Active page generation method, related device and medium

Abstract

The invention provides an activity page generation method, a related device and a medium. The activity page generation method comprises the following steps: inputting description information of a target activity into a target agent; performing information expansion on the description information through the target agent to obtain expanded information; inputting the expanded information into a first large language model to obtain a sub-process time sequence and page generation requirement information corresponding to the sub-process; selecting a picture post-processing model based on the expanded information through the target agent, and generating a material picture under the guidance of the expanded information by utilizing the selected picture post-processing model; and through the target agent, inputting page generation requirement information corresponding to each sub-process in the sub-process time sequence into the second large language model to obtain a corresponding active page, and adding the material picture to the active page to generate a target active page. According to the embodiment of the invention, the automation degree and efficiency of activity page generation can be improved. The embodiment of the invention can be applied to scenes such as artificial intelligence and page generation.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and in particular to a method for generating an activity page, a related device, and a medium. Background Art

[0002] Currently, internet applications often run promotions to increase user engagement and influence. For example, a section of an app might hold a promotion during the Mid-Autumn Festival, offering prizes for posting short videos related to the festival. This promotion involves initial planning and subsequent creation of an event page. This page allows users to participate in the promotion.

[0003] Existing methods for generating event pages typically involve manually planning a plan in the background and then manually developing the event page based on the plan, or manually planning the plan and then using the visual drag-and-drop function of a low-code platform to generate the event page. Both methods lack automation and result in low page generation efficiency. Summary of the Invention

[0004] The embodiments of the present disclosure provide an activity page generation method, related devices, and media, which can improve the automation and efficiency of activity page generation.

[0005] According to one aspect of the present disclosure, a method for generating an activity page is provided, comprising:

[0006] Inputting the description information of the target activity into the target agent;

[0007] Expanding the description information through the target agent to obtain expanded information;

[0008] Inputting the expanded information into a first large language model to obtain a sub-process timing of the target activity and page generation requirement information corresponding to the sub-process;

[0009] Selecting, by the target agent, a picture post-processing model based on the expanded information, and generating a material picture using the selected picture post-processing model under the guidance of the expanded information;

[0010] Through the target intelligent agent, for each sub-process in the sub-process sequence, the page generation requirement information corresponding to the sub-process is input into the second largest language model to obtain the activity page corresponding to the sub-process, and the material picture is added to the activity page, thereby generating a target activity page.

[0011] According to one aspect of the present disclosure, there is provided an activity page generating device, comprising:

[0012] A first input unit, configured to input description information of a target activity into a target agent;

[0013] An information expansion unit, configured to expand the description information through the target agent to obtain expanded information;

[0014] a second input unit, configured to input the expanded information into the first large language model to obtain a sub-process sequence of the target activity and page generation requirement information corresponding to the sub-process;

[0015] An image generation unit, configured to select, through the target agent, an image post-processing model based on the expanded information, and generate a material image using the selected image post-processing model under the guidance of the expanded information;

[0016] The page generation unit is used to input the page generation requirement information corresponding to each sub-process in the sub-process sequence into the second language model through the target intelligent agent, obtain the activity page corresponding to the sub-process, and add the material picture to the activity page, thereby generating a target activity page.

[0017] Optionally, the information expansion unit is specifically configured to:

[0018] Determining, through the target agent, the application to which the target activity belongs;

[0019] Obtaining a word definition library for the application;

[0020] extracting description keywords from the description information;

[0021] Based on the description keywords, searching the word interpretation library to obtain interpretation information of the description keywords;

[0022] The interpretation information is added to the description information to obtain the expanded information.

[0023] Optionally, the first large language model includes a first sub-large language model and a second sub-large language model;

[0024] The second input unit is specifically used for:

[0025] Inputting the expanded information into the first sub-large language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process;

[0026] The expanded information, the sub-process timing and the page generation requirement information are input into the second sub-large language model to obtain the revised sub-process timing and the page generation requirement information.

[0027] Optionally, the picture post-processing model includes a first picture post-processing model based on picture style and a second picture post-processing model based on application;

[0028] The image generation unit is specifically configured to:

[0029] Inputting the expanded information into a third language model through the target agent to obtain picture style keywords;

[0030] Based on the picture style keyword, selecting the first picture post-processing model from a plurality of first candidate picture post-processing models;

[0031] Determining the application to which the target activity belongs;

[0032] Based on the application, the second picture post-processing model is selected from a plurality of second candidate picture post-processing models.

[0033] Optionally, the image generation unit is further specifically configured to:

[0034] Inputting the expanded information into a stable diffusion model to obtain a material image base;

[0035] The image post-processing model is used to perform post-processing on the material image base to obtain the material image.

[0036] Optionally, the image generation unit is further specifically configured to:

[0037] Extracting forward guidance keywords and reverse guidance keywords from the expanded information;

[0038] Converting the forward guidance keyword into a first guidance vector and converting the reverse guidance keyword into a second guidance vector;

[0039] Inputting the initial basis vector into the stable diffusion model to obtain the material picture basis vector under the guidance of the first guide vector and the second guide vector;

[0040] The material image base vector is converted into the material image base.

[0041] Optionally, the image generation unit is further specifically configured to:

[0042] Initialize the to-be-diffused basis vector to the initial basis vector, and initialize the step number to 1;

[0043] Inputting the to-be-diffused base vector, the step number, the first guide vector, and the second guide vector into the stable diffusion model to obtain diffusion noise corresponding to the step number;

[0044] The base vector to be diffused is offset by the diffusion noise corresponding to the step number, the step number is increased by 1, and the process returns to the step of inputting the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model until the step number increases to a preset maximum number of steps.

[0045] Optionally, the image post-processing model includes a first image post-processing model based on image style and a second image post-processing model based on application, and the post-processing includes style change and application mark superposition;

[0046] The image generation unit is further specifically configured to:

[0047] Using the first image post-processing model, changing the style of the material image base;

[0048] The second image post-processing model is used to superimpose the application mark on the material image base after the style is changed to obtain the material image.

[0049] Optionally, the page generation requirement information includes management-side page generation requirement information and client-side page generation requirement information, and the activity page includes a management-side activity page and a client-side activity page;

[0050] The page generation unit is specifically used for:

[0051] Inputting the client page generation requirement information corresponding to the sub-process into a second language model to obtain the client activity page corresponding to the sub-process;

[0052] The management end page generation requirement information corresponding to the sub-process is input into the second language model to obtain the management end activity page corresponding to the sub-process.

[0053] Optionally, the page generating unit is further configured to:

[0054] Obtaining a first domain-specific language rule of the client;

[0055] Inputting the client page generation requirement information corresponding to the sub-process and the first domain-specific language rules into the second language model to obtain the first domain-specific language corresponding to the sub-process;

[0056] The first domain-specific language is rendered into the client active page using a domain-specific language interpreter.

[0057] Optionally, the page generating unit is further configured to:

[0058] Acquire a second domain specific language rule of the management terminal;

[0059] Inputting the management-end page generation requirement information corresponding to the sub-process and the second domain-specific language rules into the second language model to obtain the second domain-specific language corresponding to the sub-process;

[0060] The second domain-specific language is rendered into the management-end active page using a domain-specific language interpreter.

[0061] Optionally, the page generating unit is further configured to:

[0062] Generating requirement information from the page, obtaining a material picture adding position, and information on a correspondence between the material picture adding position and the material picture;

[0063] At the material picture adding position of the activity page, the material picture corresponding to the material picture adding position is added according to the corresponding relationship information.

[0064] Optionally, after adding the material picture to the activity page to generate the target activity page, the activity page generating device further includes:

[0065] A first sending unit is used to send the target format information of the response data and key indicators to the application front end through the target agent;

[0066] an acquiring unit, configured to acquire, through the application front end, the response data of the platform object to the target activity page within a predetermined time period after the target activity page is generated, and acquire the key indicator based on the response data;

[0067] A second sending unit is configured to send the response data and the key indicators to the target agent based on the target format information through the application front end;

[0068] The third input unit is used to input the response data and the key indicators into the fourth language model through the target intelligent agent to obtain a data report page.

[0069] Optionally, the third input unit is specifically configured to:

[0070] Inputting the response data and the key indicators into a fourth language model through the target agent to obtain a third domain-specific language;

[0071] The third domain specific language is rendered into the data report page using a domain specific language interpreter.

[0072] Optionally, the first input unit is specifically configured to:

[0073] determining an activity type of the target activity;

[0074] Determining the application to which the target activity belongs;

[0075] determining the target agent from a plurality of candidate agents based on the application and the activity type;

[0076] The description information of the target activity is input into the target agent.

[0077] Optionally, the second input unit is further specifically configured to:

[0078] extracting sub-process description information and page description information from the expanded information through the target agent;

[0079] Inputting the sub-process description information into the first language model to obtain the sub-process time sequence;

[0080] The sub-process sequence and the page description information are input into the first large language model to obtain page generation requirement information corresponding to the sub-process.

[0081] Optionally, after inputting the expanded information into the first large language model to obtain the sub-process timing of the target activity and page generation requirement information corresponding to the sub-process, the activity page generation apparatus further includes:

[0082] a fourth input unit, configured to input, through the target agent, for each sub-process, the sub-process timing, the page generation requirement information corresponding to the sub-process, the first page generation requirement information corresponding to a predetermined number of preceding sub-processes preceding the sub-process, and the second page generation requirement information corresponding to a predetermined number of subsequent sub-processes following the sub-process into a context continuity determination model, thereby obtaining a context continuity determination result for the sub-process;

[0083] An adjusting unit is configured to adjust the page generation requirement information corresponding to the sub-process based on the context continuity determination result through the target agent.

[0084] According to one aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned active page generation method when executing the computer program.

[0085] According to one aspect of the present disclosure, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the method for generating an active page as described above is implemented.

[0086] According to one aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. The computer program is read and executed by a processor of a computer device, so that the computer device executes the above-mentioned active page generation method.

[0087] In the disclosed embodiment, a target agent is set up and programmed to call upon a large language model to complete some tasks based on all the processes required for generating an activity page, and to utilize the target agent's own capabilities to complete some tasks. These tasks are then combined to generate the activity page. Specifically, the target agent receives a description of the target activity as input, uses its own capabilities to expand the description, inputs the expanded information into a first large language model, and uses the capabilities of the first large language model to generate a sub-process sequence and page generation requirements for each sub-process. Based on the sub-process sequence, the order of multiple consecutive target activity pages can be determined. Based on the page generation requirements for each sub-process, the text content on each target activity page can be determined. In addition to text, the target activity page must also contain source images. To obtain these source images, the target agent selects an image post-processing model based on the expanded information and uses the selected image post-processing model to generate the source images. After obtaining the source images, the target agent inputs the page generation requirements corresponding to each sub-process into a second large language model to obtain the activity page for that sub-process and adds the corresponding source images to the activity page, thereby generating the activity page. The entire process does not require any human intervention, and accuracy is guaranteed by leveraging the capabilities of the target intelligent agent itself and the large language model, thereby improving the automation and efficiency of activity page generation.

[0088] Other features and advantages of the present disclosure will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present disclosure. The purposes and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The accompanying drawings are used to provide a further understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation to the technical solution of the present disclosure.

[0090] Figure 1 is a system architecture diagram of a method for generating an activity page according to an embodiment of the present disclosure;

[0091] Figure 2A-2E This is a schematic diagram of an interface of an embodiment of the present disclosure applied in an activity page push scenario;

[0092] Figure 3is a flow chart of a method for generating an activity page according to an embodiment of the present disclosure;

[0093] Figure 4 yes Figure 3 Step 310 is a flow chart of inputting the description information of the target activity into the target agent;

[0094] Figure 5 yes Figure 3 A flow chart of step 320 for obtaining the expanded information;

[0095] Figure 6 yes Figure 5 Step 510 is a flow chart for obtaining a word definition library;

[0096] Figure 7 yes Figure 5 A schematic diagram of step 530 regarding determining interpretation information from a word interpretation library based on description keywords;

[0097] Figure 8 yes Figure 3 Step 330 is a flowchart for obtaining the sub-process timing and the page generation requirement information corresponding to the sub-process;

[0098] Figure 9 yes Figure 3 Another flowchart of step 330 regarding obtaining the sub-process timing and page generation requirement information corresponding to the sub-process;

[0099] Figure 10 It is a flow chart for determining model adjustment and page generation requirement information corresponding to sub-processes based on context continuity;

[0100] Figure 11 yes Figure 3 A flowchart of step 340 regarding selecting a post-processing model for an image;

[0101] Figure 12 yes Figure 11 A schematic diagram of step 1120 regarding selecting a first image post-processing model;

[0102] Figure 13 yes Figure 11 A schematic diagram of step 1140 regarding selecting a second image post-processing model;

[0103] Figure 14 yes Figure 3 A flow chart of step 340 for obtaining a material image;

[0104] Figure 15 yes Figure 14 A flow chart of step 1410 for obtaining a material image base based on a stable diffusion model;

[0105] Figure 16 yes Figure 14 Step 1410 is a schematic diagram of obtaining a material image base based on a stable diffusion model;

[0106] Figure 17 yes Figure 15 A schematic diagram of step 1520 regarding conversion of a guiding keyword into a corresponding guiding vector;

[0107] Figure 18 yes Figure 15 Step 1530 is a flowchart of obtaining a base vector of a material image under the guidance of the first guide vector and the second guide vector;

[0108] Figure 19 yes Figure 14 Step 1420 is a flow chart for post-processing the material image base to obtain the material image;

[0109] Figure 20 yes Figure 3 Step 350 is a flowchart for obtaining the activity page corresponding to the sub-process;

[0110] Figure 21 yes Figure 20 Step 2010 is a flowchart of obtaining the client activity page corresponding to the sub-process;

[0111] Figure 22 yes Figure 20 Step 2020 is a flowchart for obtaining the management end activity page corresponding to the sub-process;

[0112] Figure 23 yes Figure 20 A diagram of the management activity page in step 2020;

[0113] Figure 24 yes Figure 3 Step 350 is a flowchart for adding a material image to an activity page;

[0114] Figure 25 yes Figure 3 A schematic diagram of step 350 for adding a material image to the activity page;

[0115] Figure 26 is a flow chart of generating a data report page according to an embodiment of the present disclosure;

[0116] Figure 27 yes Figure 26 Step 2640 is a flowchart of using the fourth language model and the domain-specific language interpreter to render a data report page;

[0117] Figure 28 It is a framework flow chart of related technologies to generate activity pages;

[0118] Figure 29 is a framework flow chart for generating an activity page according to an embodiment of the present disclosure;

[0119] Figure 30 is a module diagram of an activity page generating device according to an embodiment of the present disclosure;

[0120] Figure 31 According to one embodiment of the present disclosure Figure 3 The terminal structure diagram of the activity page generation method shown;

[0121] Figure 32 According to one embodiment of the present disclosure Figure 3 The server structure diagram of the activity page generation method shown. DETAILED DESCRIPTION

[0122] In order to make the purpose, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not intended to limit the present disclosure.

[0123] Before further explaining the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are explained. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations:

[0124] Large Language Model (LLM): A natural language processing model based on deep learning technology. LLMs possess powerful semantic understanding and generation capabilities, capable of generating text in human-language format. LLMs have achieved significant performance improvements in various natural language processing tasks, such as machine translation, text generation, and question-answering systems. Key concepts of LLMs include the Transformer architecture, pre-training, fine-tuning, masked language modeling, and zero-shot or few-shot learning. LLMs are based on the Transformer architecture, a deep learning model with a self-attention mechanism that captures relationships between words in a text. LLMs are typically pre-trained on large amounts of text data, acquiring rich semantic information through unsupervised learning. The goal of pre-training is to train the model to effectively model and generate input text. LLMs can be fine-tuned on specific tasks to adapt to different natural language processing tasks. During fine-tuning, the model is trained on labeled data to acquire task-specific knowledge. LLMs can be pre-trained on the Masked Language Modeling (MLM) task. In the MLM task, the model needs to predict the masked words in the sentence to learn richer semantic information. LLM can be trained on zero-shot or few-shot learning tasks to address the problem of insufficient training data. By using unlabeled data for pre-training, the model can learn richer semantic information, thereby improving model performance.

[0125] Agent: In the field of machine learning (ML), an agent is an entity capable of perceiving its environment, making decisions, and executing actions. An agent can be a software program, a robot, an automated system, or any entity capable of interacting with and reacting to its environment. This entity can be a physical robot or a virtual software program. An agent operates within its environment, receiving information from it and taking actions based on predefined goals and policies to maximize its expected reward or utility. An agent typically consists of three main components: a perception, a decision-maker, and an actuator. The perception is the interface between the agent and the environment, receiving information from the environment, such as sensor data, images, and sounds. The decision-maker is the core of the agent, responsible for selecting the optimal action based on its current state and goals. The actuator is the component used by the agent to execute the selected action, such as a robot's motors or a software program's outputs. The goal of an agent is to continuously learn and improve its behavior through continuous interaction with its environment, thereby continuously improving its performance and efficiency. In machine learning, agents typically learn and improve their behavior through reinforcement learning algorithms, which enable them to adjust their behavior based on feedback from the environment to achieve optimal results.

[0126] Reinforcement learning is a machine learning paradigm designed to enable intelligent agents to learn how to make decisions through interaction with their environment to maximize future cumulative rewards. The core idea of ​​reinforcement learning is to allow the agent to try different actions, observe environmental feedback, and adjust its behavior based on this feedback to maximize long-term rewards. Reinforcement learning includes Q-learning, SARSA, and deep reinforcement learning (such as deep Q-networks and policy gradient methods).

[0127] Stable Diffusion (SD) is an unsupervised learning method. This method allows models to be trained on large datasets and then gradually transferred to specific tasks. The core idea of ​​SD is that during training, the model needs to learn how to map input data into a stable, low-dimensional latent space and gradually learn how to represent and organize information in this space. In the field of natural language processing (NLP), SD can be used for tasks such as text generation, machine translation, and text summarization. In these tasks, the model needs to learn how to map input text into a new, task-related space and effectively represent and organize information in this space.

[0128] Low-Rank Adaptation (LoRA) is a method used to address the problem of numerous parameters and limited training data in large-scale deep learning models. In traditional deep learning, model parameters typically require extensive computational resources to train, but LoRA allows for the use of less training data while maintaining high performance. The key idea behind LoRA is that during model training, only some weights are updated, while others remain unchanged. This significantly reduces computational effort and training time while maintaining high performance. LoRA is often combined with sparse coding and low-rank constraints to leverage sparsity and structural information. In natural language processing, LoRA can be used for tasks such as text generation, machine translation, and text summarization. In these tasks, models need to utilize limited training data to reduce computational effort while maintaining performance.

[0129] A Domain-Specific Language (DSL) is a computer language designed specifically for a specific domain or problem area. Unlike general-purpose programming languages ​​like Java and Python, DSLs focus on solving problems in a specific domain (industry or problem type), providing a higher level of abstraction and more intuitive expressions, making it easier for domain experts to express and solve problems within that domain. In other words, DSLs focus on solving problems in specific domains, making programming simpler and more intuitive.

[0130] Low-code refers to a development approach that uses a visual interface and minimal manual coding to create applications. This approach typically includes tools such as drag-and-drop functionality, predefined modules, templates, and automatically generated code. Low-code platforms and frameworks aim to simplify the application development process, lower the programming knowledge threshold, and make it easier for non-developers to build and deploy powerful applications. Low-code development does not completely replace traditional manual coding development methods, but in many scenarios, it can be an efficient and easy-to-use alternative.

[0131] Business Intelligence System (BI System): A system that uses software and services to convert enterprise data into useful information to help enterprise managers make more informed decisions.

[0132] System architecture and scenario description of the application of the embodiments of the present disclosure

[0133] Figure 1This is a system architecture diagram of the method for generating an activity page according to an embodiment of the present disclosure. In the activity page generation scenario, a server and multiple clients are involved. Figure 1 As shown, it includes: an object terminal 110, the Internet 120, a gateway 130, and a server 140.

[0134] The target terminal 110 is the device used by the target to view the currently generated target activity page. The target terminal 110 can be a desktop computer, laptop computer, PDA (Personal Digital Assistant), mobile phone, in-vehicle terminal, home theater terminal, dedicated terminal, and other forms. Furthermore, it can be a single device or a collection of multiple devices. For example, multiple devices connected via a local area network and sharing a common display device can work together to form a terminal. The target terminal 110 can also communicate with the Internet 120 via wired or wireless means to exchange data.

[0135] Gateway 130 is also known as an internetwork connector or protocol converter. It implements network interconnection at the transport layer and is a computer system or device that performs a conversion function. It acts as a translator between two systems using different communication protocols, data formats, or languages, or even completely different architectures. Gateway 130 can also provide filtering and security functions. The trigger operation for the current target active page sent by the object terminal 110 to the server 140 is sent to the corresponding server 140 via gateway 130. Feedback sent by the server 140 to the object terminal 110 in response to the object's trigger operation for the current target active page is also sent to the corresponding object terminal 110 via gateway 130.

[0136] The server 140 is a computer system that provides feedback to the target terminal 110 in response to the target terminal 110's triggering operation on the current target active page. Compared to the target terminal 110, the server 140 has higher requirements in terms of stability, security, and performance. The server 140 can be a high-performance computer in a network platform, a cluster of multiple high-performance computers, a portion of a high-performance computer (e.g., a virtual machine), a combination of portions of multiple high-performance computers (e.g., virtual machines), etc. The server 140 can also communicate with the Internet 120 via wired or wireless means to exchange data.

[0137] The disclosed embodiment can be applied in a variety of scenarios, such as A activity of application A, B activity of application B, etc., to provide different sections of different applications with activity pages required in different activity scenarios. To understand the activity page generation method provided by the disclosed embodiment in more detail, please refer to Figures 2A to 2EThe scene shown is viewing the generated target activity page in the instant short message application, etc.

[0138] like Figure 2A As shown, in the interface of the instant short message application in the object terminal 110, the instant short message application can be used to communicate with multiple contacts for instant messaging, and can also be used to watch videos. After triggering the "Message" option in the function bar below, the short message list is displayed. The short message list contains message bars with multiple contacts. In order to improve the stickiness of the object, the short video section of the instant short message application holds an activity during the Mid-Autumn Festival where prizes can be received by posting Mid-Autumn Festival-related short videos. In order to allow the object to quickly know the current activities of the short video section of the instant short message application, at 10:30 am, there is an activity prompt at the top of the short message list (such as Figure 2A This event notification reminds participants that new event content is available. Participants can view the pushed event content by triggering the "Short Video" option in the function bar below. Participants can also view pushed event content by triggering the "Mid-Autumn Festival Video Release Event."

[0139] like Figure 2B As shown, at 12:00 am, the subject views the activity prompt of "Mid-Autumn Festival Video Release Activity" in the interface of the instant short message application, and selects the "Mid-Autumn Festival Video Release Activity" option at the top of the trigger short message list.

[0140] like Figure 2C As shown, after triggering the "Mid-Autumn Festival Video Release Event" option at the top of the short message list, the instant short message application opens the short video push interface, and pops up an activity option box on the current interface to remind the object whether to publish a video to participate in the current activity. The object can participate in the "Mid-Autumn Festival Video Release Event" by uploading a video by triggering the "Publish Video" option in the activity option box. The object can also play the pushed content in the content push interface by triggering the "Exit" option in the activity option box. Therefore, if the object triggers the "Publish Video" option in the activity option box, the video to be published can be uploaded, and the following will be displayed after the video upload is completed. Figure 2D The interface shown.

[0141] like Figure 2D As shown, after the object uploads the video to be published, the video uploaded by the object is displayed on the short video push interface, and the option of "Video published successfully, receive the prize" is displayed on the current interface.

[0142] like Figure 2EAs shown, after the subject triggers the "Video published successfully, claim prize" option on the current interface, the current interface displays a prompt message saying "Prize claimed successfully" and the prize claimed is a "delicious mooncake." The subject can then claim the prize online or offline using the pre-set prize claim method in the instant messaging app.

[0143] Please note that when the target recipient claims a prize, the app may obtain their personal information. Before obtaining this information, you must obtain their consent. Furthermore, the collection, use, and processing of this information will comply with relevant laws, regulations, and standards. When obtaining the target recipient's consent, you can obtain their individual permission or consent through a pop-up window or by redirecting them to a confirmation page.

[0144] In the disclosed embodiment, a target agent is set up and programmed to perform tasks based on all the processes required for generating an activity page. It can also call upon the large language model to complete some tasks, leverage its own capabilities to complete other tasks, and combine these tasks to generate the activity page. This entire process requires no human intervention, and the combination of the target agent's own capabilities and the large language model ensures accuracy, improving the automation and efficiency of activity page generation.

[0145] General description of the embodiments of the present disclosure

[0146] According to one embodiment of the present disclosure, a method for generating an activity page is provided.

[0147] The activity page generation method dynamically generates activity pages for specific activities within a section of a website or app to increase user engagement. These activities can be held at specific times, such as weekdays, weekends, and holidays. In other words, by hosting an activity within a section of a website or app at a specific time, users can participate in the activity, thereby increasing their visits to the website or app. The target user is the person who will view the generated activity page. The content of the activity page can include images, videos, articles, applications, links, and other types of content.

[0148] Currently, internet applications often run promotions to increase user engagement and influence. For example, a section of an app might hold a promotion during the Mid-Autumn Festival, offering prizes for posting short videos related to the festival. This promotion involves initial planning and subsequent creation of an event page. This page allows users to participate in the promotion.

[0149] Related art methods for generating event pages typically involve manually planning a plan in the background and then manually developing the event page based on the plan, or manually planning the plan and then using the visual drag-and-drop function of a low-code platform to generate the event page. Both methods lack automation and result in low page generation efficiency.

[0150] In this disclosed embodiment, a target agent is set up and programmed to perform tasks based on all the processes required for generating an activity page. It can then invoke the large language model to complete some tasks, leverage its own capabilities to complete other tasks, and combine these tasks to generate the activity page. This entire process requires no human intervention, and the combination of the target agent's own capabilities and the large language model ensures accuracy, improving the automation and efficiency of activity page generation.

[0151] The activity page generation method of the embodiment of the present disclosure is executed on the server 140. After determining the target activity page corresponding to the target activity, a reminder message corresponding to the target activity page is pushed to the interface of the target terminal 110 via the network and the Internet 120.

[0152] like Figure 3 As shown, according to one embodiment of the present disclosure, the activity page generation method includes:

[0153] Step 310: Input the description information of the target activity into the target agent;

[0154] Step 320: Expand the description information through the target agent to obtain expanded information;

[0155] Step 330: Input the expanded information into the first language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process;

[0156] Step 340: The target agent selects an image post-processing model based on the expanded information, and uses the selected image post-processing model to generate a material image under the guidance of the expanded information.

[0157] Step 350: Through the target intelligent agent, for each sub-process in the sub-process sequence, the page generation requirement information corresponding to the sub-process is input into the second largest language model to obtain the activity page corresponding to the sub-process, and the material picture is added to the activity page, thereby generating the target activity page.

[0158] The advantage of this embodiment of steps 310-350 is that the target agent receives the target activity's descriptive information as input, leverages its own capabilities to expand the descriptive information, inputs the expanded information into the first large language model, and leverages the capabilities of the first large language model to generate the sub-process sequence and page generation requirements for each sub-process. Based on the sub-process sequence, the order of the multiple consecutive target activity pages generated can be determined. Based on the page generation requirements for each sub-process, the text content on each target activity page can be determined. In addition to text, the target activity page must also contain source images. To obtain these source images, the target agent selects an image post-processing model based on the expanded information and uses the selected image post-processing model to generate the source images. After obtaining the source images, the target agent inputs the page generation requirements corresponding to each sub-process into the second large language model to generate the activity page for that sub-process. The corresponding source images are then added to the activity page, thereby generating the activity page. The entire process requires no human intervention and leverages the combined capabilities of the target agent and the large language model to ensure accuracy, improving the automation and efficiency of activity page generation.

[0159] The above steps 310 to 350 are described in detail below.

[0160] Detailed description of step 310

[0161] In step 310, the target activity description information is used to represent text information that details the specific content and form of the target activity. The target activity description information can be pre-defined by the initiator of the target activity. The target activity description information may include the target activity's name, time, location, content, purpose, participants, activity process, and relevant regulations (such as participation conditions and rewards). The target activity page generated based on the target activity description information can help participating target participants understand the specific details of the target activity so that they can make appropriate decisions, such as whether to participate in the activity and how to participate in the activity.

[0162] A target agent is an entity used to assist the large language model in generating the target activity page. The target agent in the disclosed embodiment can independently complete tasks at various stages of generating the target activity page, and combine these tasks to achieve activity page generation. In other words, the disclosed embodiment utilizes a single target agent throughout the entire process of generating the target activity page, assisting the large language model in completing certain tasks. This entire process requires no human intervention, effectively improving the automation and efficiency of activity page generation.

[0163] In one embodiment, Figure 4 As shown, step 310 includes:

[0164] Step 410: Determine the activity type of the target activity;

[0165] Step 420: Determine the application to which the target activity belongs;

[0166] Step 430: Determine a target agent from a plurality of candidate agents based on the application and activity type;

[0167] Step 440: Input the description information of the target activity into the target agent.

[0168] In step 410, the activity type of the target activity refers to the type of activity at a specific time. For example, activity types include holiday promotions, weekend promotions, and daily promotions. Determining the activity type of the target activity helps clarify the nature and characteristics of the target activity. By guiding the target agent and the large language model to generate corresponding target activity pages for different activity types, it can effectively improve the stickiness of a specific section of an application at different times.

[0169] In step 420, the application to which the target activity belongs refers to the platform that displays the target activity page corresponding to the target activity. Different applications belong to different application fields or business scenarios. For example, the target activity may include product promotion rewards, video publishing rewards, document upload rewards, etc., and the application to which the target activity belongs may be an instant short messaging application, a short video application, an A document application, etc. By determining the application to which the target activity belongs, the selection of the target agent corresponding to the target activity can be based on the specific application context, thereby improving the accuracy of the activity page generation.

[0170] In step 430, determining a target agent from multiple candidate agents based on the application and activity type includes: determining a first score of the target activity based on the application; determining a second score of the target activity based on the activity type; determining a total score of the target activity based on the first score and the second score of the target activity; and determining a target agent from multiple candidate agents based on the total score of the target activity.

[0171] The first score of the target activity determined based on the application can be obtained by searching a first comparison table. Table 1 shows an example of the first comparison table.

[0172] application First Score Application A 100 Application B 80 Application C 60 Application D 40 other 20

[0173] Table 1

[0174] Based on the above example, the target activity belongs to application B. Looking up Table 1, the corresponding first score is 80; the target activity belongs to application E. Looking up Table 1, application E belongs to Other, and the corresponding first score is 20.

[0175] The second score of the target activity is determined based on the activity type and can be obtained by searching the second comparison table. Table 2 shows an example of the second comparison table.

[0176]

[0177]

[0178] Table 2

[0179] Based on the above example, if the activity type of the target activity is a holiday promotion activity, and Table 2 is searched, the corresponding second score is 100; if the activity type of the target activity is a daily promotion activity, and Table 2 is searched, the corresponding second score is 20.

[0180] Based on the first score and the second score of the target activity, a total score of the target activity is determined, which may be obtained by averaging or weighted averaging.

[0181] In one embodiment, the total score can be determined by taking the average of the first and second scores. For example, if the first score determined based on the application to which the target activity belongs is 60, and the second score determined based on the activity type of the target activity is 60, then the total score is (60+60) / 2=60. The advantage of using the average method to calculate the total score is that the activity type of the target activity and the application to which the target activity belongs have the same impact on the calculation of the total score, thereby improving the fairness of determining the target agent.

[0182] In another embodiment, the total score can be calculated using the weighted average of the first score and the second score. In this embodiment, it is first necessary to set weights for the first score corresponding to the application to which the target activity belongs and the second score corresponding to the activity type of the target activity. For example, the weight of the first score is 0.6, the weight of the second score is 0.4, the first score is 80, and the second score is 60, then the total score is 80*0.6+60*0.4=72. The advantage of using the weighted average method to calculate the total score is that different weights can be flexibly set for the first score corresponding to the application to which the target activity belongs and the second score corresponding to the activity type of the target activity according to the needs of the actual application, thereby improving the flexibility of determining the target intelligent agent.

[0183] Based on the total score of the target activity, the target agent is determined from multiple candidate agents, which can be found by searching the third comparison table. Table 3 shows an example of the third comparison table.

[0184] Total score Candidate Agents Less than or equal to 100 and greater than 75 Candidate Agent A Less than or equal to 75 and greater than 50 Candidate Agent B Less than or equal to 50 and greater than 25 Candidate Agent C Less than or equal to 25 and greater than 0 Candidate Agent D

[0185] Table 3

[0186] Based on the above example, the total score of the target activity is 70. Looking up Table 3, the corresponding target agent is candidate agent B; the total score of the target activity is 50. Looking up Table 3, the corresponding target agent is candidate agent C.

[0187] It is understandable that different agents may have different functions and characteristics, and different agents may be suitable for different application scenarios or business fields. Selecting a target agent suitable for the target activity from multiple candidate agents can better match the needs of the activity and ensure that the selected target agent has better applicability in a specific application scenario, thereby ensuring that the selected target agent has better performance and effect when processing the target activity. Therefore, after obtaining the total score of the target activity, the embodiment of the present disclosure can determine the most suitable target agent from multiple candidate agents. In the present disclosure, the accuracy is ensured by means of the coordination of the selected target agent's own capabilities and the capabilities of the large language model, and the degree of automation and efficiency of activity page generation can be improved.

[0188] In step 440, after the target agent is determined, the target activity description information is input into the target agent. In other words, detailed description information of the target activity (such as the activity name, time, location, content, purpose, participants, and activity process) is input into the target agent so that the target agent can understand and process this information and make appropriate decisions or actions.

[0189] The advantage of steps 410-440 is that the total score is determined based on the target activity's application and activity type. The scores for the target activity's application and activity type can be adjusted based on actual application needs, thereby determining the target agent corresponding to the target activity. This provides increased flexibility in determining the target agent. Thus, by understanding and processing the input target activity description information based on the determined target agent, the automation and accuracy of activity page generation can be improved.

[0190] Detailed description of step 320

[0191] In step 320, after the target agent receives the description information of the target activity, it does not directly give the description information of the target activity to the large language model, but first expands the description information to obtain the expanded information. It is understandable that due to the presence of professional terms in specific fields in specific applications, these professional terms are incomprehensible to the subsequent first large language model. In order for the first large language model to accurately understand the meaning expressed in the description information, it is necessary to input the explanation information corresponding to the professional terms in the specific field (such as glossary) into the first large language model to assist the first large language model in accurately identifying the complete activity copy information.

[0192] The expanded information refers to a collection of the original descriptive information of the target activity and the interpretation information related to the descriptive information output by the target agent. For example, the descriptive information for the target activity on virtual social platform A includes: "Decorate on the island three times to receive a reward." Since the first language model does not understand the specific meanings of "island" and "decorate," the target agent then expands the description to include: "The island refers to the subject's personal virtual space on virtual social platform A, similar to a virtual "island," where the subject can personalize their personal space, post updates, and share their life." The expanded information for "decorate" includes: "Decorate refers to the subject decorating and personalizing their personal space on the island, including changing the background, adding decorative items, and setting music." Therefore, the expanded information may include: "The subject decorates and personalizes their personal space on virtual social platform A three times, such as changing the background, adding decorative items, and setting music, to receive a reward."

[0193] In one embodiment, Figure 5 As shown, step 320 includes:

[0194] Step 510: Obtain a word definition library through the target agent;

[0195] Step 520: extract description keywords from the description information;

[0196] Step 530: Based on the description keyword, search the word interpretation library to obtain interpretation information of the description keyword;

[0197] Step 540: Add the interpretation information to the description information to obtain expanded information.

[0198] In step 510, the acquisition process of the target agent has been described in detail in the above embodiment and will not be repeated here to save space. The target agent understands and analyzes the input descriptive information of the target activity and determines the word interpretation library corresponding to the target activity. The word interpretation library refers to a database or dictionary containing information such as phrase interpretations, definitions, and examples of various vocabulary related to the target activity. Therefore, the word interpretation library can include vocabulary interpretations (such as the basic meaning, part of speech, pronunciation, meaning, and collocation of words), word examples (i.e., examples of various vocabulary to help the target agent understand the usage of the vocabulary in different contexts), synonyms and antonyms (i.e., synonyms and antonyms related to the target vocabulary to help the target agent better understand the semantic relationship of the vocabulary), vocabulary association information (i.e., information on the association of the vocabulary with other vocabulary, concepts, or topics to help the target agent perform semantic understanding and information retrieval), and professional interpretations of the vocabulary (i.e., professional interpretations, definitions, and examples of vocabulary in specific fields or professions to meet the needs of different fields). The word interpretation library can be used to help the target agent understand and process natural language, thereby better understanding the needs of the target activity.

[0199] It is understandable that in actual applications, the word definition library can be continuously updated and maintained according to a preset time period to reflect language changes and new vocabulary usage.

[0200] In one embodiment, Figure 6 As shown, step 510 includes:

[0201] Step 610: Determine the application to which the target activity belongs through the target agent;

[0202] Step 620: Obtain the word definition library of the application.

[0203] In step 610, to accurately determine the word definition library corresponding to the target activity, the target agent is first used to determine the application to which the target activity belongs. For example, if the description of the target activity contains keywords such as "video publishing" and "Application A," the target agent can determine that the application to which the target activity belongs is short video application A.

[0204] In step 620, the embodiment of the present disclosure may pre-set corresponding word definition libraries for different applications. Thus, after determining the application to which the target activity belongs, the word definition library of the application may be obtained by matching multiple word definition libraries according to the application to which it belongs.

[0205] In another embodiment, a knowledge graph can be pre-built based on multiple activities and corresponding word definition libraries. In this case, there is a correspondence between different activities and word definition libraries. In this way, the target agent can match the word definition library in the knowledge graph based on the target activity to determine the word definition library corresponding to the target activity.

[0206] In step 520, description keywords are phrases extracted from the target activity's description that represent key information about the activity. These description keywords help people quickly understand the theme and key points of the target activity description. For example, if the target activity description includes "Attach a high-definition, detailed image of a mooncake to the right of the activity description," the extracted description keywords would include "activity description text," "right side," "a piece," "mooncake," and "high-definition, detailed image."

[0207] It is understood that the process of extracting descriptive keywords from the description information may include steps such as text preprocessing, word segmentation, part-of-speech tagging, keyword extraction, irrelevant word filtering, keyword organization, and keyword weight calculation. The specific process of extracting descriptive keywords from the description information can be flexibly selected based on actual needs and is not specifically limited here. Therefore, through these steps, representative and important descriptive keywords can be extracted from the description information, providing a useful information foundation for further information processing and application.

[0208] It should be noted that text preprocessing refers to the process of removing punctuation, numbers, special characters, and other information from descriptions, as well as converting the text in the descriptions to lowercase. Text preprocessing helps standardize text formatting and reduce noise and interference. Word segmentation is the process of segmenting the text of the description into word sequences. Word segmentation helps convert the text of the description into a computer-processable format and prepares for subsequent keyword extraction. Part-of-speech tagging is the process of tagging the segmented words and identifying each word's part of speech (e.g., noun, verb, adjective, etc.). Part-of-speech tagging helps identify and extract descriptive keywords related to the topic of the description. Keyword extraction is the process of using keyword extraction algorithms (such as Term Frequency-Inverse Document Frequency (TF-IDF) and TextRank) to weight the words in the description and extract words with higher weights as descriptive keywords. Keyword extraction algorithms consider the frequency of words in the text and their importance in the corpus to determine descriptive keywords. Irrelevant word filtering is the process of removing words that are irrelevant to the topic of the description based on domain knowledge or specific requirements. For example, common meaningless words and general vocabulary are removed to retain keywords relevant to the topic. Keyword sorting refers to the process of filtering and sorting extracted keywords, removing duplicate keywords, and integrating related terms to obtain a clearer and more meaningful list of descriptive keywords. Keyword weighting is the process of calculating the weight or importance of keywords as needed to further determine the core keywords in the description information. For example, algorithms can be used to sort keywords to identify the most important keywords.

[0209] In step 530, after obtaining the description keyword and the word definition library, the target agent can search the word definition library based on the description keyword to obtain the definition information of the description keyword. The definition information refers to the text information related to the description keyword that the target agent matched in the word definition library. In other words, the target agent matches the description keyword with the candidate keywords in the word definition library. If a match is successful, the definition information corresponding to the candidate keyword is used as the definition information of the description keyword.

[0210] like Figure 7 As shown, the word interpretation library at this time is the obtained word interpretation library of application A, which stores multiple candidate keywords and corresponding interpretation information. For example, the interpretation information of the candidate keyword "ultra-clear details" is "4K resolution, i.e., 4096×2160 resolution", the interpretation information of the candidate keyword "high-definition details" is "1280x720 resolution", the interpretation information of the candidate keyword "standard-definition details" is "720x480 resolution", and the interpretation information of the candidate keyword "best picture quality" is "highest color saturation and highest clarity", etc. In this way, if the obtained description keyword is "ultra-clear details picture", it can be determined that the description keyword successfully matches the candidate keyword "ultra-clear details", and the interpretation information of the description keyword is "4K resolution, i.e., 4096×2160 resolution".

[0211] In step 540, after determining the interpretation information for the multiple description keywords, the interpretation information for the multiple description keywords is added to the original description information to obtain expanded information for the target activity. In this way, the interpretation information for the multiple description keywords can be added to the original description information and input into the first large language model, thereby ensuring the accuracy of the generated activity page based on the capabilities of the large language model.

[0212] The advantage of the above steps 510-540 is that by obtaining a word interpretation library and automatically extracting descriptive keywords, the impact of human factors on information can be reduced, thereby improving the accuracy of information processing. By searching the word interpretation library to obtain interpretation information for descriptive keywords, the descriptive information can be made richer and more detailed, providing more background knowledge and contextual information. Adding interpretation information to the descriptive information can enhance the semantic association of keywords in the descriptive information, making the information more coherent and easier to understand. Therefore, by using the target intelligent agent's own capabilities to expand the description of the target activity, the entire process is automated, reducing the need for manual operation, and effectively improving the degree of automation and efficiency of activity page generation.

[0213] Detailed description of step 330

[0214] In step 330, the first large language model is a model used to generate page generation requirement information for each sub-process of the target activity within the sub-process sequence. The first large language model is constructed based on the LLM model structure and can generate text output that conforms to language logic and semantic requirements based on input information. The disclosed embodiments can leverage the capabilities of large language models to generate sub-process sequence and page generation requirement information for the target activity that conforms to logic and semantics based on the expanded input information, thereby further guiding and improving the execution of the target activity and the generation of related target activity pages.

[0215] The sub-process sequence refers to the display order of multiple sub-processes corresponding to the target activity. Based on the sub-process sequence, the order of the multiple consecutive target activity pages that are finally generated can be obtained. In other words, a first target activity page containing the target activity is pushed to the target object on the target activity application. The target object triggers link 1 on the first target activity page to enter the second target activity page, and then triggers link 2 on the second target activity page to enter the third target activity page, and so on. In this case, the sub-process sequence can be understood as data that sorts out the relationship between the multiple sub-processes of the target activity and the target activity pages corresponding to the sub-processes.

[0216] Each sub-process corresponds to a target activity page, and the page generation requirement information corresponding to the sub-process is used to represent the detailed text information when the activity page corresponding to the sub-process is displayed. The page generation requirement information corresponding to the sub-process can include text information on key elements such as the purpose of the activity, the application to which the activity belongs, the activity links, the activity form, and the activity style. The activity links can represent the process steps required to execute the activity. The activity form refers to the implementation form of the activity, such as online activities, offline activities, and a combination of online and offline activities. The activity style refers to the presentation method and style of the target activity page corresponding to the target activity. The activity style can represent the style requirements of the target activity page in terms of page design, layout, color matching, interaction method, font style, etc. The activity style can also require the overall style of the page, such as cartoon style, watercolor style, etc. In other words, based on the page generation requirement information corresponding to the sub-process, complete text information related to the design of each target activity page can be obtained.

[0217] For example, application A needs to launch a Mid-Autumn Festival video release event on the application platform three days before the Mid-Autumn Festival. Figure 2A-2E As shown, the sub-process timing at this time is used to represent the timing relationship of the four sub-processes, that is, the four target activity pages, respectively as shown in Figure 2B 、 Figure 2C 、 Figure 2D and Figure 2E In this way, the sub-process sequence can represent the trigger Figure 2B You can enter the activity link in Figure 2C The page triggers Figure 2C You can enter the activity link in Figure 2D The page triggers Figure 2D You can enter the activity link in Figure 2E page.

[0218] In one embodiment, the expanded information is input into the first large language model to directly obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process.

[0219] In another embodiment, Figure 8 As shown, step 330 includes:

[0220] Step 810: Extract sub-process description information and page description information from the expanded information through the target agent;

[0221] Step 820: Input the sub-process description information into the first language model to obtain the sub-process time sequence;

[0222] Step 830: Input the sub-process sequence and page description information into the first language model to obtain page generation requirement information corresponding to the sub-process.

[0223] In step 810, the sub-process description information is a detailed description of each sub-process of the target activity. The sub-process description information of each sub-process may include detailed information such as the specific operations, time sequence, required resources, and related personnel of the corresponding sub-process, so that the application system or personnel can clearly understand and execute these sub-processes.

[0224] Page description information refers to the design requirements and specifications for pages associated with each sub-process of the target activity. Each sub-process's page description may include a detailed description of the corresponding sub-process's page layout, color scheme, interaction methods, content presentation format, and other aspects to ensure that the page meets the activity's requirements and provides a good user experience. Therefore, by extracting sub-process and page description information from the expanded information, the target agent can better understand and execute the target activity, ensuring that the activity page corresponding to each sub-process meets the activity's requirements.

[0225] In step 820, the sub-process description information is input into the first large language model to obtain the sub-process sequence. The sub-process sequence has been described in detail in the above embodiment.

[0226] In step 830, after determining the sub-process timing of the target activity, the target agent may input the sub-process timing and page description information into the first large language model to obtain page generation requirement information corresponding to the sub-process.

[0227] The advantage of steps 810-830 described above is that, by using the target agent to extract sub-process description information and page description information from the expanded information, automated information extraction and processing can be achieved, reducing human intervention and improving efficiency. By inputting the sub-process description information and page description information into the first large language model, richer and more detailed information related to the sub-process can be obtained, facilitating a more comprehensive understanding of the target activity and page generation requirements. Therefore, using the target agent and the first large language model for information extraction and processing reduces the impact of human factors on information and effectively improves the accuracy and consistency of information.

[0228] In another embodiment, the first large language model includes a first sub-large language model and a second sub-large language model. Both the first sub-large language model and the second sub-large language model are models constructed based on the model structure of LLM.

[0229] Based on this, Figure 9 As shown, step 330 includes:

[0230] Step 910: Input the expanded information into the first sub-large language model to obtain the sub-process sequence of the target activity and the page generation requirement information corresponding to the sub-process;

[0231] Step 920: Input the expanded information, sub-process timing and page generation requirement information into the second sub-large language model to obtain the revised sub-process timing and page generation requirement information.

[0232] In step 910, the expanded information is input into the first sub-large language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process. This process can be referred to in step 330 of the above embodiment, except that the first large language model is replaced with the first sub-large language model contained in the first large language model. This process will not be further described here. In other words, the capabilities of the first sub-large language model can be utilized to help parse and understand the input information, thereby extracting the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process from the expanded input information.

[0233] In step 920, since the first sub-large language model may have errors in understanding and processing the expanded information, the accuracy of the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process is affected. The disclosed embodiment can use the second sub-large language model to compare and optimize the information output by the first sub-large language model with the expanded information to improve the quality and consistency of the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process. In other words, the second sub-large language model can be used as a correction model. After the expanded information, sub-process timing and page generation requirement information are input into the second sub-large language model through the target intelligent agent, a more accurate and reliable revised sub-process timing and page generation requirement information can be obtained.

[0234] The advantage of the above steps 910 to 920 is that the first sub-large language model and the second sub-large language model are used to jointly determine the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process. The first sub-large language model serves as a model for initially identifying the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process, and the second sub-large language model serves as a model for correcting the output result of the first sub-large language model. Therefore, through the combination of multiple different large language models, the accuracy and reliability of the output sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process can be improved. In addition, the entire process does not require any human intervention, and the accuracy is guaranteed by means of the cooperation of the target intelligent agent's own capabilities and the large language model capabilities, thereby improving the degree of automation and efficiency of activity page generation.

[0235] Detailed description of the continuity adjustment of the sub-process in the embodiment of the present disclosure

[0236] Because the multiple sub-processes of the target activity derived from the first language model may have inconsistent context within the corresponding pages, for example, the previous sub-process may require an image to be displayed after triggering a link to the page, but the page for the subsequent sub-process does not contain an image. This means that the context of the two sub-processes is inconsistent, indicating a problem with the sequential continuity of the sub-processes, which requires adjustment.

[0237] Based on this, in one embodiment, Figure 10 As shown, after step 330, the activity page generation method of the embodiment of the present disclosure further includes:

[0238] Step 1010: For each sub-process, the target agent inputs the sub-process sequence, the page generation requirement information corresponding to the sub-process, the first page generation requirement information corresponding to a predetermined number of preceding sub-processes preceding the sub-process, and the second page generation requirement information corresponding to a predetermined number of subsequent sub-processes following the sub-process into a context continuity determination model to obtain a context continuity determination result for the sub-process.

[0239] Step 1020: Through the target intelligent agent, based on the context continuity determination result, adjust the page generation requirement information corresponding to the sub-process.

[0240] In step 1010, the page generation requirement information corresponding to the sub-process at this time refers to the page generation requirement information of the sub-process that currently needs to perform context continuity judgment. The predecessor sub-process refers to the sub-process that appears before the sub-process that currently needs to perform context continuity judgment. In other words, the page corresponding to the predecessor sub-process may appear on the application interface earlier than the page corresponding to the sub-process that needs to perform context continuity judgment. The subsequent sub-process refers to the sub-process that appears after the sub-process that currently needs to perform context continuity judgment. In other words, the page corresponding to the subsequent sub-process may appear on the application interface later than the page corresponding to the sub-process that needs to perform context continuity judgment.

[0241] The first page generation requirement information refers to the page generation requirement information of the sub-process that appears before the previous sub-process that needs to be judged for context continuity. The second page generation requirement information refers to the page generation requirement information of the subsequent sub-process that appears after the sub-process that needs to be judged for context continuity. Since the sub-process sequence can represent the interlaced relationship of the steps of each sub-process, when determining the context continuity of the sub-process that currently needs to be judged, it is necessary to simultaneously examine the page generation information of the sub-processes related to the previous and subsequent moments of the sub-process. At this time, the sub-process sequence, the page generation requirement information corresponding to the sub-process, the first page generation requirement information corresponding to a predetermined number of previous sub-processes before the sub-process, and the second page generation requirement information corresponding to a predetermined number of subsequent sub-processes after the sub-process are input into the context continuity determination model to obtain the context continuity determination result of the sub-process. The context continuity determination result of the sub-process is used to characterize the continuity of the sub-process currently being judged and the sub-processes related to the previous and subsequent moments.

[0242] It is understood that the predetermined number of preceding sub-processes and the predetermined number of subsequent sub-processes corresponding to a sub-process can be the same, or can be flexibly adjusted according to actual needs, and are not specifically limited here. For example, if it is necessary to examine the continuity of a sub-process within the entire sub-process sequence, then the predetermined number of preceding sub-processes corresponding to the sub-process is the sum of the number of sub-processes from the first sub-process to the previous sub-process of the current sub-process, and the predetermined number of subsequent sub-processes corresponding to the sub-process is the sum of the number of sub-processes from the next sub-process to the last sub-process of the current sub-process. If it is only necessary to examine the continuity of a sub-process and its three adjacent sub-processes, then the predetermined number of preceding sub-processes corresponding to the sub-process can be 3.

[0243] In step 1020, if the context continuity determination result for a sub-process indicates an anomaly in the sub-process's continuity with a preceding or subsequent sub-process, the target agent adjusts the page generation requirement information corresponding to the sub-process based on the context continuity determination result. Consequently, the adjusted page generation information for each sub-process ensures context continuity, thereby guaranteeing the accuracy of target activity page generation.

[0244] The advantage of the above steps 1010-1020 is that, through the target agent, the execution order of the sub-processes and the page generation requirements are determined for each sub-process based on the continuity of the context information, thereby ensuring that the generated pages are coherent and consistent in content and order. By using the target agent to determine the results based on the context continuity and adjust the page generation requirement information corresponding to the sub-process, the generated page can be made more in line with the needs of the target activity, thereby improving the accuracy and quality of page generation. In addition, by making adjustments through the target agent, the page generation requirement information is automatically adjusted, saving the time and cost of manual intervention and improving the efficiency of page generation.

[0245] Detailed description of step 340

[0246] In step 340, after determining the text in the activity page of each sub-process according to the first large language model, by target intelligent agent, based on the information after expansion, select picture post-processing model, and utilize the picture post-processing model of selection, generate material picture under the guidance of the information after expansion.The picture post-processing model refers to the model of the material picture needed in the target activity page that is used to generate the target activity.That is to say, utilize selected picture post-processing model, under the guidance of the information after expansion, can generate material picture, to ensure that the material picture generated is associated with the information after expansion.And, on the target activity page that finally displays, add material picture, can make the activity page have certain creativity and personalization, help to promote the attractiveness of page and information expression effect, thereby improve object stickiness.The material picture at this moment can be all material pictures that comprise in the sub-process sequential order.

[0247] In one embodiment, the image post-processing model is equivalent to a filter model. Combined with the target agent and the large language model, it can generate material images that meet the requirements of the target activity. The image post-processing model includes a first image post-processing model based on the image style and a second image post-processing model based on the application. The first image post-processing model and the second image post-processing model can be models built based on LoRA. Image styles include cartoon style, watercolor style, anime style, etc. The application has been described in detail in the above embodiment and will not be repeated here.

[0248] It should be noted that all large language models and image post-processing models in the disclosed embodiments can determine training samples based on their application in specific scenarios and complete model training based on these training samples. The disclosed embodiments use the trained models to generate the target activity page. To save space, the model training process is not described in detail.

[0249] Based on this, Figure 11 As shown, the target agent selects an image post-processing model based on the expanded information, including:

[0250] Step 1110: Input the expanded information into the third language model through the target agent to obtain image style keywords;

[0251] Step 1120: Select a first image post-processing model from a plurality of first candidate image post-processing models based on the image style keyword;

[0252] Step 1130: Determine the application to which the target activity belongs;

[0253] Step 1140: Select a second image post-processing model from a plurality of second candidate image post-processing models based on the application.

[0254] In step 1110, the third language model is used to extract image style keywords from the expanded information. This can be the first or second language model, or another language model other than the first and second language models. Image style keywords are keywords that characterize the style of the generated source image. Image style keywords include: abstract, realistic, landscape, figure, still life, documentary, art, black and white, color, retro, fashion, nature, city, abstract, fantasy, science fiction, etc.

[0255] In step 1120, the first picture post-processing model refers to a model used to adjust the picture style of the target activity. The second candidate picture post-processing model refers to a processing model corresponding to each pre-set picture style. For example, if the picture style of the target activity is science fiction, then the corresponding first picture post-processing model may be a model that can adjust the picture to a science fiction style picture. Since one picture style can correspond to one first candidate picture post-processing model, after determining the picture style keyword of the target activity, based on the picture style keyword, a first picture post-processing model that matches the picture style keyword is selected from multiple first candidate picture post-processing models. It can be understood that if there are multiple picture style keywords, then multiple first picture post-processing models can be selected from multiple first candidate picture post-processing models. Figure 12As shown, after the target agent inputs the expanded information into the third language model, the resulting image style keywords include "animation" and "ultra-high-definition details." Based on these two image style keywords, a model can be selected from multiple first candidate image post-processing models, determining the first candidate post-processing model corresponding to "animation" and the first candidate post-processing model corresponding to "ultra-high-definition details." These two first candidate post-processing models can then be used as the first image post-processing model.

[0256] In step 1130, the application to which the target activity belongs is determined through the target agent, and the specific process of determination has been described in detail in the above step 610 and will not be repeated here.

[0257] In step 1140, the second image post-processing model refers to a model for adding information related to the application to which the target activity belongs to the image. The second candidate image post-processing model refers to a pre-set processing model corresponding to the application to which each image belongs. The application-related information of the target activity can be an application tag, or a specific style corresponding to the application, which is not specifically limited here. For example, if the application-related information of the target activity is an application tag, then the corresponding second image post-processing model can be a model that can adjust the image to a picture with the application tag corresponding to the application to which the target activity belongs added. Figure 13 As shown, second candidate image post-processing models corresponding to applications A to F are predefined. If the target agent determines that the application to which the target activity belongs is application B, then based on application B, a model is selected from multiple second candidate image post-processing models, and the second candidate image post-processing model corresponding to application B can be used as the second image post-processing model.

[0258] The advantage of the above steps 1110 to 1140 is that, through the joint action of the target intelligent agent and the third language model, more accurate and comprehensive picture style keywords can be obtained, which helps to more accurately describe and identify the style characteristics of the picture. Based on the picture style keywords and the application to which the target activity belongs, the first picture post-processing model for changing the picture style and the second picture post-processing model for superimposing application tags are determined. If so, the most suitable post-processing model can be selected according to the needs of the application and the characteristics of the picture style keywords, thereby improving the effect and quality of picture processing. Therefore, the embodiment of the present disclosure can comprehensively consider the style characteristics, application requirements and post-processing effects of the target activity when adding pictures to the activity page, thereby improving the accuracy and applicability of the generated material pictures.

[0259] In one embodiment, Figure 14 As shown, the selected image post-processing model is used to generate material images under the guidance of the expanded information, including:

[0260] Step 1410: Input the expanded information into a stable diffusion model to obtain a material image base;

[0261] Step 1420: Use the image post-processing model to perform post-processing on the material image base to obtain the material image.

[0262] In step 1410, the stable diffusion model refers to a model used to generate a material image base. The material image base refers to an image that contains the basic structure and form of the material image. The material image base provides a foundation for subsequent image post-processing. Generating the material image base through the stable diffusion model can control the basic structure and form of the final material image to a certain extent, providing a good starting point for subsequent post-processing. Therefore, the style in the material image base is not specifically limited, because the image post-processing model will be used to change the image later. It should be noted that the stable diffusion model refers to a model constructed based on the SD algorithm.

[0263] In one embodiment, Figure 15 As shown, step 1410 includes:

[0264] Step 1510: extracting positive guidance keywords and negative guidance keywords from the expanded information;

[0265] Step 1520: Convert the forward guidance keyword into a first guidance vector, and convert the reverse guidance keyword into a second guidance vector;

[0266] Step 1530: Input the initial basis vector into the stable diffusion model to obtain the material image basis vector under the guidance of the first guide vector and the second guide vector;

[0267] Step 1540: Convert the material image base vector into the material image base.

[0268] In step 1510, the forward guidance keywords refer to keywords related to the expected results or the characteristics of the required material pictures. The forward guidance keywords may include keywords that describe the content, characteristics or style of the required material pictures. In the subsequent processing process, these keywords may be used to guide the model to generate material pictures that are more in line with expectations. The reverse guidance keywords refer to keywords that are contradictory or opposite to the characteristics of the required material pictures. The reverse guidance keywords may be used to help the model avoid generating material picture content or characteristics that are contradictory to these keywords. By using forward guidance keywords and reverse guidance keywords, the stable diffusion model can be guided to a certain extent, so that it can understand the characteristics of material pictures that are more in line with expectations, so as to obtain a more accurate material picture base, thereby improving the accuracy of the generated material pictures. This guidance method can control the generation process of the material pictures to a certain extent, so that the generated material pictures are more in line with expectations. As Figure 16 As shown, the forward guidance keywords extracted from the expanded information may include "masterpiece", "animation", "best quality", "ultra-clear details", and these keywords are used to characterize the image features that the material image to be generated contains. The reverse guidance keywords extracted from the expanded information may include "low resolution", "text", "worst quality", "blur", etc. These keywords are used to characterize the image features that the material image to be generated should avoid containing. In this way, the material image base is obtained by guiding the stable diffusion model based on the forward guidance keywords and the reverse guidance keywords.

[0269] In step 1520, a positive guide keyword corresponds to a first guide vector, and a negative guide keyword corresponds to a second guide vector. In one embodiment, the first guide vector corresponding to the positive guide keyword and the second guide vector corresponding to the negative guide keyword can be determined by table lookup. In another embodiment, if the positive guide keyword or the negative guide keyword is a character-type keyword, the character-type keyword can be input into the embedding layer to obtain the guide vector corresponding to the character-type keyword; if the positive guide keyword or the negative guide keyword is a numerical feature, the numerical feature is numerically mapped to obtain the guide vector corresponding to the numerical feature. In this embodiment, the advantage of using different vectorization methods for different types of positive guide keywords and negative guide keywords is that the features can be vectorized according to the characteristics of different types of guide keywords, thereby ensuring the accuracy of the feature vectorization results.

[0270] A vector is an array of values ​​in different dimensions, and it is a point in a multidimensional space. The line segment between the point and the origin in the multidimensional coordinate system has a size and direction, which are the size and direction of the vector. Each of the above values ​​is the point value projected on the corresponding coordinate axis in the multidimensional coordinate system, that is, a vector element. Vector elements can be numerical values ​​or symbols. For example, the vector elements of the first guide vector and the second guide vector in the embodiment of the present disclosure can be numerical values. Figure 17As shown in the figure, there are four positive guidance keywords, namely "masterpiece", "animation", "best quality", and "ultra-clear details". After vectorizing these four positive guidance keywords, the first guidance vector corresponding to "masterpiece" is [1, 2, 2, 0, 2, 1, 0, 0], the first guidance vector corresponding to "animation" is [5, 8, 6, 0, 2, 1, 0, 0], the first guidance vector corresponding to "best quality" is [10, 20, 30, 40, 68, 50, 60, 10], and the first guidance vector corresponding to "ultra-clear details" is [10, 40, 50, 30, 20, 76, 90, 100]. There are six negative guidance keywords, namely "low resolution", "text", "missing fingers", "extra digits", "worst quality", and "normal quality". After vectorizing these six reverse guidance keywords, the second guidance vector corresponding to “low resolution” is [58, 54, 42, 30, 44, 54, 64, 68], the second guidance vector corresponding to “text” is [10, 40, 50, 30, 20, 80, 90, 104], the second guidance vector corresponding to “missing fingers” is [5, 6, 8, 9, 10, 13, 5, 32], the second guidance vector corresponding to “extra digits” is [505, 76, 68, 97, 10, 13, 55, 320], the second guidance vector corresponding to “worst quality” is [203, 33, 53, 66, 9, 9, 31, 84], and the second guidance vector corresponding to “normal quality” is [99, 23, 24, 18, 6, 8, 20, 54].

[0271] In step 1530, the initial basis vector can be any vector. This is because the stable diffusion model, guided by the guide vector, can transform any initial basis vector into a material image basis vector. In other words, using the stable diffusion model to generate an image under the guidance of the first and second guide vectors can ensure that the generated material image is more consistent with expectations. The material image basis vector refers to the vector form of the material image basis.

[0272] The stable diffusion model includes a diffusion noise addition process (i.e., the forward process) and a diffusion denoising process (i.e., the reverse process). During the model training process of the stable diffusion model, noise can be gradually added to the original image vector according to a fixed distribution. In this process, the information of the original image is gradually replaced by noise, which is similar to diffusion in space. After a preset number of steps, a vector of a Gaussian-distributed noise image is obtained. The diffusion denoising operation is then performed on the vector of the image in the last step. The noise corresponding to the number of steps in the diffusion noise addition process is predicted by the neural network, and the image is gradually restored to its original state. If so, the performance of the stable diffusion model can be determined by comparing the restored image vector with the original image vector at the beginning.

[0273] In one embodiment, Figure 18 As shown, step 1530 includes:

[0274] Step 1810: Initialize the base vector to be diffused as the initial base vector, and initialize the step number to 1;

[0275] Step 1820: Input the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model to obtain the diffusion noise corresponding to the step number;

[0276] Step 1830: Use the diffusion noise corresponding to the step number to offset the base vector to be diffused, increase the step number by 1, and return to the step of inputting the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model until the step number increases to the preset maximum number of steps.

[0277] In step 1810, the initialization step number refers to the initial step number of the reverse process of the stable diffusion model. It should be understood that the total number of steps in the forward process and the reverse process of the stable diffusion model are the same. For example, if the reverse process of the stable diffusion model has a total of 100 steps, the initialization step number is 1, and the preset maximum number of steps is 100. The basis vector to be diffused refers to the vector currently requiring stable diffusion. When the step number is initialized, the basis vector to be diffused is the initial basis vector.

[0278] In step 1820, the base vector to be diffused, the step number, the first guiding vector, and the second guiding vector are input into the stable diffusion model. Specifically, the base vector to be diffused is concatenated with the step number so that the stable diffusion model knows which step the base vector to be diffused belongs to. Simultaneously, the guidance provided by the first and second guiding vectors is combined to generate the diffusion noise corresponding to the step number.

[0279] In step 1830, after determining the diffusion noise corresponding to the step number, the diffusion noise corresponding to the step number is used to offset the base vector to be diffused. Using the diffusion noise corresponding to the step number to offset the base vector to be diffused can be understood as the reverse operation of the diffusion noise addition process, that is, offsetting the diffusion effect of the noise to update the base vector to be diffused. In this way, the base vector to be diffused can be guided to gradually approach the base vector of the material image. The step number is increased by 1, and the process returns to the step of inputting the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model (i.e., step 1820) until the step number is increased to the preset maximum number of steps. At this point, the base vector to be diffused is the base vector of the material image.

[0280] The advantage of the above steps 1810 to 1830 is that by using the first guiding vector and the second guiding vector to guide the stable diffusion model to generate the material image base vector, the generation process can be controlled to a certain extent, so that the generated material image is more in line with expectations.

[0281] In step 1540, the material image basis vector is converted into the final material image basis. The process of converting the vector into an image can use methods such as inverse mapping, neural network decoding, and generative adversarial network (GAN) to achieve the conversion of the material image basis vector into the material image basis.

[0282] The advantage of steps 1510-1540 is that the introduction of guiding keywords and the vectorization process can help the model better understand and apply the augmented input information. Using the first and second guiding vectors to guide the stable diffusion model in generating the base vectors for the source image can control the generation process to a certain extent, ensuring that the generated source image is more consistent with expectations.

[0283] In step 1420, the image post-processing model is used to perform post-processing on the source image base to obtain the source image. The image post-processing model is a model that adjusts color, contrast, sharpening, and adds filters and special effects to the source image base to obtain the final source image. The image post-processing model includes a first image post-processing model based on image style and a second image post-processing model based on application. Post-processing includes style changes and application markup overlays. The first and second image post-processing models have been described in detail in the above embodiments and will not be repeated here to save space.

[0284] Based on this, Figure 19 As shown, step 1420 includes:

[0285] Step 1910: using the first image post-processing model to change the style of the material image base;

[0286] Step 1920: Using the second image post-processing model, apply a marker to the base of the material image after the style is changed to obtain the material image.

[0287] In step 1910, the source image base is input into the first image post-processing model, and the style of the source image base is changed to obtain a source image base with a changed style. If there are multiple first image post-processing models, the styles of the source image base can be changed sequentially by the multiple first image post-processing models in a random order to obtain a source image base with a changed style.

[0288] In step 1920, the stylized image base is input into the second image post-processing model, and the application mark is superimposed on the stylized image base to obtain the image base. For example, if the second image post-processing model is the second candidate image post-processing model corresponding to application B, the application mark of application B will be superimposed on the stylized image base according to the second image post-processing model.

[0289] The advantage of the above steps 1910-1920 is that the style of the material image base is changed through the first image post-processing model, and the application mark is superimposed on the material image base after the style change through the second image post-processing model, so that the generated material image can better meet the expected style and application mark requirements.

[0290] In one embodiment, after obtaining the source image, the source image and page generation requirement information can be input into a third image post-processing model to obtain a revised source image. The third image post-processing model then serves as a model for the output of the revised image post-processing model, thereby improving the accuracy of the generated source image.

[0291] The advantage of steps 1410-1420 above is that by using the image post-processing model to post-process the material image base, the material image base can be adjusted and optimized according to specific needs, so that the final material image meets the requirements of the target activity. The entire process combines the stable diffusion model and the image post-processing model, which can, to a certain extent, ensure that the generated material images have a certain quality and style, while also having a certain degree of flexibility and controllability. The entire process does not require any human intervention, and the accuracy is guaranteed by the combination of the target agent's own capabilities and the large language model capabilities, thereby improving the automation and efficiency of activity page generation.

[0292] Detailed description of step 350

[0293] In step 350, through the target agent, for each sub-process in the sub-process sequence, the page generation requirement information corresponding to the sub-process is input into the second largest language model to obtain the activity page corresponding to the sub-process, and the material image is added to the activity page, thereby generating the target activity page. The second largest language model refers to the large language model for generating the activity page corresponding to the sub-process. The second largest language model here can be the first largest language model, or it can be another large language model other than the first largest language model. At this time, the activity page corresponding to the generated sub-process is a page that only contains text, numbers, and other special symbols. In other words, the activity page corresponding to the generated sub-process does not contain images.

[0294] In one embodiment, the page generation requirement information includes management-side page generation requirement information and client-side page generation requirement information, and the active page includes a management-side active page and a client-side active page.

[0295] Management page generation requirements refer to the requirements for generating the backend management interface for the target activity. These requirements may include design requirements for the backend management page (e.g., layout, functional modules, etc.), data presentation, and permission management. When developing management pages, it's important to understand the functionality and interface design required by administrators to better meet their needs. The management activity page refers to the page presented on the backend management interface for the target activity.

[0296] Client page generation requirements refer to the requirements for generating the target activity's application frontend interface. These requirements may include frontend page design requirements (such as responsive design and interactive methods), data presentation methods, and user experience. When developing client pages, it's important to understand the target audience's needs and expectations in order to provide pages and functionality that meet their expectations. The client activity page refers to the page that the target activity presents on the application frontend interface.

[0297] In one embodiment, Figure 20 As shown, the page generation requirement information corresponding to the sub-process is input into the second language model to obtain the activity page corresponding to the sub-process, including:

[0298] Step 2010: Input the client page generation requirement information corresponding to the sub-process into the second language model to obtain the client activity page corresponding to the sub-process;

[0299] Step 2020: Input the management-end page generation requirement information corresponding to the sub-process into the second language model to obtain the management-end activity page corresponding to the sub-process.

[0300] In step 210, when generating the client activity page, the client page generation requirement information corresponding to the sub-process can be input into the second largest language model to obtain the client activity page corresponding to the sub-process. Therefore, for multiple sub-processes included in the sub-process sequence, the client activity page corresponding to each sub-process can be obtained using the second largest language model.

[0301] In one embodiment, Figure 21 As shown, step 2010 includes:

[0302] Step 2110: Obtain the first domain-specific language rule of the client;

[0303] Step 2120: Input the client page generation requirement information corresponding to the sub-process and the first domain-specific language rules into the second language model to obtain the first domain-specific language corresponding to the sub-process;

[0304] Step 2130: Utilize the domain specific language interpreter to render the first domain specific language into a client active page.

[0305] In step 2110, the first domain-specific language rules are language rules pre-defined for the client in a specific domain. The first domain-specific language rules may include computer language designed for terminology, rules, grammar, etc. specific to the business domain. Using the first domain-specific language rules, it is possible to ensure that the generated page meets the requirements of the client in the specific domain.

[0306] In step 2120, the client page generation requirement information corresponding to the sub-process and the first domain-specific language rules are input into the second large language model to obtain the first domain-specific language corresponding to the sub-process. The output of the second large language model is a computer language model. In this case, the first domain-specific language refers to a domain-specific language that can fully describe the client active page. This is because the disclosed embodiments can leverage the powerful language understanding and generation capabilities of the second large language model to generate the code language of the sub-process that conforms to the domain-specific language rules.

[0307] In step 2130, the client active page can be an H5 active page, i.e., a web page developed based on HTML5 (H5) technology. H5 active pages typically feature rich visual effects, interactivity, and animation, providing a more vivid and engaging page experience. A domain-specific language interpreter is a tool that can render a domain-specific language into an actual executable page. Therefore, through the target agent, using the domain-specific language interpreter, the first domain-specific language can be rendered into a client active page.

[0308] The advantage of the above steps 2110-2130 is that the use of domain-specific language rules can ensure that the generated client active page meets the requirements of the specific domain, thereby improving the accuracy and quality of page generation. The use of the second largest language model can help understand and generate text or code that meets the language rules of the specific domain, thereby improving the automation and intelligence level of the generation process. The use of a domain-specific language interpreter can render the language rules of a specific domain into executable pages, simplifying the page generation process and improving efficiency. In general, the entire process combines domain-specific language rules, a large language model, and a domain-specific language interpreter to help achieve a more intelligent, accurate, and efficient client page generation process.

[0309] In step 220, when generating the management-side active page, the management-side page generation requirements corresponding to the sub-process can be input into the second largest language model to obtain the management-side active page corresponding to the sub-process. Therefore, for multiple sub-processes included in the sub-process sequence, the management-side active page corresponding to each sub-process can be obtained using the second largest language model.

[0310] In one embodiment, Figure 22 As shown, step 2020 includes:

[0311] Step 2210: Obtain the second domain-specific language rule of the management end;

[0312] Step 2220: Input the management-end page generation requirement information corresponding to the sub-process and the second domain-specific language rules into the second language model to obtain the second domain-specific language corresponding to the sub-process;

[0313] Step 2230: Utilize the domain specific language interpreter to render the second domain specific language into a management end active page.

[0314] In step 2210, the second domain-specific language rules refer to language rules pre-defined for the administrator in a specific domain. The second domain-specific language rules may include computer language designed for terminology, rules, grammar, etc. specific to the business domain. The second domain-specific language rules ensure that the generated pages meet the requirements of the administrator in the specific domain.

[0315] In step 2220, the management-side page generation requirements corresponding to the sub-process and the second domain-specific language rules are input into the second large language model to obtain the second domain-specific language corresponding to the sub-process. In this case, the second domain-specific language refers to a domain-specific language that can fully describe the management-side active page. This is because the disclosed embodiments can leverage the powerful language understanding and generation capabilities of the second large language model to generate code language for the sub-process that conforms to the domain-specific language rules.

[0316] In step 2230, the target agent uses a domain specific language interpreter to render the second domain specific language into a management terminal active page. Figure 23 As shown, the management-side page generation requirement information is for generating a demand sheet containing multiple contract information. The management-side page generation requirement information corresponding to the sub-process and the second domain-specific language rules are input into the second language model to obtain the second domain-specific language 2310 corresponding to the sub-process. In this way, the second domain-specific language can be rendered into the management-side active page 2320 using the domain-specific language interpreter.

[0317] It should be noted that the management end can clearly see the operations of the objects in the client, that is, it can query the real-time data of the objects in the client and perform real-time analysis on the object data in the client.

[0318] The advantage of the above steps 2210-2230 is that the use of domain-specific language rules can ensure that the generated management-side active page meets the requirements of the specific domain, thereby improving the accuracy and quality of page generation. The use of the second largest language model can help understand and generate text or code that meets the language rules of the specific domain, thereby improving the automation and intelligence level of the generation process. The use of a domain-specific language interpreter can render the language rules of a specific domain into executable pages, simplifying the page generation process and improving efficiency. In general, the entire process combines domain-specific language rules, a large language model, and a domain-specific language interpreter to help achieve a more intelligent, accurate, and efficient client page generation process.

[0319] The advantage of the above steps 2010 to 2020 is that the client page generation requirement information and the management end page generation requirement information corresponding to the sub-process are respectively input into the second large language model, which can flexibly generate activity pages for different ends. The embodiment of the present disclosure renders a deliverable page by fine-tuning the domain-specific language and combining it with a low-code engine. The entire process does not require any human intervention, and the accuracy is guaranteed by the combination of the target intelligent agent's own capabilities and the large language model capabilities, thereby improving the degree of automation and efficiency of activity page generation.

[0320] In one embodiment, Figure 24 As shown, add the material image to the activity page to generate the target activity page, including:

[0321] Step 2410: Generate requirement information from the page, obtain the location where the material picture is to be added, and the corresponding relationship information between the location where the material picture is to be added and the material picture;

[0322] Step 2420: Add the material picture corresponding to the material picture adding position on the activity page according to the corresponding relationship information.

[0323] In step 2410, the page generation requirement information refers to the requirement information for generating the current active page. The material image addition location refers to the location within the active page where the material image is added. The material image addition location determines the active page to which the material image is added and its specific location within the corresponding active page. The material image addition location corresponds to the material image, ensuring that the material image does not overlap or conflict with other images.

[0324] Therefore, by obtaining the location information for adding material images, it is possible to ensure that the location where the material images need to be added is accurately found in the activity page. By obtaining the corresponding relationship information between the material images and the locations where the material images are added, it is possible to ensure that each material image is placed in the correct location to avoid confusion or errors.

[0325] In step 2420, after determining the location where the material picture is added and the corresponding relationship information between the location where the material picture is added and the material picture, the material picture corresponding to the location where the material picture is added is added to the location where the material picture is added on the activity page according to the corresponding relationship information, thereby ensuring that the material picture is correctly placed at the designated location on the activity page. Figure 25 As shown, for example, the page generation requirement information is "a little girl in an animated style sitting and playing guitar, and the image of the little girl should be placed in the right area of ​​the page." After obtaining the material image 2510, the target agent can use the page generation requirement information to obtain the corresponding relationship information between the material image addition location "the image should be placed in the right area of ​​the page" and the material image 2510, that is, the material image 2510 should be in the right area of ​​the page. In this way, the material image corresponding to the material image addition location is added to the material image addition location of the generated activity page 2520 according to the corresponding relationship information, resulting in the target activity page 2530 after the material image is added.

[0326] The advantage of the above steps 2410-2420 is that by clearly specifying the location where each material picture is to be added, the efficiency of page generation can be improved, and the workload of subsequent adjustments and modifications can be reduced. According to the correspondence information, the material pictures are added to the specified locations to avoid errors or confusion. Therefore, in the embodiment of the present disclosure, during the target activity page generation process, the material pictures are accurately placed at the specified locations based on the correspondence information between the material picture addition location and the material picture, which can effectively improve the accuracy and efficiency of page generation.

[0327] Detailed description of generating a data report page in the embodiment of the present disclosure

[0328] In one embodiment, Figure 26 As shown, after step 350, the activity page generation method of the embodiment of the present disclosure further includes:

[0329] Step 2610: Send the target format information of the response data and key indicators to the application front end through the target agent;

[0330] Step 2620: Obtain, through the application front end, response data of the platform object to the target activity page within a predetermined time period after the target activity page is generated, and obtain key indicators based on the response data;

[0331] Step 2630: Send the response data and key indicators to the target agent based on the target format information through the application front end;

[0332] Step 2640: Input the response data and key indicators into the fourth language model through the target intelligent agent to obtain a data report page.

[0333] In step 2610, the target intelligent agent will send the target format information of the response data and key indicators to the application front end to ensure that the application front end can accurately understand and process the data. Key indicators refer to the indicators that the application front end needs to view. These key indicators may include visits, click-through rates, conversion rates, etc. Response data refers to the data generated after the platform object responds to the target activity page. The target format information may include the structure of the data, field name, data type, etc. There are differences between the application front end and the client and management end. The application front end is used to view the data of the key indicators generated by the client. The management end can view the data of all indicators generated by the client. Therefore, by sending the target format information, it can be ensured that the application front end can correctly parse and process the response data and key indicators, avoid problems caused by data format errors, and reduce the workload of data parsing and conversion. It should be noted that the application front end can be a BI system or other, which is not specifically limited here.

[0334] In step 2620, due to the data recycling of the application front end, it is not recycled immediately, but periodically. Therefore, through the application front end, the response data of the platform object to the target activity page is obtained in a predetermined time period after the target activity page is generated, and the key indicators are obtained based on the response data. The predetermined time period can be flexibly set according to actual needs. For example, the predetermined time period is 1 minute, that is, after the target activity page is generated, the response data of the platform object to the target activity page will be obtained once every 1 minute, and the key indicators are obtained based on the response data. Therefore, by obtaining the response data within the predetermined time period after the target activity page is generated, the performance and effect of the activity page can be understood in a timely manner. In addition, the key indicators obtained can provide a basis for subsequent data analysis, helping to evaluate the effect and optimization direction of the activity page.

[0335] In step 2630, the application front end sends the acquired response data and key indicators to the target agent based on the target format information. This ensures that the target agent can accurately receive and process the data.

[0336] In step 2640, the target agent inputs the received response data and key indicators into the fourth language model to generate a data report page. The fourth language model can be one of the first, second, and third language models, or a language model other than the first, second, and third language models. This data report page can include various charts, data analysis results, and the like to help users better understand and analyze the effectiveness of the activity page.

[0337] In one embodiment, Figure 27 As shown, step 2640 includes:

[0338] Step 2710: Input the response data and key indicators into the fourth language model through the target agent to obtain the third domain-specific language;

[0339] Step 2720: Utilize the domain specific language interpreter to render the third domain specific language into a data report page.

[0340] In step 2710, the third domain-specific language is a domain-specific language that can fully describe the response data and key indicators in the data report page. This is because the embodiment of the present disclosure can utilize the powerful language understanding and generation capabilities of the fourth language model to accurately generate code language that can represent the data report page.

[0341] In step 2720, the target agent uses a domain-specific language interpreter to render the third domain-specific language into a data report page. In this way, the application front end can view the data report page containing the response data and key indicators.

[0342] The advantage of steps 2710-2720 above is that using the fourth language model can help understand and generate text or code that conforms to domain-specific language rules, improving the automation and intelligence of the generation process. Using a domain-specific language interpreter, domain-specific language rules can be rendered as executable pages, simplifying the page generation process and improving efficiency. Overall, the entire process, combining the large language model and domain-specific language interpreter, can help achieve a more intelligent, accurate, and efficient data report page generation process.

[0343] The advantage of steps 2610-2640 above is that, based on the target format information, response data and key indicators are sent to the target agent, ensuring accurate data transmission and reception. The fourth language model enables automated data analysis and report generation, saving labor costs and time. Furthermore, the generated data report page can intuitively display the effectiveness of the activity page and key indicators, helping users better understand the data.

[0344] Detailed description of the framework process for generating an activity page in the embodiment of the present disclosure

[0345] Before the emergence of generative AI capabilities such as LLM and SD, the entire process of generating activity pages and finally forming data reports for related technologies involved multiple links. Figure 28 As shown, the entire process may include drafting page generation requirements, outputting interaction drafts, outputting design drafts, developing client-side activity pages, setting up the management interface, collecting activity data, and generating data reports. The interaction draft refers to the interaction between the object and the application platform. For example, after the object clicks a link on the page and completes the relevant information, the application platform will provide feedback, and the object will perform a series of subsequent operations based on this feedback, thus achieving interaction. The design draft refers to the design of the page when the interaction between the object and the application platform is completed. Because each process link takes a certain number of days, excluding backend time consumption, it takes approximately 23 days from planning to launch. However, this approach requires manual intervention at each link and cannot cover the entire chain, resulting in relatively limited efficiency in activity page generation.

[0346] The disclosed embodiments can combine the current generative capabilities, by leveraging the capabilities of the target agent itself and the capabilities of the large language model, to open up the factor production capabilities of the entire process of the operation scenario and truly improve the efficiency of the entire process. Figure 29 As shown, by using the target agent in conjunction with the first language model, the sub-process sequence and the page generation requirement information of each sub-process can be generated. By using the target agent in conjunction with the image post-processing model, the material image added to the active page can be accurately obtained. By using the target agent in conjunction with the second language model, the active page of the sub-process can be obtained, and the corresponding material image can be added to the active page to generate the target active page. When forming a data report, the target agent and the fourth language model can be used in conjunction to form a data report page required by the application front end.

[0347] From this, it can be seen that the disclosed embodiment can closely combine the capabilities of the target intelligent body with the operation scenario, open up the automated production of production factors in the entire chain, avoid the missing elements caused by information communication in each link, and thus avoid problems such as inconsistent understanding of information, thereby improving the degree of automation and efficiency of activity page generation.

[0348] In the specific experiments on the embodiments of the present disclosure and related technologies, it can be known that the activity page generation method provided by the embodiments of the present disclosure can save 30%-50% of the development time at the front end, and improve the overall efficiency of the whole process by at least 30%. This is also because the entire process of activity page generation involves the participation of the target intelligent agent, which links each link together to form a complete business closed loop, thereby reducing the problem of cognitive inconsistency in each link. Moreover, according to the coordinated use of the target intelligent agent and the large language model, the pages of the client and the management end can be directly generated, which increases the creative ability of the operators. Therefore, the entire process of the embodiment of the present disclosure does not require any human intervention, and the accuracy is guaranteed by means of the coordination of the target intelligent agent's own capabilities and the large language model capabilities, thereby improving the degree of automation and efficiency of activity page generation.

[0349] Description of the apparatus and device of the present disclosure

[0350] It is to be understood that, although the steps in the above-mentioned flowcharts are shown in sequence according to the arrow representations, these steps are not necessarily performed in sequence according to the order represented by the arrows. Unless otherwise specified in the present embodiment, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0351] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on the descriptive information of the target activity and other data related to the characteristics of the target activity, the permission or consent of the organizer of the target activity will be obtained first, and the collection, use and processing of such data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain the descriptive information of the target activity, the separate permission or consent of the organizer of the target activity will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the separate permission or consent of the organizer of the target activity, the necessary descriptive information of the target activity for the normal operation of the embodiment of the present application will be obtained.

[0352] Figure 30 This is a schematic diagram of the structure of the activity page generation device 3000 provided in an embodiment of the present disclosure. The activity page generation device 3000 includes:

[0353] The first input unit 3010 is used to input the description information of the target activity into the target agent;

[0354] The information expansion unit 3020 is used to expand the description information through the target agent to obtain expanded information;

[0355] The second input unit 3030 is used to input the expanded information into the first large language model to obtain the sub-process sequence of the target activity and the page generation requirement information corresponding to the sub-process;

[0356] The image generation unit 3040 is configured to select an image post-processing model based on the expanded information through the target agent, and generate a material image using the selected image post-processing model under the guidance of the expanded information;

[0357] The page generation unit 3050 is used to input the page generation requirement information corresponding to each sub-process in the sub-process sequence into the second largest language model through the target intelligent agent, obtain the activity page corresponding to the sub-process, and add the material picture to the activity page, thereby generating the target activity page.

[0358] Optionally, the information expansion unit 3020 is specifically configured to:

[0359] Determine the application to which the target activity belongs through the target agent;

[0360] Get the word definition library of the application;

[0361] Extract description keywords from description information;

[0362] Based on the description keywords, search the word interpretation database to obtain the interpretation information of the description keywords;

[0363] The paraphrase information is added to the description information to obtain the expanded information.

[0364] Optionally, the first large language model includes a first sub-large language model and a second sub-large language model;

[0365] The second input unit 3030 is specifically used for:

[0366] Inputting the expanded information into the first sub-large language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process;

[0367] The expanded information, sub-process timing and page generation requirement information are input into the second sub-large language model to obtain the revised sub-process timing and page generation requirement information.

[0368] Optionally, the picture post-processing model includes a first picture post-processing model based on picture style and a second picture post-processing model based on application;

[0369] The image generation unit 3040 is specifically configured to:

[0370] Through the target agent, the expanded information is input into the third language model to obtain the image style keywords;

[0371] Selecting a first image post-processing model from a plurality of first candidate image post-processing models based on the image style keyword;

[0372] Identify the app to which the target activity belongs;

[0373] Based on the application, a second picture post-processing model is selected from a plurality of second candidate picture post-processing models.

[0374] Optionally, the picture generating unit 3040 is further specifically configured to:

[0375] Input the expanded information into the stable diffusion model to obtain the material image base;

[0376] Using the image post-processing model, post-processing is performed on the material image base to obtain the material image.

[0377] Optionally, the picture generating unit 3040 is further specifically configured to:

[0378] Extracting positive guidance keywords and negative guidance keywords from the expanded information;

[0379] Converting the forward guidance keyword into a first guidance vector and converting the reverse guidance keyword into a second guidance vector;

[0380] Inputting the initial basis vector into the stable diffusion model to obtain the material image basis vector under the guidance of the first guide vector and the second guide vector;

[0381] Converts a material base vector to a material base.

[0382] Optionally, the picture generating unit 3040 is further specifically configured to:

[0383] Initialize the base vector to be diffused as the initial base vector, and initialize the step number to 1;

[0384] Input the base vector to be diffused, the step number, the first guide vector and the second guide vector into the stable diffusion model to obtain the diffusion noise corresponding to the step number;

[0385] The diffusion noise corresponding to the step number is used to offset the base vector to be diffused, and the step number is increased by 1, and the process returns to the step of inputting the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model until the step number increases to a preset maximum number of steps.

[0386] Optionally, the image post-processing model includes a first image post-processing model based on image style and a second image post-processing model based on application, and the post-processing includes style change and application mark superposition;

[0387] The picture generation unit 3040 is further specifically configured to:

[0388] Using the first image post-processing model, the style of the material image base is changed;

[0389] The second image post-processing model is used to superimpose the applied marker on the style-changed material image base to obtain the material image.

[0390] Optionally, the page generation requirement information includes management-side page generation requirement information and client-side page generation requirement information, and the activity page includes management-side activity page and client-side activity page;

[0391] The page generation unit 3050 is specifically used for:

[0392] Inputting the client page generation requirement information corresponding to the sub-process into the second language model to obtain the client activity page corresponding to the sub-process;

[0393] Input the management end page generation requirement information corresponding to the sub-process into the second language model to obtain the management end activity page corresponding to the sub-process.

[0394] Optionally, the page generating unit 3050 is further specifically configured to:

[0395] Get the first domain-specific language rule of the client;

[0396] Inputting the client page generation requirement information corresponding to the sub-process and the first domain-specific language rules into the second language model to obtain the first domain-specific language corresponding to the sub-process;

[0397] The first domain specific language is rendered into a client active page using a domain specific language interpreter.

[0398] Optionally, the page generating unit 3050 is further specifically configured to:

[0399] Obtain the second domain-specific language rules from the management side;

[0400] Inputting the management end page generation requirement information corresponding to the sub-process and the second domain specific language rules into the second language model to obtain the second domain specific language corresponding to the sub-process;

[0401] The second domain specific language is rendered into an active management page using a domain specific language interpreter.

[0402] Optionally, the page generating unit 3050 is further specifically configured to:

[0403] Generate requirement information from the page, obtain the location where the material image is to be added, and the corresponding relationship information between the location where the material image is to be added and the material image;

[0404] In the material picture adding position on the activity page, add the material pictures corresponding to the material picture adding position according to the corresponding relationship information.

[0405] Optionally, after adding the material image to the activity page to generate the target activity page, the activity page generating apparatus further includes:

[0406] A first sending unit (not shown) is used to send the target format information of the response data and key indicators to the application front end through the target agent;

[0407] An acquisition unit (not shown) is used to acquire, through the application front end, response data of the platform object to the target activity page within a predetermined period of time after the target activity page is generated, and to acquire key indicators based on the response data;

[0408] A second sending unit (not shown) is used to send the response data and key indicators to the target agent based on the target format information through the application front end;

[0409] The third input unit (not shown) is used to input the response data and key indicators into the fourth language model through the target intelligent agent to obtain a data report page.

[0410] Optionally, the third input unit (not shown) is specifically used to:

[0411] Through the target agent, the response data and key indicators are input into the fourth language model to obtain the third domain-specific language;

[0412] The domain-specific language interpreter is used to render the third domain-specific language into a data report page.

[0413] Optionally, the first input unit 3010 is specifically configured to:

[0414] Determine the activity type of the target activity;

[0415] Identify the app to which the target activity belongs;

[0416] Determine the target agent from multiple candidate agents based on the application and activity type;

[0417] Input the description information of the target activity into the target agent.

[0418] Optionally, the second input unit 3030 is further specifically configured to:

[0419] Extract sub-process description information and page description information from the expanded information through the target agent;

[0420] Input the sub-process description information into the first language model to obtain the sub-process time sequence;

[0421] The sub-process timing and page description information are input into the first language model to obtain page generation requirement information corresponding to the sub-process.

[0422] Optionally, after inputting the expanded information into the first language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process, the activity page generation apparatus further includes:

[0423] A fourth input unit (not shown) is configured to input, through the target agent, for each sub-process, the sub-process sequence, the page generation requirement information corresponding to the sub-process, the first page generation requirement information corresponding to a predetermined number of preceding sub-processes preceding the sub-process, and the second page generation requirement information corresponding to a predetermined number of subsequent sub-processes following the sub-process into the context continuity determination model to obtain a context continuity determination result for the sub-process;

[0424] An adjusting unit (not shown) is configured to adjust page generation requirement information corresponding to a sub-process based on a result of context continuity determination through a target agent.

[0425] Reference Figure 31 , Figure 31 The following is a block diagram of the structure of a terminal that implements the method for generating an activity page according to an embodiment of the present disclosure. The terminal includes: a radio frequency (RF) circuit 3110, a memory 3115, an input unit 3130, a display unit 3140, a sensor 3150, an audio circuit 3160, a wireless fidelity (WiFi) module 3170, a processor 3180, and a power supply 3190. It will be understood by those skilled in the art that Figure 31 The terminal structure shown does not constitute a limitation on the mobile phone or computer, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0426] The RF circuit 3110 may be used for receiving and sending signals during information transmission or calls. In particular, after receiving downlink information from the base station, it is sent to the processor 3180 for processing; in addition, the designed uplink data is sent to the base station.

[0427] The memory 3115 may be used to store software programs and modules. The processor 3180 executes various functional applications and data processing of the content terminal by running the software programs and modules stored in the memory 3115 .

[0428] The input unit 3130 may be configured to receive input digital or character information and generate key signal input related to the settings and function control of the content terminal. Specifically, the input unit 3130 may include a touch panel 3131 and other input devices 3132 .

[0429] The display unit 3140 may be configured to display input information or provided information and various menus of the content terminal. The display unit 3140 may include a display panel 3141.

[0430] The audio circuit 3160 , the speaker 3161 , and the microphone 3162 may provide an audio interface.

[0431] In this embodiment, the processor 3180 included in the terminal can execute the activity page generation method of the previous embodiment.

[0432] The terminals of the embodiments of the present disclosure include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc. The embodiments of the present invention can be applied to various scenarios, including but not limited to content recommendation, data screening, etc.

[0433] Figure 32 This is a block diagram of the structure of a portion of the server 140 for implementing the method for generating an active page according to an embodiment of the present disclosure. The server 140 may vary significantly due to different configurations or performances, and may include one or more central processing units (CPUs) 3222 (e.g., one or more processors) and memory 3232, and one or more storage media 2630 (e.g., one or more mass storage devices) for storing application programs 3242 or data 3244. The memory 3232 and storage medium 3230 may be temporary storage or permanent storage. The program stored in the storage medium 3230 may include one or more modules (not shown), each module of which may include a series of instruction operations on the server 140. Furthermore, the central processing unit 3222 may be configured to communicate with the storage medium 3230 to execute a series of instruction operations in the storage medium 3230 on the server 140.

[0434] The server 140 may also include one or more power supplies 3226, one or more wired or wireless network interfaces 3250, one or more input and output interfaces 3258, and / or one or more operating systems 3241, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0435] The central processing unit 3222 in the server 140 may be configured to execute the activity page generating method according to the embodiment of the present disclosure.

[0436] The embodiments of the present disclosure further provide a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the activity page generation methods of the aforementioned embodiments.

[0437] The present disclosure also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device performs the above-mentioned delivery process.

[0438] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present disclosure and the accompanying drawings are used to distinguish between similar contents and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein, for example, can be implemented in orders other than those illustrated or described herein. In addition, the terms "comprises" and "comprising," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0439] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated content, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and following associated content is in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0440] It should be understood that in the description of the embodiments of the present disclosure, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to exclude the number itself, and above, below, within, etc. are understood to include the number itself.

[0441] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0442] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0443] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0444] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server 140, or network device, etc.) to execute all or part of the steps of the various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0445] It should also be understood that the various implementations provided in the embodiments of the present disclosure can be combined arbitrarily to achieve different technical effects.

[0446] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A method for generating an activity page, characterized in that: include: Inputting the description information of the target activity into the target agent; Expanding the description information through the target agent to obtain expanded information; Inputting the expanded information into a first large language model to obtain a sub-process timing of the target activity and page generation requirement information corresponding to the sub-process; Selecting, by the target agent, a picture post-processing model based on the expanded information, and using the selected picture post-processing model to generate a material picture under the guidance of the expanded information; Through the target intelligent agent, for each sub-process in the sub-process sequence, the page generation requirement information corresponding to the sub-process is input into the second largest language model to obtain the activity page corresponding to the sub-process, and the material picture is added to the activity page, thereby generating a target activity page.

2. The method for generating an activity page according to claim 1, wherein: The step of expanding the description information by the target agent to obtain expanded information includes: Determining, through the target agent, the application to which the target activity belongs; Obtaining a word definition library for the application; extracting description keywords from the description information; Based on the description keywords, searching the word interpretation library to obtain interpretation information of the description keywords; The interpretation information is added to the description information to obtain the expanded information.

3. The method for generating an activity page according to claim 1, wherein: The first large language model includes a first sub-large language model and a second sub-large language model; Inputting the expanded information into the first large language model to obtain the sub-process timing of the target activity and page generation requirement information corresponding to the sub-process includes: Inputting the expanded information into the first sub-large language model to obtain the sub-process timing of the target activity and the page generation requirement information corresponding to the sub-process; The expanded information, the sub-process timing and the page generation requirement information are input into the second sub-large language model to obtain the revised sub-process timing and the page generation requirement information.

4. The method for generating an activity page according to claim 1, wherein: The picture post-processing model includes a first picture post-processing model based on picture style and a second picture post-processing model based on application; The selecting, by the target agent, a picture post-processing model based on the expanded information includes: Inputting the expanded information into a third language model through the target agent to obtain picture style keywords; Based on the picture style keyword, selecting the first picture post-processing model from a plurality of first candidate picture post-processing models; Determining the application to which the target activity belongs; Based on the application, the second picture post-processing model is selected from a plurality of second candidate picture post-processing models.

5. The method for generating an activity page according to claim 1, wherein: The step of generating a material image using the selected image post-processing model under the guidance of the expanded information includes: Inputting the expanded information into a stable diffusion model to obtain a material image base; The image post-processing model is used to perform post-processing on the material image base to obtain the material image.

6. The method for generating an activity page according to claim 5, wherein: Inputting the expanded information into a stable diffusion model to obtain a material image base includes: Extracting forward guidance keywords and reverse guidance keywords from the expanded information; Converting the forward guidance keyword into a first guidance vector and converting the reverse guidance keyword into a second guidance vector; Inputting the initial basis vector into the stable diffusion model to obtain the material picture basis vector under the guidance of the first guide vector and the second guide vector; The material image base vector is converted into the material image base.

7. The method for generating an activity page according to claim 6, wherein: Inputting the initial basis vector into the stable diffusion model to obtain the material picture basis vector under the guidance of the first guide vector and the second guide vector includes: Initialize the to-be-diffused basis vector to the initial basis vector, and initialize the step number to 1; Inputting the to-be-diffused base vector, the step number, the first guide vector, and the second guide vector into the stable diffusion model to obtain diffusion noise corresponding to the step number; The base vector to be diffused is offset by the diffusion noise corresponding to the step number, the step number is increased by 1, and the process returns to the step of inputting the base vector to be diffused, the step number, the first guide vector, and the second guide vector into the stable diffusion model until the step number increases to a preset maximum number of steps.

8. The method for generating an activity page according to claim 5, wherein: The image post-processing model includes a first image post-processing model based on image style and a second image post-processing model based on application, wherein the post-processing includes style change and application mark superposition; The post-processing is performed on the material image base by using the image post-processing model to obtain the material image, including: Using the first image post-processing model, changing the style of the material image base; The second image post-processing model is used to superimpose the application mark on the material image base after the style is changed to obtain the material image.

9. The method for generating an activity page according to claim 1, wherein: The page generation requirement information includes management-side page generation requirement information and client-side page generation requirement information, and the activity page includes management-side activity page and client-side activity page; The step of inputting the page generation requirement information corresponding to the sub-process into a second language model to obtain an active page corresponding to the sub-process includes: Inputting the client page generation requirement information corresponding to the sub-process into a second language model to obtain the client activity page corresponding to the sub-process; The management end page generation requirement information corresponding to the sub-process is input into the second language model to obtain the management end activity page corresponding to the sub-process.

10. The method for generating an activity page according to claim 9, wherein: The step of inputting the client page generation requirement information corresponding to the sub-process into a second language model to obtain the client activity page corresponding to the sub-process includes: Obtaining a first domain-specific language rule of the client; Inputting the client page generation requirement information corresponding to the sub-process and the first domain-specific language rules into the second language model to obtain the first domain-specific language corresponding to the sub-process; The first domain-specific language is rendered into the client active page using a domain-specific language interpreter.

11. The activity page generation method according to claim 9, characterized in that: The step of inputting the management terminal page generation requirement information corresponding to the sub-process into a second language model to obtain the management terminal activity page corresponding to the sub-process includes: Acquire a second domain specific language rule of the management terminal; Inputting the management-end page generation requirement information corresponding to the sub-process and the second domain-specific language rules into the second language model to obtain the second domain-specific language corresponding to the sub-process; The second domain-specific language is rendered into the management-end active page using a domain-specific language interpreter.

12. The method for generating an activity page according to claim 1, wherein: Adding the material image to the activity page to generate a target activity page includes: Generating requirement information from the page, obtaining a material picture adding position, and information on a correspondence between the material picture adding position and the material picture; At the material picture adding position of the activity page, the material picture corresponding to the material picture adding position is added according to the corresponding relationship information.

13. The activity page generation method according to claim 1, characterized in that: After adding the material picture to the activity page to generate the target activity page, the activity page generation method further includes: Sending the target format information of the response data and key indicators to the application front end through the target agent; Obtaining, through the application front end, the response data of the platform object to the target activity page within a predetermined time period after the target activity page is generated, and obtaining the key indicator based on the response data; Sending the response data and the key indicators to the target agent based on the target format information through the application front end; The response data and the key indicators are input into the fourth language model through the target agent to obtain a data report page.

14. The activity page generation method according to claim 13, characterized in that: The target agent inputs the response data and the key indicators into the fourth language model to obtain a data report page, including: Inputting the response data and the key indicators into a fourth language model through the target agent to obtain a third domain-specific language; The third domain specific language is rendered into the data report page using a domain specific language interpreter.

15. An activity page generating device, characterized in that: include: A first input unit, configured to input description information of a target activity into a target agent; An information expansion unit, configured to expand the description information through the target agent to obtain expanded information; a second input unit, configured to input the expanded information into the first large language model to obtain a sub-process sequence of the target activity and page generation requirement information corresponding to the sub-process; An image generation unit, configured to select, through the target agent, an image post-processing model based on the expanded information, and generate a material image using the selected image post-processing model under the guidance of the expanded information; The page generation unit is used to input the page generation requirement information corresponding to each sub-process in the sub-process sequence into the second language model through the target intelligent agent, obtain the activity page corresponding to the sub-process, and add the material picture to the activity page, thereby generating a target activity page.

16. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the activity page generating method according to any one of claims 1 to 14 is implemented.

17. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the activity page generating method according to any one of claims 1 to 14 is implemented.

18. A computer program product, comprising a computer program, wherein the computer program is read and executed by a processor of a computer device, so that the computer device executes the method for generating an active page according to any one of claims 1 to 14.

Images