A page data generation method, device, equipment and readable storage medium

Abstract

This invention discloses a page data generation method, apparatus, device, and readable storage medium, applied in the field of data processing technology. The method includes: acquiring a preset number of intelligent agents corresponding to the current business scenario, and prompt words corresponding to each intelligent agent; wherein the intelligent agents are obtained by decomposing the linear page generation process based on the current business scenario; determining the target data proportion corresponding to each intelligent agent according to requirements; and generating page data using the preset number of intelligent agents based on the target data proportion and prompt words to obtain target page data. Compared to the current method where the generation, filtering, and correction of each piece of data requires a large amount of manual effort, this application constructs a method based on multi-agent collaboration. Based on multiple intelligent agents corresponding to the linear page generation process, it can automatically and efficiently generate page data that conforms to business logic and rules, is precisely controllable in style and architecture, and has controllable quantity and type.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and readable storage medium for generating page data. Background Technology

[0002] Currently, the main technical means for page data synthesis and generation in the industry can be summarized into the following three categories: First: manual data construction, which relies entirely on manual methods to create data from scratch; Second: modular platform synthesis, which synthesizes page structure and data by combining predefined modules. The results generated by this platform are often only a draft, and a large number of developers are still needed for manual fine-tuning and secondary development; Third: code generation applications / platforms, which use applications focused on code generation or low-code / no-code generation platforms on the Web (Internet) to transform high-level descriptions into specific page code, thereby achieving page data synthesis, but still requires a lot of front-end modification work.

[0003] It is evident that improving the efficiency of page data generation is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a page data generation method, apparatus, device and readable storage medium, which solves the technical problem of low page data generation efficiency in the prior art.

[0005] To solve the above-mentioned technical problems, the present invention provides a page data generation method, comprising: Obtain a preset number of intelligent agents corresponding to the current business scenario, and the prompt words corresponding to each intelligent agent; wherein, the intelligent agent is a large language model obtained by decomposing the page linear generation process based on the current business scenario; Determine the target data percentage for each agent based on requirements; Based on the target data ratio and the prompt words, page data is generated using the preset number of intelligent agents to obtain the target page data.

[0006] Optionally, before obtaining the preset number of intelligent agents corresponding to the current business scenario and the prompt words corresponding to each intelligent agent, the following steps are also included: The linear page generation process based on the current business scenario is broken down to obtain the intelligent agent and the corresponding data structure content of each intelligent agent; Each agent is split according to the data structure content to obtain a sub-agent corresponding to each agent; wherein, the agent includes the sub-agent.

[0007] Optionally, the linear page generation process based on the current business scenario is broken down to obtain the intelligent agent and the data structure content corresponding to each intelligent agent, including: Based on the current business scenario, the page linear generation process is broken down into a requirement draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent; The data structure content corresponding to the intelligent agent that parses the requirements draft includes content, user requirements, and news events; The data structure content corresponding to the line drawing design agent includes the main title of the page, the subtitle of the page, the card content, and the line drawing page; The data structure content corresponding to the header image generating agent includes the prompt words used to generate the header image and the address after the header image is generated; The data structure content corresponding to the visual enhancement agent includes color planning, card-in-card materials, and the final page. The data structure content corresponding to the project code generation agent includes the multi-turn dialogue content during generation and the final project generation address.

[0008] Optionally, based on the target data proportion and the prompt words, page data is generated using the preset number of intelligent agents to obtain target page data, including: Based on user needs, the intelligent agent is used to analyze the requirements draft and combine it with news source search for news events to form a target requirements draft; Based on the target requirements draft, the line drawing design agent is used to generate line drawing data including the target main title, target subtitle, target card content, and target line drawing page. The header image generation agent extracts the main title and subtitle from the line drawing data to form the header image and color scheme. Based on the header image and color scheme, the visual enhancement agent is used to perform visual processing on the target line drawing page to form a page preview draft. The project code is used to generate an intelligent agent that converts the preview image into the target page data.

[0009] Optionally, each agent can be split according to the data structure content to obtain sub-agents corresponding to each agent, including: The requirement draft parsing agent is divided into a text generation sub-agent, a tag generation sub-agent, and a key information extraction sub-agent.

[0010] Optionally, after determining the target data percentage for each agent based on requirements, the following may also be included: Determine the amount of original data for each agent, the proportion of the target data for each agent, and the data expansion factor for each agent; Based on the original data volume and the data expansion factor, determine the minimum number of agents after expansion for each agent; The minimum total capacity of the total dataset is determined based on the minimum number of data agents after expansion and the proportion of target data for each agent. The initial target data generation amount for each agent is determined based on the minimum total capacity, the target data proportion, and the minimum quantity after expansion for each agent. The final target data generation amount for each intelligent agent is determined based on the initial target data generation amount and the target data proportion corresponding to each intelligent agent. Accordingly, based on the target data proportion and the prompt words, page data is generated using the preset number of intelligent agents to obtain target page data, including: Based on the final target data volume and the prompt words corresponding to each agent, page data is generated using the preset number of agents to obtain the target page data.

[0011] Optionally, after generating the target page data using the preset number of intelligent agents based on the target data proportion and the prompt words, the method further includes: Determine the actual data percentage corresponding to each agent in the target page data; The actual data proportion is compared with the target data proportion to identify the data types that have not reached the predetermined quantity and to determine the intelligent agents corresponding to the data types that need to be supplemented. The target amount of supplementary data is determined based on the actual data percentage and the target data percentage. Based on the target amount of supplementary data, data is generated using the agents corresponding to the data types to be supplemented, until the amount of data corresponding to each agent meets the set requirements.

[0012] The present invention also provides a page data generation apparatus, comprising: The agent determination module is used to obtain a preset number of agents corresponding to the current business scenario, and the prompt words corresponding to each agent; wherein, the agent is a large language model obtained by decomposing the page linear generation process of the current business scenario; The target data percentage determination module is used to determine the target data percentage for each agent according to requirements. The target page data generation module is used to generate page data based on the target data ratio and the prompt words, using the preset number of intelligent agents to obtain the target page data.

[0013] The present invention also provides a page data generation device, comprising: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the page data generation method described above.

[0014] The present invention also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the page data generation method described above.

[0015] The present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the above-described page data generation method.

[0016] As can be seen, this invention obtains a preset number of intelligent agents corresponding to the current business scenario, as well as prompt words corresponding to each intelligent agent; wherein, the intelligent agents are obtained by decomposing the page linear generation process based on the current business scenario; the target data proportion corresponding to each intelligent agent is determined according to the requirements; based on the target data proportion and prompt words, page data is generated using the preset number of intelligent agents to obtain the target page data. Compared with the current method where the generation, filtering and correction of each piece of data requires a lot of manual effort, this application constructs a method based on multi-agent collaboration, which, based on multiple intelligent agents corresponding to the page linear generation process, can automatically and efficiently generate page data that conforms to business logic and rules, is precisely controllable in style and architecture, and has controllable quantity and type.

[0017] In addition, the present invention also provides a page data generation apparatus, device and readable storage medium, which also have the above-mentioned beneficial effects. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a page data generation method provided in an embodiment of the present invention; Figure 2 An overall process specification and data structure diagram provided for embodiments of the present invention; Figure 3 A flowchart illustrating a page data generation method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a page data generation device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a page data generation device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Some terms that appear in the description of the embodiments of this application are subject to the following interpretation: Large Language Models (LLMs) are artificial intelligence models designed to understand and generate human language. They are trained on massive amounts of text data and can perform a wide range of tasks, including text summarization, translation, and sentiment analysis. LLMs are characterized by their sheer size, containing billions of parameters that help them learn complex patterns in language data. These models are typically based on deep learning architectures, such as transformers, which contributes to their impressive performance across various NLP tasks.

[0022] Website banners: These typically appear horizontally on web pages, with the most common sizes being 468x60 and 468x80, and a larger size of 728x90 is also currently available. They are usually placed below the navigation bar to attract user attention, highlight the promotional theme, and drive traffic to event or product pages. Most are static images in PNG (Portable Web Graphics) or JPG (Joint Image Experts Group) formats, but some also include animated elements such as GIF (Graphics Interchange Format) and SWF (Animation Format).

[0023] Large-scale intelligent agents: A large-scale intelligent agent is an intelligent software entity built on a large-scale pre-trained language model, possessing autonomous perception, decision-making, and execution capabilities. This agent interacts with users, the environment, or other systems through natural language, can understand complex intentions, plan and execute multi-step tasks, and achieve continuous performance optimization and adaptation during the interaction process through context learning, tool invocation, and environmental feedback.

[0024] Hallucination refers to the phenomenon where content generated by large language models or other generative AI models appears reasonable and fluent, but is actually inconsistent with the provided source information, contradicts facts, or is purely fabricated. Hallucination is a byproduct of the inherently probabilistic generation of large models and the limitations of training data. The longer the output, the higher the probability that the model will output hallucinatory content.

[0025] Please refer to Figure 1 , Figure 1 A flowchart illustrating a page data generation method provided in an embodiment of the present invention. The method may include: S101, obtain a preset number of intelligent agents corresponding to the current business scenario, and the prompt words corresponding to each intelligent agent; wherein, the intelligent agent is obtained by splitting the page linear generation process based on the current business scenario.

[0026] Each step in this embodiment can be executed by a designated electronic device, which can be a server, portable terminal, or other form. This electronic device contains memory modules, the specific number of which is not limited. This embodiment does not limit specific business scenarios. For example, the business scenario in this embodiment can be a vertical business scenario such as finance, education, or healthcare. This embodiment does not limit specific quantities; the preset quantities may differ for different business scenarios. This preset data is mainly determined based on the number of steps in the current business scenario's linear page generation process. The intelligent agent in this embodiment is a large language model, which can be GPT (Generative Pre-trained Transformer), DeepSeek, or Wenxin Yiyan, etc. The linear page generation process in this embodiment refers to the logical evolution path of transforming unstructured requirements into structured, renderable final page data. For example, by deeply analyzing the actual vertical page production work scenario, the traditional or ideal complete generation process can be decomposed and modeled into a pipeline where multiple professional intelligent agents collaborate sequentially. These intelligent agents can include at least: a requirements draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent. This step laid the architectural foundation for the division of labor and collaboration throughout the system.

[0027] It should be further noted that, based on any of the above embodiments, in order to improve the accuracy of agent construction, before obtaining the preset number of agents corresponding to the current business scenario and the prompt words corresponding to each agent, the following may also be included: S01, based on the current business scenario, the page linear generation process is broken down to obtain intelligent agents and the corresponding data structure content of each intelligent agent; S02, split each agent according to the data structure content to obtain the sub-agents corresponding to each agent; where an agent includes sub-agents.

[0028] In this embodiment, the intelligent agents and data structures obtained from different business scenarios vary. The data structures corresponding to different intelligent agents also differ. This embodiment allows for the classification and identification of different intelligent agents, and the addition of necessary features to the data structures of different intelligent agents. For each defined intelligent agent, based on the business characteristics of its responsible sub-tasks, its internal functions are further decomposed and restructured. For example, the requirement draft parsing intelligent agent is further refined into a text generation module, a tag generation module, and a key information extraction module. This deep refinement ensures a high degree of focus and optimization for each processing unit, guaranteeing the generation of high-quality, structured data.

[0029] It should be further explained that, in order to improve the accuracy of the decomposition, the above-mentioned linear page generation process based on the current business scenario is decomposed to obtain intelligent agents and the corresponding data structure content for each intelligent agent, which may include: S011, based on the current business scenario's linear page generation process, is broken down into a requirement draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent.

[0030] In this embodiment, the intelligent agents can include at least: a requirements draft analysis agent, a line art design agent, a header image generation agent, a visual enhancement agent, and a project code generation agent. This step lays the architectural foundation for the division of labor and collaboration in the entire system. The functions of each intelligent agent are as follows: Requirements Draft Analysis Agent: Responsible for analyzing the original requirements and generating a structured requirements outline by calling keywords and web query APIs (interfaces). Line Art Design Agent: An agent that builds the basic framework of the landing page based on the requirements draft and user needs. Header Image Generation Agent: Generates a header banner image that matches the theme and scene of the landing page based on the main / subtitles and descriptions in the requirements draft. Style Enhancement Agent: Analyzes the color scheme and visual style of the header image, designs a complete color scheme, and intelligently redraws and colors the line art. Project Code Generation Agent: Reconstructs the generated HTML (Hypertext Markup Language) code into a modern front-end scaffolding project to improve code maintainability and secondary development efficiency.

[0031] S012, Determine the data structure content corresponding to the intelligent agent in the requirements draft analysis, including content, user requirements, and news events.

[0032] In this embodiment, the requirement draft is a natural language requirement of the user's own needs, including the following: Content: The content result generated by the requirement draft agent or the operational requirements obtained from the project operation department and uploaded; User requirements: The requirements input by the user in the requirement draft; News events: Relevant hot events searched from external sources.

[0033] S013, determine the data structure content corresponding to the line drawing design agent, including the main title of the page, the subtitle of the page, the card content, and the line drawing page.

[0034] In this embodiment, the line art is a design draft of the overall landing page based on the user's requirements. It includes the following: Main title: The main title of the page; Subtitle: The subtitle of the page; CTA: An action CTA (a conditionally rendered component) button and its text information (if empty, the button is not present); Card content: Includes each card's serial number, card title, the address of the filled card, style template, and the individual HTML code for the filled card. The line art page is the complete line art drawing assembled with the main and subtitles and the content of each card, according to the wireframe requirements.

[0035] S014, Determine the data structure content corresponding to the header image generating agent, including the prompt words used to generate the header image and the address after the header image is generated.

[0036] In this embodiment, the header image is generated based on the title and content of the requirements draft, and includes the following content: prompt words: prompt words used to generate the header image after preprocessing based on the line drawing and the content of the requirements draft; URL (Uniform Resource Locator): The address of the generated header image; base64 (Binary-to-Text Encoding Scheme): The base64 information converted from the header image.

[0037] S015, determine the data structure content corresponding to the visual reinforcement agent, including color planning, card materials, and the final page.

[0038] This embodiment uses a visual enhancement agent to perform visual enhancement. Visual enhancement involves extracting a color scheme from the header image and line art, then coloring the entire page and filling in the page's elements. This includes: Color Scheme: The color scheme extracted based on the header image's style. Card Elements: The process of filling the page with elements designed according to the card style, including images, SVG (Scalable Vector Graphics) icons, and dynamic components. Final Page: The final design draft assembled by combining the HTML and header image, enhancing the style, and then completing the design.

[0039] S016, Determine the data structure content corresponding to the intelligent agent that generates the project code, including the multi-turn dialogue content during generation and the final project generation address.

[0040] In this embodiment, code generation involves inputting a visually enhanced HTML page into the project code generation agent to generate a corresponding dynamic front-end project. This project includes the following: Generation process: the multi-turn dialogue content during generation; Project address: the final generated project address. The entire data structure is defined as follows: { "Requirements":{ Content: "# Subject: XX Bureau Releases List of 'X Drugs'!...", "Additional Information": { User Request: "Please provide a landing page for user acquisition that incorporates trending news..." "News Event": "Event 1: The State Administration of Drug Administration releases the first edition of the 'Drug Catalog'\nRelated News:..." } } Line art: { Main title: "XX Bureau Releases List of 'X Drugs'", Subtitle: "...", "CTA": "..." "Card content": { "Header material address": "http: / / ..." "Card Filling":[ { "Serial Number": "1", Title: "XX Type of Enterprises Will See a Surge in Prices..." "Card address after filling": "http: / / ...", "Style Template": "<!DOCTYPE html> \n ...", Card Contents:<!DOCTYPE html> \n ..." }, ... ] }, "Line art page": "<!DOCTYPE html> \n\n..." }, "Header Image": { "Prompt": "Please generate a header image for me based on the following theme:\nMain title:...", "url": "...", "base64": "..." }, "Visual Enhancement": { Color Planning: ### I. Color Extraction and Proportion Allocation #### ...", "In-card materials": { "picture": [ { Prompt: "Please provide me with an image based on the original title and content. Main title: ... " "url": "...", "base64": "..." }, ... ], "SVG icon": [ { Prompt: "Please provide me with an SVG icon based on the original title and content. Main title:..." "url": "...", "base64": "..." }, ... ], "Dynamic Components": [ { "Prompt": "Please provide me with a dynamic component based on the original title and content". "content": "..." }, ] }, Final Page: "..." }, "Code generation": { "Generation process": [ { "Role": "system", "content": "..." }, { "role": "user", "content": "..." }, ... { "Role": "assistant", "content": "..." }, ] Project address: http: / / ... }

[0041] It should be further explained that, based on any of the above embodiments, the above-mentioned S103, based on the target data ratio and prompt words, uses a preset number of intelligent agents to generate page data to obtain target page data, may include: S1031, Based on user needs, use the intelligent agent to analyze the requirements draft, and combine it with news source search for news events to form a target requirements draft; S1032, Based on the target requirements draft, use line art to design an intelligent agent to form line art data including the target main title, target subtitle, target card content, and target line art page; S1033, using a header image generation agent to extract the main title and subtitle from line drawing data, forming a header image and color scheme; S1034, Based on the header image and color scheme, a visual enhancement agent is used to visually process the target line drawing page to form a page preview draft; S1035 uses project code to generate an intelligent agent that converts the preview draft into target page data.

[0042] For easier understanding, please refer to Figure 2 , Figure 2 To provide an overall process specification and data structure diagram for embodiments of the present invention, the following process specification was created: First, the requirement draft parsing agent will generate a requirement draft based on user needs and news events searched from news sources. The line art design agent will generate line art data for the main title, subtitle, action CTA button, card content, and the line art HTML page based on the requirement draft. The header image generation agent will extract the main title and subtitle from the request draft to generate the header image and color scheme. The visual enhancement agent will visually enhance the line art HTML page by combining the header image color scheme and style information, generating an HTML preview. Finally, the project code generation agent will convert the HTML preview into a Vite / React (a React (core code generation target framework) front-end project built on Vite (a front-end build tool)) front-end project for subsequent adjustments and release.

[0043] It should be further explained that, based on any of the above embodiments, the above-mentioned splitting of each agent according to the data structure content to obtain the corresponding sub-agents can include: splitting the requirement draft parsing agent into a text generation sub-agent, a tag generation sub-agent, and a key information extraction sub-agent. This embodiment does not limit the specific method of splitting, such as type splitting: splitting the data into multiple sub-agents according to the type, level, and dependency relationship of the data, so as to process and generate data in a more detailed manner. For example, a complex multi-agent can be split into multiple sub-functional modules, each generating data independently. Hierarchical analysis: considering the hierarchical relationship of the agents, the top-level agent and its sub-agents are split to ensure that the agents at each level are fully covered. Through this splitting step, the agents can be operated and controlled more finely, which is convenient for subsequent data processing. For ease of understanding, the code data structure after splitting can be as follows: { Type: "Request Draft" "Requirements":{ Content: "# Subject: XX Bureau Releases List of 'X Drugs'!...", "Additional Information": { User Request: "Please provide a landing page for user acquisition that incorporates trending news..." "News Event": "Event 1: The State Administration of Drug Administration releases the first edition of the 'Drug Catalog'\nRelated News:..." }, } } ... { Type: Line Art "Line art": { Main title: "XX Bureau Releases List of 'X Drugs'", Subtitle: "...", "CTA": "..." "Card content": { "Header material address": "http: / / ..." "Card Filling":[ { "Serial Number": "1", Title: "XX Type of Enterprises Will See a Surge in Prices..." "Card address after filling": "http: / / ...", "Style Template": "<!DOCTYPE html> \n ...", Card Contents:<!DOCTYPE html> \n ..." }, ... ] }, "Line art page": "<!DOCTYPE html> \n\n..." } } ... { Category: "Header Image" "Header Image": { "Prompt": "Please generate a header image for me based on the following theme:\nMain title:...", "url": "...", "base64": "..."} } ... { Category: Visual Enhancement "Visual Enhancement": { "Line art page": "<!DOCTYPE html> \n\n..." "Header Image":{ "url": "...", "base64": "..." }, Color Planning: ### I. Color Extraction and Proportion Allocation #### ...", "In-card materials": { "picture": [ { Prompt: "Please provide me with an image based on the original title and content. Main title: ... " "url": "...", "base64": "..." }, ... ], "SVG icon": [ { Prompt: "Please provide me with an SVG icon based on the original title and content. Main title:..." "url": "...", "base64": "..." }, ... ], "Dynamic Components": [ { "Prompt": "Please provide me with a dynamic component based on the original title and content". "content": "..." }, ... ] }, Final Page: "<!DOCTYPE html> \n" } } ... { Category: Code Generation "Code generation": { "Generation process": [ { "Role": "system", "content": "..." }, { "role": "user", "content": "..." }, { "Role": "assistant", "content": "..." }, ... ] Project address: http: / / ... } }

[0044] S102, determine the target data percentage for each agent based on requirements.

[0045] This embodiment determines the target data proportion to ascertain the data generation ratio and quantity target that each agent needs to undertake in the overall task. This step is crucial for translating diversity planning into concrete executable parameters, guiding the subsequent generation process and ensuring the structural balance of the dataset. This embodiment requires statistical analysis of different types of data (one type per agent) to obtain their quantity distribution. This step helps clarify the data distribution and, for subsequent calculations based on the target data proportion, determines and accurately calculates the data generation ratio and quantity target that each agent in the pipeline needs to undertake in the overall task.

[0046] It should be further noted that, based on any of the above embodiments, in order to improve the accuracy of page data generation, after determining the target data proportion corresponding to each intelligent agent according to requirements, the following may also be included: Step 1: Determine the amount of original data for each agent, the proportion of target data for each agent, and the data expansion factor for each agent.

[0047] Based on the statistically derived distribution, developers can customize the data generation ratio for each agent and scenario in this embodiment. Dataset collection: Input parameters: Indicates the first The number of original records (original data volume) of the data type; Indicates the first The target proportion of class data (target data proportion); : indicates the first The expansion factor of class data.

[0048] Step 2: Determine the minimum number of agents after expansion based on the original data volume and the data expansion factor.

[0049] This embodiment determines the minimum quantity that each type of data must reach after expansion in order to ensure the diversity of that type of data. .

[0050] Step 3: Determine the minimum total capacity of the dataset based on the minimum number of agents after expansion and the proportion of target data for each agent.

[0051] In this embodiment, to simultaneously satisfy the thresholds for all categories, it is necessary to find the category with the most severe restrictions and use it as a benchmark to deduce the required total dataset size. Therefore, the minimum total capacity satisfies... .

[0052] Step 4: Determine the initial target data generation amount for each agent based on the minimum total capacity, the target data ratio, and the minimum number of agents after expansion.

[0053] This embodiment is based on minimum total capacity. and target percentage Data allocation. This is to prevent calculation errors from causing certain data types to fall below their hard threshold. Introducing the formula ;in ,and It is a set of integers.

[0054] Step 5: Determine the final target data generation amount for each agent based on the initial target data generation amount and the target data proportion corresponding to each agent.

[0055] This embodiment derives the required generation amounts for each part. Calculate the final total data pool size And perform a final proportional calibration to ensure that the final output strictly conforms to... Distribution requirements.

[0056] .

[0057] For example, in this scenario, it is necessary to generate a dataset for training a multimodal large model, which includes intermediate products (wireframes), component materials (header images), and final products (visually enhanced drafts).

[0058] Scene setup and parameters: Dataset ,in For the requirements draft, Line drawing This is the header image. For visual enhancement, This is a code project.

[0059] Raw data ( Requirements draft Wireframe Header image ; Visual enhancement draft Code Project .

[0060] Target ratio ( Requirements draft Wireframe Header image ; Visual enhancement draft Code Project .

[0061] Minimum expansion factor ( ): ; ; ; ; Calculate the minimum amount of data required to guarantee diversity for each data type: ; Determine the minimum total capacity: .

[0062] Determine the final output: .

[0063] .

[0064] Accordingly, based on the target data proportion and prompt words, a preset number of agents are used to generate page data to obtain the target page data. This can include: based on the final target data generation volume and prompt words corresponding to each agent, a preset number of agents are used to generate page data to obtain the target page data. This embodiment provides a specific process for generating page data based on the target proportion, improving the accuracy of page data generation.

[0065] S103, based on the target data ratio and prompt words, uses a preset number of intelligent agents to generate page data and obtain the target page data.

[0066] In this embodiment, a customized prompt word template closely aligned with the task is designed for each agent. These prompt words not only contain task instructions but also specific quantity requirements, format specifications, quality standards, and business rules calculated based on the target data proportion, ensuring that the large language model can understand and execute refined control intentions during generation. This embodiment inputs the prompt words to the model instance of the corresponding agent, initiating a large-scale, parallel data generation task to produce preliminary page data, code, or descriptions. The process of generating page data based on a preset number of agents can be, for example, in the current business scenario, including a first agent, a second agent, a third agent, and a fourth agent. The process of generating target page data involves: determining the first data obtained by the first agent based on requirements; generating second data using the second agent based on the first data; generating third data using the third agent based on the second data; and generating the target page data using the fourth agent based on the third data. In this invention, page data refers to a set of structured information generated for a marketing landing page of a specific vertical business, used for training or inputting the model. This data consists of structured data from each sub-agent. This embodiment, after obtaining page data, does not specify how to use it. Real-world vertical marketing landing page data is scarce and involves privacy concerns. Generated page data can be generated on a large scale for model training. By allowing the model to learn from these synthetic page data with clear marketing intent and business structure, the model can master the copywriting style, conversion path design, and visual layout patterns specific to that vertical. Each industry has its own compliance requirements, user psychology, and communication strategies. Generating page data allows these rules and logic to be injected during the generation process. After learning, the model's generated results will naturally follow these implicit rules, rather than generating a generic webpage without industry-specific characteristics. Since the page data is self-generated, the changes to each variable can be precisely controlled (e.g., changing the title, color, or button text). This facilitates research into "which marketing elements led to changes in conversion rates," thereby guiding the model to optimize and generate better business goals (high conversion, high click-through rates). In other words, the page data generated in this embodiment can be used to further train the agent, enabling it to generate pages that better meet requirements and improve the efficiency of subsequent page generation; or to generate pages that better meet user needs.

[0067] It should be noted that, in this embodiment, developers can customize the prompts based on different intelligent agents. A specific set of prompts can be customized according to different engineering processes to generate content of varying granularity and for different scenarios, facilitating subsequent development.

[0068] like: The following prompts can be used to expand news events and user needs for producing draft requests: Based on the following, propagate hypothetical news events and user needs.

[0069] News Events: {{ news_event}} User requirements: {{ requirement}} The output should be in JSON format, as follows: { "News Events": ... User requirements: ... } Example for reference: ...

[0070] It should be further explained that, based on any of the above embodiments, after generating page data using a preset number of intelligent agents based on the target data ratio and prompt words to obtain the target page data, the process may further include: determining the actual data ratio corresponding to each intelligent agent in the target page data; comparing the actual data ratio with the target data ratio to identify data types that have not reached a predetermined number and determining the intelligent agents corresponding to the data types to be supplemented; determining the target supplementary data volume based on the actual data ratio and the target data ratio; and generating data using the intelligent agents corresponding to the data types to be supplemented based on the target supplementary data volume until the data volume corresponding to each intelligent agent meets the set requirements. This embodiment performs a series of post-processing operations on the generated page data, including format standardization, syntax checking, and basic logic verification, and automatically filters and removes data entries with obvious errors, non-compliance with specifications, or low quality using rules or lightweight models to ensure the reliability of the data foundation for subsequent stages. It also performs data ratio verification and gap identification: after completing the first round of generation and cleaning, the system recalculates the actual ratio distribution of the generated data and compares it with the target data ratio. Through this process, it accurately identifies which types or scenarios of data have not yet reached the predetermined quantity or diversity requirements, i.e., it identifies data gaps. Further targeted iterative data generation: For the identified data gap types, the system restarts the generation process. However, this generation will be more targeted: on the one hand, carefully designed prompts for this type are reused; on the other hand, successfully generated high-quality data of the same type are incorporated into the context as minority examples, guiding the model to generate supplementary data with non-repetitive content and novel perspectives while maintaining style consistency, thereby efficiently filling gaps and enriching data diversity. Further closed-loop iterative optimization until the target is met: The generation of page data to iterative data generation forms a cyclical iterative optimization closed loop. The system continuously performs a cycle of generation-cleaning-evaluation-targeted regeneration, with each iteration striving to narrow the gap between data distribution and quality goals. This cycle will continue until all preset data types fully meet the established requirements in terms of quantity, distribution, diversity, and overall quality, ultimately outputting a high-quality, balanced vertical page dataset or set of pages.

[0071] It should be noted that the research objective of this invention focuses on marketing landing pages within vertical business sectors. Marketing landing pages have a clear business purpose—guiding users to complete conversions (such as registration, purchase, etc.). Their page structure, content layout, and visual elements differ significantly from ordinary information pages or blog pages. Existing public datasets mostly contain general web pages or e-commerce product pages, but marketing landing page data for specific vertical sectors (such as finance, education, and healthcare) is extremely scarce, and real data often involves business privacy and is difficult to obtain. Therefore, this invention fills this gap by generating marketing page data that conforms to business logic. Generated data allows for flexible control of variables, generating diverse page variations, facilitating a systematic exploration of the impact of marketing elements—something difficult to achieve with real data.

[0072] It should be further explained that, based on any of the above embodiments, the present invention can utilize a lightweight diagnostic analyzer (a small model that can be based on rules or fine-tuned). When the result given by an agent is marked as "poor quality," the diagnostic analyzer will: locate the defect pattern: for example, the requirement draft parsing agent frequently omits "price terms," ​​or the project code generation agent always incorrectly imports a non-existent library. Comparative analysis: compare the input / output of the failed task with successful cases. Generate a diagnostic report: infer the prompt word-level reasons that led to the failure. For example: "The instructions in the current prompt word regarding 'extracting key information' are too vague, causing the model to ignore numerical information"; or "The prompt word lacks constraints on the XX version interface changes, resulting in the generation of deprecated configurations." This embodiment shifts the focus of quality control from post-event screening to pre-event prevention and process improvement. The system can automatically diagnose defects in the tool (prompt word) itself.

[0073] This invention provides a page data generation method, which may include: S101, obtaining a preset number of intelligent agents corresponding to the current business scenario, and prompt words corresponding to each intelligent agent; wherein, the intelligent agents are obtained by decomposing the page linear generation process based on the current business scenario; S102, determining the target data proportion corresponding to each intelligent agent according to requirements; S103, generating page data using the preset number of intelligent agents based on the target data proportion and prompt words, to obtain target page data. Compared with the current method where the generation, filtering, and correction of each piece of data requires a large amount of manual labor, this application constructs a method based on multi-agent collaboration. Based on multiple intelligent agents corresponding to the page linear generation process, it can automatically and efficiently generate page data that conforms to business logic and rules, is precisely controllable in style and architecture, and has controllable quantity and type.

[0074] For a clearer understanding of this invention, please refer to the following details. Figure 3 , Figure 3A flowchart illustrating a page data generation method provided in this embodiment of the invention may specifically include: S201. Conduct in-depth analysis of the actual vertical page production work scenarios and construct intelligent agents. The intelligent agents should include at least: a requirements draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent.

[0075] S202. For each assigned intelligent agent, based on the business characteristics of the sub-tasks it is responsible for, perform a secondary decomposition of its internal functions to obtain the sub-intelligent agents corresponding to each intelligent agent.

[0076] S203. Determine the target generated data quantity for each agent based on the original data quantity, target ratio, and minimum expansion factor for each sub-agent.

[0077] S204. Design a set of customized prompt word templates for each agent that closely match its task; the prompt word templates include task instructions, target data generation quantity, format specifications, quality standards, and business rules.

[0078] S205. Input the prompt words into the model instance of the corresponding agent to start a large-scale, parallel data generation task and produce corresponding requirement drafts, image information, page code and other data.

[0079] S206. Conduct rigorous, multi-level quality screening and cleaning of the initially generated data, and systematically remove data entries that contain errors, do not conform to established specifications, or are substandard in quality.

[0080] S207. After the data is generated, recalculate the proportions of each category and compare them with the previously set matching requirements. Filter out data with insufficient categories and perform further processing.

[0081] S208. For the data gaps identified in the previous step, use step S204 to perform diversity generalization, using existing data as a few text examples to generate targeted data to enrich the diversity of the data.

[0082] S209, Closed-loop iterative optimization: Repeat S205 to S208 to continuously iterate and optimize until the sample data meets the requirements of quantity, diversity, quality, and coverage.

[0083] This invention defines a set of specifications for mobile page data in vertical business applications, clarifying the stages, nodes, and characteristics of data generation. This specification not only provides a basis for subsequent data segmentation according to agent responsibilities but also, based on this specification, decomposes complete data into specific data types corresponding to different agents, thus laying the foundation for subsequent model training and data reconstruction. Based on data distribution and proportion, a methodology for data allocation is constructed to control data balance and quality.

[0084] The following describes a page data generation apparatus provided by an embodiment of the present invention. The page data generation apparatus described below can be referred to in correspondence with the page data generation method described above.

[0085] Please refer to the details. Figure 4 , Figure 4 A schematic diagram of a page data generation device provided in an embodiment of the present invention may include: The agent determination module 100 is used to obtain a preset number of agents corresponding to the current business scenario, and the prompt words corresponding to each agent; wherein, the agent is a large language model obtained by decomposing the page linear generation process based on the current business scenario. The target data proportion determination module 200 is used to determine the target data proportion for each agent according to requirements. The target page data generation module 300 is used to generate page data based on the target data ratio and the prompt words, using the preset number of intelligent agents to obtain the target page data.

[0086] Furthermore, based on any of the above embodiments, the page data generation apparatus may further include: The first splitting module is used to split the page linear generation process based on the current business scenario to obtain the intelligent agent and the data structure content corresponding to each intelligent agent. The second splitting module is used to split each agent according to the data structure content to obtain a sub-agent corresponding to each agent; wherein, the agent includes the sub-agent.

[0087] Furthermore, based on any of the above embodiments, the first splitting module may include: The first splitting unit is used to split the page linear generation process according to the current business scenario, resulting in a requirement draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent; The first data structure content determination unit is used to determine the data structure content corresponding to the demand draft parsing agent, including content, user demand, and news events. The second data structure content determination unit is used to determine the data structure content corresponding to the line drawing design agent, including the main title of the page, the subtitle of the page, the card content, and the line drawing page. The third data structure content determination unit is used to determine the data structure content corresponding to the header image generating agent, including the prompt words used to generate the header image and the address after the header image is generated; The fourth data structure content determination unit is used to determine the data structure content corresponding to the visual enhancement agent, including color planning, card-in-card materials, and the final page; The fifth data structure content determination unit is used to determine the data structure content corresponding to the project code generation agent, including the multi-turn dialogue content during generation and the final project generation address.

[0088] Furthermore, based on any of the above embodiments, the target page data generation module 300 may include: The target requirement draft determination unit is used to analyze the intelligent agent based on the user's requirements and combine it with news source search for news events to form a target requirement draft. The line drawing data determination unit is used to form line drawing data including target main title, target subtitle, target card content and target line drawing page by using the line drawing design agent according to the target requirement draft; The header image scheme determination unit is used to extract the main title and subtitle from the line drawing data using the header image generation agent to form the header image and color scheme. The page preview draft determination unit is used to perform visual processing on the target line drawing page based on the header image and color scheme using the visual enhancement agent to form a page preview draft. The target page data determination unit is used to convert the preview draft into the target page data using the project code generation agent.

[0089] Furthermore, based on any of the above embodiments, the second splitting module may include: The second splitting unit is used to split the requirement draft parsing agent into a text generation sub-agent, a tag generation sub-agent, and a key information extraction sub-agent.

[0090] Furthermore, based on any of the above embodiments, the page data generation apparatus may further include: The initial data determination module is used to determine the amount of original data corresponding to each intelligent agent, the proportion of the target data corresponding to each intelligent agent, and the data expansion multiple corresponding to each intelligent agent; The module for determining the minimum number after expansion is used to determine the minimum number of agents after expansion based on the original data volume and the data expansion factor. The minimum capacity determination module is used to determine the minimum total capacity of the total dataset based on the minimum number of data after expansion corresponding to each agent and the proportion of the target data. The initial target data generation volume determination module is used to determine the initial target data generation volume for each intelligent agent based on the minimum total capacity, the target data ratio, and the minimum quantity after expansion for each intelligent agent. The final target generated data volume determination module is used to determine the final target generated data volume for each intelligent agent based on the initial target generated data volume and the target data proportion corresponding to each intelligent agent; Correspondingly, the target page data generation module 300 may include: The target page data generation unit is used to generate page data using the preset number of intelligent agents based on the final target data volume and the prompt words corresponding to each intelligent agent, thereby obtaining the target page data.

[0091] Furthermore, based on any of the above embodiments, the page data generation apparatus may further include: The actual data proportion determination module is used to determine the actual data proportion corresponding to each intelligent agent in the target page data; The module for determining the data type to be supplemented is used to compare the actual data ratio with the target data ratio, identify the data types to be supplemented that have not reached the predetermined number, and determine the intelligent agent corresponding to the data type to be supplemented. The target supplementary data volume determination module is used to determine the target supplementary data volume based on the actual data proportion and the target data proportion; and to generate data using the agents corresponding to the data types to be supplemented based on the target supplementary data volume until the data volume corresponding to each agent meets the set requirements.

[0092] It should be noted that the order of the modules and units in the above-mentioned page data generation device can be changed without affecting the logic.

[0093] This invention provides a page data generation device, which may include: an agent determination module 100, used to acquire a preset number of agents corresponding to the current business scenario, and a prompt word corresponding to each agent; wherein the agent is a large language model obtained by decomposing the linear page generation process of the current business scenario; a target data proportion determination module 200, used to determine the target data proportion corresponding to each agent according to requirements; and a target page data generation module 300, used to generate page data using the preset number of agents based on the target data proportion and the prompt word, to obtain target page data. Compared with the current method where the generation, filtering, and correction of each piece of data requires a large amount of manual labor, this application constructs a method based on multi-agent collaboration, which, based on multiple agents corresponding to the linear page generation process, can automatically and efficiently generate page data that conforms to business logic and rules, is precisely controllable in style and architecture, and has controllable quantity and type.

[0094] The following describes a page data generation device provided by an embodiment of the present invention. The page data generation device described below can be referred to in correspondence with the page data generation method described above.

[0095] Please refer to Figure 5 , Figure 5 A schematic diagram of a page data generation device provided in an embodiment of the present invention may include: Memory 10 is used to store computer programs; Processor 20 is used to execute computer programs to implement the page data generation method described above.

[0096] The memory 10, processor 20, and communication interface 30 all communicate with each other through the communication bus 40.

[0097] In this embodiment of the invention, the memory 10 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment of the invention, the memory 10 may store programs for implementing the following functions: Obtain a preset number of intelligent agents corresponding to the current business scenario, and the prompt words corresponding to each intelligent agent; wherein, the intelligent agents are obtained by decomposing the page linear generation process based on the current business scenario; Determine the target data percentage for each agent based on requirements; Based on the target data proportion and prompt words, a preset number of intelligent agents are used to generate page data to obtain the target page data.

[0098] In one possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0099] Furthermore, memory 10 may include read-only memory and random access memory, providing instructions and data to the processor. A portion of the memory may also include NVRAM. The memory stores operating systems and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0100] Processor 20 can be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic device. Processor 20 can be a microprocessor or any conventional processor. Processor 20 can call programs stored in memory 10.

[0101] The communication interface 30 can be an interface for the communication module, used to connect with other devices or systems.

[0102] Of course, it should be noted that, Figure 5 The structure shown does not constitute a limitation on the page data generation device in the embodiments of the present invention. In practical applications, the page data generation device may include devices such as... Figure 5 More or fewer components as shown, or combinations of certain components.

[0103] The following describes the readable storage medium (i.e., computer-readable storage medium) provided in the embodiments of the present invention. The computer-readable storage medium described below can be referred to in correspondence with the page data generation method described above.

[0104] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the page data generation method described above.

[0105] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0107] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0108] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] The foregoing has provided a detailed description of a page data generation method, apparatus, device, and readable storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for generating page data, characterized in that, include: Obtain a preset number of intelligent agents corresponding to the current business scenario, and the prompt words corresponding to each intelligent agent; wherein, the intelligent agent is a large language model obtained by decomposing the page linear generation process based on the current business scenario; Determine the target data percentage for each agent based on requirements; Based on the target data ratio and the prompt words, page data is generated using the preset number of intelligent agents to obtain the target page data.

2. The page data generation method according to claim 1, characterized in that, Before obtaining the preset number of intelligent agents corresponding to the current business scenario and the prompt words corresponding to each intelligent agent, the process also includes: The linear page generation process based on the current business scenario is broken down to obtain the intelligent agent and the corresponding data structure content of each intelligent agent; Each agent is split according to the data structure content to obtain a sub-agent corresponding to each agent; wherein, the agent includes the sub-agent.

3. The page data generation method according to claim 2, characterized in that, The linear page generation process based on the current business scenario is broken down to obtain the intelligent agent and the corresponding data structure content for each intelligent agent, including: Based on the current business scenario, the page linear generation process is broken down into a requirement draft parsing intelligent agent, a line drawing design intelligent agent, a header image generation intelligent agent, a visual enhancement intelligent agent, and a project code generation intelligent agent; The data structure content corresponding to the intelligent agent that parses the requirements draft includes content, user requirements, and news events; The data structure content corresponding to the line drawing design agent includes the main title of the page, the subtitle of the page, the card content, and the line drawing page; The data structure content corresponding to the header image generating agent includes the prompt words used to generate the header image and the address after the header image is generated; The data structure content corresponding to the visual enhancement agent includes color planning, card-in-card materials, and the final page. The data structure content corresponding to the project code generation agent includes the multi-turn dialogue content during generation and the final project generation address.

4. The page data generation method according to claim 3, characterized in that, Based on the target data proportion and the prompt words, page data is generated using the preset number of intelligent agents to obtain target page data, including: Based on user needs, the intelligent agent is used to analyze the requirements draft and combine it with news source search for news events to form a target requirements draft; Based on the target requirements draft, the line drawing design agent is used to generate line drawing data including the target main title, target subtitle, target card content, and target line drawing page. The header image generation agent extracts the main title and subtitle from the line drawing data to form the header image and color scheme. Based on the header image and color scheme, the visual enhancement agent is used to perform visual processing on the target line drawing page to form a page preview draft. The project code is used to generate an intelligent agent that converts the preview image into the target page data.

5. The page data generation method according to claim 3, characterized in that, Based on the data structure content, each agent is split into sub-agents corresponding to each agent, including: The requirement draft parsing agent is divided into a text generation sub-agent, a tag generation sub-agent, and a key information extraction sub-agent.

6. The page data generation method according to any one of claims 1 to 5, characterized in that, After determining the target data percentage for each agent based on requirements, the following steps are also included: Determine the amount of original data for each agent, the proportion of the target data for each agent, and the data expansion factor for each agent; Based on the original data volume and the data expansion factor, determine the minimum number of agents after expansion for each agent; The minimum total capacity of the total dataset is determined based on the minimum number of data agents after expansion and the proportion of target data for each agent. The initial target data generation amount for each agent is determined based on the minimum total capacity, the target data proportion, and the minimum quantity after expansion for each agent. The final target data generation amount for each intelligent agent is determined based on the initial target data generation amount and the target data proportion corresponding to each intelligent agent. Accordingly, based on the target data proportion and the prompt words, page data is generated using the preset number of intelligent agents to obtain target page data, including: Based on the final target data volume and the prompt words corresponding to each agent, page data is generated using the preset number of agents to obtain the target page data.

7. The page data generation method according to claim 1, characterized in that, After generating page data using the preset number of intelligent agents based on the target data proportion and the prompt words to obtain the target page data, the process further includes: Determine the actual data percentage corresponding to each agent in the target page data; The actual data proportion is compared with the target data proportion to identify the data types that have not reached the predetermined quantity and to determine the intelligent agents corresponding to the data types that need to be supplemented. The target amount of supplementary data is determined based on the actual data percentage and the target data percentage. Based on the target amount of supplementary data, data is generated using the agents corresponding to the data types to be supplemented, until the amount of data corresponding to each agent meets the set requirements.

8. A page data generation device, characterized in that, include: The agent determination module is used to obtain a preset number of agents corresponding to the current business scenario, and the prompt words corresponding to each agent; wherein, the agent is a large language model obtained by decomposing the page linear generation process of the current business scenario; The target data percentage determination module is used to determine the target data percentage for each agent according to requirements. The target page data generation module is used to generate page data based on the target data ratio and the prompt words, using the preset number of intelligent agents to obtain the target page data.

9. A page data generation device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the page data generation method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the steps of the page data generation method as described in any one of claims 1 to 7.

Images