loop technical process from data acquisition to report construction. This closed-loop process sequentially executes six stages: data acquisition, data structuring, knowledge enhancement, model reasoning, result generation, and report construction. Feedback paths are established between each stage, enabling the output of subsequent stages to back-optimize the parameter configuration of preceding stages, thereby forming an adaptive iterative optimization system.

[0064] In the data acquisition phase, the data acquisition module retrieves raw webpage data in parallel from multiple network data sources based on a preset industry theme and the time range provided by the user interaction module. It records the acquisition time, source domain, and network response status for each data entry. The output of this phase is the raw webpage data stream.

[0065] In the data structuring stage, the data cleaning unit extracts the main text, filters noise, and standardizes the format of each original webpage, resulting in cleaned text. The entity recognition unit extracts key entity information such as company names, product names, numerical indicators, and time points from the cleaned text and associates and stores these entities with text paragraphs. The timeliness assessment unit calculates a timeliness weight for each text based on the authority of the source, publication time, and frequency of citation. The output of this stage is structured data with entity tags and timeliness weights.

[0066] In the knowledge enhancement stage, the industry big data model analysis module calls upon a pre-built industry knowledge vector library to convert key query terms or paragraphs in the structured data into vectors, and retrieves several high-quality industry knowledge paragraphs with similar semantics from the vector library. The retrieved knowledge paragraphs are then concatenated with the original structured data to form enhancement prompts, serving as contextual input for subsequent model inference. The core function of this stage is to compensate for potential terminology gaps or insufficient background information in the original data, ensuring that the model can rely on authoritative prior industry knowledge during inference.

[0067] In the model inference phase, the pre-trained language model, fine-tuned with industry knowledge, receives knowledge-enhanced prompts and sequentially completes multiple inference tasks, including trend identification, competitive landscape analysis, market size prediction, and sentiment and risk assessment. The model outputs multiple structured analytical conclusions, each accompanied by a confidence score and the identifier of the original data source cited.

[0068] In the results generation stage, the text generation unit in the report generation module expands the brief analytical conclusions produced by the model inference into logically complete report paragraphs; the chart generation unit extracts numerical sequences from the analytical conclusions, automatically adapts the chart type, and generates visual charts with data labels and analytical annotations.

[0069] In the report construction phase, the report structure arrangement unit dynamically plans the order of report chapters and the logical order of paragraphs within each chapter based on the content type and quantity of the analysis conclusions; the text and image integration unit integrates and typesets the generated text paragraphs and charts according to the arrangement structure, and adds a cover, table of contents and reference notes, finally outputting a complete report document.

[0070] The aforementioned closed-loop process also includes multiple feedback paths to enhance the system's adaptability. First, after the report is built, the user interaction module receives user modification instructions for the preview report and sends the modified content back as feedback data to the result generation stage. This is used to adjust the extended style of the text generation unit and the type selection preference of the chart generation unit. Second, analysis conclusions generated in the model inference stage with confidence scores below a preset threshold are fed back to the data acquisition stage, triggering supplementary crawling tasks for relevant keywords to obtain more evidence data before re-executing the inference. Third, if knowledge paragraphs retrieved in the knowledge enhancement stage semantically conflict with the original data, the system marks the conflict and feeds it back to the data structuring stage, prompting the entity recognition unit to update its recognition rules or vocabulary. Through these feedback paths, the entire system forms a continuously self-optimizing closed loop. Each report generation process accumulates experience data, continuously improving the quality and adaptability of subsequent reports.

[0071] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent industry report generation platform that integrates large-scale industry models and web data scraping, characterized in that, include: The data acquisition module is used to capture raw data related to the industry from multiple network data sources and preprocess the raw data to obtain structured data. The industry big data model analysis module is used to analyze the structured data based on a pre-trained language model fine-tuned with industry knowledge and generate industry analysis conclusions. The report generation module is used to automatically generate an industry report containing textual descriptions and visual charts based on the industry analysis conclusions.

2. The platform according to claim 1, characterized in that, The data acquisition module includes a data crawling unit, which is used to access a preset network data source at regular intervals through a distributed crawler to crawl raw web page data; and a data cleaning unit, which is used to remove noise information from the raw web page data to obtain cleaned text. An entity recognition unit is used to extract key entity information from the cleaned text and associate and store the key entity information with the cleaned text; a timeliness evaluation unit is used to score the timeliness and credibility of the cleaned text based on the data source, publication time, and citation frequency, and generate a timeliness weight.

3. The platform according to claim 1, characterized in that, The industry big data model analysis module includes an industry knowledge vector library, which stores text paragraphs extracted from high-quality industry documents and their vectorized representations; a trend recognition unit, used to input the structured data into the pre-trained language model, identify hot events and trends within the industry, and generate event summaries; a competitive landscape analysis unit, used to analyze the market share changes and competitive dynamics of major companies based on the enterprise entity information in the structured data, and generate competitive landscape analysis text; and a market size prediction unit, used to extract historical statistical information from the structured data, combine it with industry growth rates to infer the current market size, and generate future prediction descriptions. The sentiment and risk assessment unit is used to score public opinion information on sentiment and assess its risk level, generating risk analysis conclusions.

4. The platform according to claim 3, characterized in that, When generating analysis conclusions, the trend identification unit, competitive landscape analysis unit, market size prediction unit, and sentiment and risk assessment unit call the vectorized representation in the industry knowledge vector library for comparison and verification, so as to improve the consistency between the analysis conclusions and industry professional terminology.

5. The platform according to claim 1, characterized in that, The report generation module includes a report structure arrangement unit, which is used to automatically plan the report chapter structure based on the industry analysis conclusions and determine the logical order of paragraphs within each chapter; and a text generation unit, which is used to expand and refine the industry analysis conclusions to generate a report text with logical depth. The chart generation unit is used to extract numerical sequences from the industry analysis conclusions, automatically select the chart type, and generate visual charts with data labels and analytical annotations. The text and image fusion unit is used to associate the report text with the visualization charts and to arrange the layout according to the structure planned by the report structure arrangement unit to generate a complete report document.

6. The platform according to claim 5, characterized in that, The chart generation unit automatically selects one of the following chart types: pie chart, line chart, bar chart, radar chart, or bubble chart, and adapts it according to the data type.

7. The platform according to claim 1, characterized in that, It also includes a user interaction module, which receives the industry analysis topic, time range and key focus areas set by the user, and passes the user settings to the data acquisition module to adjust the data capture strategy, and to the report generation module to customize the report content.

8. The platform according to claim 7, characterized in that, The user interaction module is also used to provide a report preview interface, receive user instructions to modify the report content, and record the modification behavior as feedback data.

9. The platform according to claim 7, characterized in that, The user interaction module supports exporting the generated report as PDF, HTML, or PPT format, and adapts the layout according to the selected format.

10. The platform according to claim 7, characterized in that, The user interaction module is also used to automatically save each generated report version and build an index to support the tracing and comparison of historical reports.

Images