Modular website integrated generation system and method based on multi-modal AI and closed-loop optimization

Abstract

The invention discloses a modularized website integrated generation system and method based on multi-modal AI and closed-loop optimization, particularly relates to the technical field of website integrated generation, can accurately grasp user behavior characteristics by collecting and processing multi-modal data in real time, and improves the user experience. Personalized content is automatically generated and a website structure is dynamically adjusted according to user behavior characteristics, functional components are intelligently selected through a deep learning model, and a response type layout is constructed, so that high matching of website content and user requirements is realized, and the user experience is improved in a high-concurrency scene. A load balancing and distributed data processing framework is utilized to efficiently distribute requests and process large-scale data, system response delay is effectively reduced, the stability of feedback circulation is ensured through cache optimization and continuous monitoring of system loads and feedback delay, the situation of too high system loads or feedback lagging is recognized and dealt with in time, and the service life of the system is prolonged. If closed-loop optimization is unstable, a fault-tolerant mechanism and a data cleaning strategy are automatically triggered, and continuous and efficient operation of the system is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of website integrated generation, more particularly, the present application relates to a modular website integrated generation system and method based on multi-modal AI and closed-loop optimization. BACKGROUND

[0002] With the development of Internet technology, the modular website integrated generation method based on multi-modal AI and closed-loop optimization gradually becomes the key to improving user experience and system adaptive ability. However, in a high-concurrency environment, the system faces two main challenges: high-concurrency data processing and response delay, and unstable feedback loop in the closed-loop optimization process. These two challenges interact with each other, directly affecting the stability of the system and user experience.

[0003] In a high-concurrency scenario, the system needs to handle a large amount of multi-modal data from users at the same time, including text, images, videos, etc. The transmission and processing of these data consume a lot of computing resources, resulting in prolonged response time and limited processing capacity of the system. At the same time, the closed-loop optimization system relies on real-time user behavior data for feedback and dynamic adjustment. When the system is affected by high-concurrency load, the collection and transmission of user behavior data are delayed or lost, resulting in distorted feedback links and the optimization process unable to timely reflect the actual needs of users. This data feedback lag makes the website content, layout and recommendations unable to quickly adapt to changes in users, resulting in content and layout update lag.

[0004] In addition, when the system faces high concurrency, it may waste traffic due to improper resource scheduling. In this case, the wrong feedback in the closed-loop optimization may guide the system to prioritize processing low-priority or irrelevant requests, further increasing unnecessary burden and traffic consumption. This interaction forms a vicious cycle, which cannot optimize user experience and cannot effectively utilize system resources.

[0005] Therefore, the mutual influence between high concurrency and unstable feedback loop ultimately leads to inconsistent user experience and optimization lag, seriously affecting the responsiveness and accuracy of the system, and thus affecting the continuous participation and satisfaction of users. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a modular website integrated generation system and method based on multi-modal AI and closed-loop optimization to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The modular website integrated generation method based on multi-modal AI and closed-loop optimization comprises the following steps: Real-time acquisition of users' text, voice, and image multimodal data, followed by denoising, temporal alignment, and synchronization processing; extraction of user behavior features; and generation of a unified data stream. The data stream is input into the content generation engine, which selects functional components and builds the website structure based on a deep learning model, and dynamically adjusts the page layout using responsive design. The content generation engine generates personalized content based on user behavior characteristics and delivers the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. User behavior feedback data is transmitted to the optimization engine, which adjusts content, recommendations, and layout based on real-time feedback. Under high-concurrency feedback adjustment, requests are distributed through load balancing, large-scale data is processed using a distributed data processing framework, and response latency is reduced through caching optimization. Continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability during closed-loop optimization, and determine whether fault tolerance mechanisms and data cleaning strategies are triggered.

[0008] In a preferred embodiment, the denoising includes text denoising, speech denoising, and image denoising; The time alignment includes assigning a uniform timestamp to each modality of data to ensure consistency of data from different modalities on the timeline; and aligning multimodal data using a time alignment algorithm. The extraction of user behavior features includes text feature extraction, voice feature extraction, and image feature extraction.

[0009] In a preferred embodiment, inputting the data stream into the content generation engine includes inputting the generated multimodal feature stream into the content generation engine; the content generation engine parses the input multimodal feature stream, extracts key information, and forms a structured data representation; The process of selecting functional components and constructing the website structure based on a deep learning model includes using a deep learning model to analyze the input multimodal feature stream and predict user needs and preferences; based on the prediction results, the content generation engine selects functional components and content display formats; based on the selected functional components, the content generation engine automatically generates the initial structural layout of the website; and binding the selected functional components with user data to ensure that the website content is updated in real time according to user characteristics. The method of dynamically adjusting the page layout using responsive design includes automatically identifying device information based on the user's device type and selecting a responsive design strategy; and automatically adjusting the page layout based on the responsive design framework.

[0010] In a preferred embodiment, the content generation engine generates personalized content based on user behavior characteristics, including receiving user behavior characteristic data from the multimodal data processing stage to form a personalized behavior profile of the user; the content generation engine uses a natural language generation model to analyze user behavior characteristics and generate personalized content that meets the user's interests and needs; the generated personalized content is bound to the user's behavior characteristics and needs. The process of delivering personalized content to the recommendation system and pushing relevant content and functions includes delivering the generated personalized content to the recommendation system as input data for the recommendation engine. The recommendation system uses collaborative filtering and content-based deep learning recommendation models to analyze the correlation between the content and other user behaviors, and generates a recommendation list of relevant content. Based on the recommendation list generated by the personalized content, the recommendation system pushes content and functions related to user needs. The monitoring and collection of user behavior feedback data includes real-time monitoring of user behavior after the user interacts with the recommended content, and collecting the user behavior feedback data into a behavior database.

[0011] In a preferred embodiment, the process of distributing requests through load balancing under high-concurrency feedback adjustment, and using a distributed data processing framework to process large-scale data, includes triggering a high-concurrency state warning when the number of user requests exceeds a preset request threshold; using a load balancer to intelligently distribute high-concurrency requests, with the load balancer dynamically selecting the most suitable server node for processing based on the request source, request type, and current server load; and transmitting the user requests distributed to each server node to the distributed data processing framework. The reduction of response latency through caching optimization includes accelerating access through caching mechanisms, storing hot data, recommendation results, and user historical behavior data in the cache.

[0012] In a preferred embodiment, the high-concurrency data processing load includes server CPU utilization, memory utilization, and I / O latency. The feedback loop delay includes network latency, data transmission latency, and processing latency.

[0013] In a preferred embodiment, for each monitoring period, a weighted sum of the high-concurrency data processing load and feedback loop latency is calculated and defined as the total load latency index, as follows: ,in The total load delay index, For high-concurrency data processing loads, For feedback loop delay, These are preset proportional coefficients for high-concurrency data processing load and feedback loop latency, respectively. All are greater than 0.

[0014] In a preferred embodiment, the stability of closed-loop optimization is analyzed based on the total load delay index, specifically as follows: if the total load delay index exceeds a preset total load delay index threshold, the closed-loop optimization of the system is considered to be in an unstable state. If the closed-loop optimization is in an unstable state, the fault tolerance mechanism and data cleaning strategy will be triggered.

[0015] In a preferred embodiment, the modular website integrated generation system based on multimodal AI and closed-loop optimization includes a data acquisition and preprocessing module, a content generation and structure building module, a personalized content generation and recommendation module, a feedback and optimization adjustment module, a high-concurrency request processing and optimization module, and a performance monitoring and fault tolerance module. The data acquisition and preprocessing module is used to collect users' text, voice, and image multimodal data in real time and perform noise reduction, time alignment and synchronization processing, extract user behavior features, and generate a unified data stream; The content generation and structure building module is used to input data streams into the content generation engine, select functional components based on deep learning models and build the website structure, and dynamically adjust the page layout using responsive design. The personalized content generation and recommendation module is used by the content generation engine to generate personalized content based on user behavior characteristics, and then pass the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. The feedback and optimization module is used to transmit user behavior feedback data to the optimization engine and adjust the content, recommendations and layout based on real-time feedback; The high-concurrency request processing and optimization module is used to distribute requests through load balancing under high-concurrency feedback adjustment, process large-scale data using a distributed data processing framework, and reduce response latency through caching optimization. The performance monitoring and fault tolerance module is used to continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability in closed-loop optimization, and determine whether to trigger fault tolerance mechanisms and data cleaning strategies.

[0016] The technical effects and advantages of this invention are as follows: 1. This invention is based on a modular website integrated generation method using multimodal AI and closed-loop optimization. By collecting and processing multimodal data in real time, it can accurately grasp user behavior characteristics and automatically generate personalized content and dynamically adjust the website structure according to these characteristics. Through deep learning models, it intelligently selects functional components and builds a responsive layout, achieving a high degree of matching between website content and user needs. In high-concurrency scenarios, it utilizes load balancing and distributed data processing frameworks to efficiently allocate requests and process large-scale data, effectively reducing system response latency. Through caching optimization and continuous monitoring of system load and feedback latency, it ensures the stability of the feedback loop, promptly identifies and responds to situations of excessive system load or delayed feedback. If the closed-loop optimization is found to be unstable, it automatically triggers fault tolerance mechanisms and data cleaning strategies to ensure the continuous and efficient operation of the system.

[0017] 2. This invention significantly improves the responsiveness and accuracy of the system, effectively solving the problems of data processing latency and closed-loop optimization distortion under high concurrency. The user experience is no longer affected by excessive system load. Website content, layout, and recommendations can quickly adapt to user needs based on real-time feedback, avoiding delays in content and function updates. Through adaptive optimization, it can dynamically adjust according to users' real-time behavior and needs, improving user satisfaction and engagement. At the same time, resource utilization is optimized, avoiding traffic waste and ensuring stability and efficiency under high concurrency. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the system in Embodiment 2 of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Figure 1 The present invention provides a modular website integrated generation method based on multimodal AI and closed-loop optimization, comprising the following steps: Real-time acquisition of users' text, voice, and image multimodal data, followed by denoising, temporal alignment, and synchronization processing; extraction of user behavior features; and generation of a unified data stream. The data stream is input into the content generation engine, which selects functional components and builds the website structure based on a deep learning model, and dynamically adjusts the page layout using responsive design. The content generation engine generates personalized content based on user behavior characteristics and delivers the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. User behavior feedback data is transmitted to the optimization engine, which adjusts content, recommendations, and layout based on real-time feedback. Under high-concurrency feedback adjustment, requests are distributed through load balancing, large-scale data is processed using a distributed data processing framework, and response latency is reduced through caching optimization. Continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability during closed-loop optimization, and determine whether fault tolerance mechanisms and data cleaning strategies are triggered.

[0021] Real-time acquisition of users' text, voice, and image multimodal data, followed by denoising, temporal alignment, and synchronization processing; extraction of user behavior features; and generation of a unified data stream. In this embodiment of the invention, collecting user text data includes capturing user-inputted text information in real time through a natural language processing interface (such as an API or an embedded speech recognition module) or obtaining user voice input through speech-to-text technology. The collection of user voice data includes using audio capture devices (such as microphones) or audio acquisition systems to record user voice information in real time, and converting it into text or extracting voice features through speech recognition models (such as DeepSpeech, Google Speech-to-Text). The collection of user image data includes real-time acquisition of user image information or video streams through a camera or image acquisition device. The images can be static pictures or dynamic video frames. The denoising includes text denoising, speech denoising, and image denoising; Text denoising: Cleaning and denoising text data to remove irrelevant characters, stop words, and spelling errors, ensuring the accuracy and clarity of the text data; Speech denoising: Remove background noise using speech enhancement algorithms (such as WavNet and SpecAugment), extract effective speech signals, and convert the speech signals into clear text using acoustic models (such as DeepSpeech); Image denoising: removing noise from image or video data by applying image processing algorithms (such as Gaussian filtering and median filtering) to remove blur and noise, ensuring that the image quality is suitable for subsequent processing; The temporal alignment includes assigning a uniform timestamp to each modality of data (text, speech, and image) to ensure consistency of data across different modalities on the timeline; aligning multimodal data using time alignment algorithms (such as Dynamic Time Warping (DTW)) to ensure accurate correspondence between speech, text, and image data within the same time period; and synchronizing the timestamps of dynamic video streams and speech data to ensure spatiotemporal consistency of multimodal data.

[0022] The extraction of user behavior features includes text feature extraction, voice feature extraction, and image feature extraction; Text feature extraction: Use pre-trained natural language processing models (such as BERT, GPT) to extract features from text data, obtain word vectors or sentence vectors, and extract features such as user sentiment and intent; Speech feature extraction: Utilize speech processing models (such as MFCC, VGGish, DeepSpeech) to extract features from speech data, such as audio spectrum features, intonation, and speech rate; Image feature extraction: Extracting spatial features from images or video streams using computer vision models such as Convolutional Neural Networks (CNN) (e.g., ResNet, VGG) to identify information such as objects, expressions, and poses; By fusing features from text, speech, and image data to form a multimodal feature vector, concatenation, weighted fusion, or deep fusion models (such as the multimodal Transformer) can be used to merge the features of each modality. Finally, the fused multimodal feature stream is output in a unified data format (such as JSON or Protobuf) to generate a unified data stream.

[0023] The data stream is input into the content generation engine, which selects functional components and builds the website structure based on a deep learning model, and dynamically adjusts the page layout using responsive design. In this embodiment of the invention, inputting the data stream into the content generation engine includes inputting the generated multimodal feature stream (including fused features of text, speech, images, etc.) into the content generation engine. This data stream contains user behavioral characteristics, emotional states, preferences, etc. The content generation engine parses the input multimodal feature stream, extracts key information (such as user needs, sentiment analysis results, behavior predictions, etc.), forms a structured data representation, and prepares for functional module selection and page structure construction. The process of selecting functional components and constructing the website structure based on deep learning models involves using deep learning models (such as neural networks and recommendation system models) to analyze the multimodal feature stream of the input and predict user needs and preferences. For example, based on the user's historical behavior, preferences, and current emotional state, it predicts the functional modules the user might need, such as recommended content, interactive modules, and navigation elements. Based on the prediction results, the content generation engine selects the most suitable functional components (such as text display areas, recommendation engines, video players, and comment sections) and content display formats to ensure that the website structure is consistent with user needs. The deep learning model can optimize the module selection strategy through training data (such as A / B test results and user feedback). Based on the selected functional components, the content generation engine automatically generates the initial structural layout of the website. This layout includes the arrangement order, size, and interaction methods of each functional area. Algorithms (such as rule-based layout engines and optimization algorithms) are used to ensure that the structure is reasonable, concise, and conforms to user interaction habits. The selected functional components are bound to user data (such as personalized content and recommendation algorithm results) to ensure that the website content is updated in real time according to user characteristics. For example, user interest tags are bound to the recommended content module to dynamically load and display relevant content. The dynamic adjustment of page layout using responsive design includes automatically identifying device information based on the user's device type (such as mobile phone, tablet, PC, etc.) and selecting an appropriate responsive design strategy. The system can determine this based on the device's screen size, resolution, browser characteristics, etc.; and automatically adjust the page layout based on responsive design frameworks (such as CSS Grid, Flexbox). Through fluid layout and media query technology, page elements (such as navigation bars, images, and recommendation areas) will be dynamically adjusted according to different devices, ensuring the website provides the best user experience regardless of the device. Under different devices, the arrangement order, size, or function of certain content modules may need to be adjusted. For example, on mobile phones, a multi-column layout may be changed to a single-column layout, and image sizes and button layouts may be adjusted. The loading order of page content can also be optimized based on network conditions and device performance to ensure loading speed and smoothness. Based on the generated page layout and design, the content generation engine outputs the corresponding HTML, CSS, and JavaScript code to form a complete responsive website page.

[0024] The content generation engine generates personalized content based on user behavior characteristics and delivers the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. In this embodiment of the invention, the content generation engine generates personalized content based on user behavior characteristics. This includes receiving user behavior characteristic data from the multimodal data processing stage. This data stream includes the user's historical behavior data (such as clicks, browsing, and purchase records), sentiment analysis results, interest tags, etc., forming a personalized user behavior profile. The content generation engine uses a Natural Language Generation (NLG) model to analyze user behavior characteristics and generate personalized content that matches the user's interests and needs. This content can be in the form of text, images, videos, or other multimedia. The content generation process is customized based on factors such as user preferences, emotional state, and behavioral predictions. The generated personalized content is bound to the user's behavior characteristics and needs. For example, the system may generate related articles, product recommendations, or interactive content based on the user's recently browsed product types or search history. This content, combined with the user profile, ensures that the displayed content maximizes its relevance to the user's current needs. The process of delivering personalized content to the recommendation system for pushing relevant content and functions includes: sending the generated personalized content to the recommendation system as input data for the recommendation engine; the recommendation system using collaborative filtering and content-based deep learning recommendation models (such as neural collaborative filtering, matrix factorization, etc.) to analyze the correlation between the content and other user behaviors, generating a recommendation list of relevant content; and the recommendation system pushing content and functions related to user needs based on the recommendation list generated by the personalized content. The pushed content may include articles, products, videos, advertisements, or interactive modules, and the recommended content will be updated in real time to ensure its timeliness and relevance. The monitoring and collection of user behavior feedback data includes real-time monitoring of user behavior (such as clicks, dwell time, feedback, purchases, etc.) after users interact with recommended content, and collecting user behavior feedback data (click-through rate, dwell time, scroll depth, page bounce rate, number of comments, purchase behavior, etc.) into the behavior database. User interaction data is recorded in real time and transmitted to the feedback monitoring system to provide a basis for subsequent optimization and content adjustment.

[0025] In a further embodiment of the present invention, an SEO adaptive engine is embedded in the content generation engine to dynamically optimize the website's search engine visibility. The SEO adaptive engine constructs a search engine algorithm change monitoring network, which automatically analyzes keyword weight changes and link quality assessment standards by periodically acquiring official algorithm update announcements or index weight change information from mainstream search engines such as Baidu and Google. Based on a Long Short-Term Memory (LSTM) network model, time-series prediction of SEO weight change trends is performed to obtain keyword popularity evolution curves and page weight fluctuation ranges.

[0026] Based on the prediction results, the SEO adaptive engine automatically optimizes and adjusts the meta tags (including title, description, and keywords) of website pages; it also restructures the internal link structure, prioritizes indexing and optimizes anchor points for high-authority pages; and adaptively adjusts the density of the page's main text content and the distribution ratio of keywords based on the content semantic density model. This process forms a closed-loop optimization path: prediction-adjustment-monitoring-re-prediction, ensuring that the website's content maintains a stable search engine indexing weight after algorithm updates, thereby improving the external search accessibility and dynamic adaptability of the automatically generated website.

[0027] User behavior feedback data is transmitted to the optimization engine, which adjusts content, recommendations, and layout based on real-time feedback. In this embodiment of the invention, user behavior feedback data is transmitted to the optimization engine. Adjustments to content, recommendations, and layout based on real-time feedback include: after receiving the feedback data, the optimization engine performs data cleaning and preprocessing to remove noise and ensure high data quality and accuracy; the optimization engine further analyzes the feedback data to identify changes in user interests, shifts in preferences, or dissatisfaction; the optimization engine uses analytical models (such as deep learning, cluster analysis, etc.) to model and analyze the user feedback data to identify which content is liked by users and which content fails to meet their needs. Based on the feedback analysis results, the optimization engine adjusts its content generation strategy, for example, by replacing recommended content, adjusting the display order of content, or recommending new types of content to users; the optimization engine optimizes the website's page layout based on the user behavior feedback data and the adjustment results of the recommendation system, adjusting the display method of page elements by analyzing user interaction paths (such as dwell time, clicked areas, page browsing order, etc.), which may include: The position and size of the recommendation area are dynamically adjusted.

[0028] Optimize the layout of the navigation bar or buttons to improve usability.

[0029] Adjust the display strategy for advertising and interactive modules to improve user engagement.

[0030] Based on the optimization engine's adjustments, the system uses front-end update mechanisms (such as AJAX requests and WebSockets) to update new content, recommendations, and layout configurations to the user's browsing pages in real time. These adjustments directly impact user experience, ensuring a more personalized page presentation through seamless dynamic updates.

[0031] Under high-concurrency feedback adjustment, requests are distributed through load balancing, large-scale data is processed using a distributed data processing framework, and response latency is reduced through caching optimization. In this embodiment of the invention, requests are distributed through load balancing under high-concurrency feedback adjustment conditions. A distributed data processing framework is used to process large-scale data. This includes triggering a high-concurrency alert when the number of user requests exceeds a preset request threshold. User requests may involve various operations such as content display, recommendation queries, and data submission. Each request contains user behavior data, request type, device information, etc. A load balancer (such as Nginx, HAProxy, AWS ELB) is used to intelligently distribute high-concurrency requests. The load balancer dynamically selects the most suitable server node for processing based on the request source, request type, and current server load, ensuring even distribution of requests across multiple servers and avoiding single-point overload. User requests distributed to each server node are then passed to a distributed data processing framework (such as Apache Kafka, Flink, Spark). The system utilizes this framework to perform parallel processing and real-time streaming processing of large amounts of data, ensuring efficient handling of the data volume and computational demands brought by concurrent requests. The reduction of response latency through caching optimization includes accelerating access through caching mechanisms (such as Redis and Memcached), storing hot data, recommendation results, and user historical behavior data in the cache, and reading directly from the cache in subsequent requests instead of recalculating or querying the database every time, thereby significantly reducing response time. At the same time, it combines a content delivery network (CDN) to accelerate static resources (such as images, CSS, and JS files). When a user makes a request, CDN nodes will select the optimal node to provide resources based on the user's geographical location, reducing network latency and improving page loading speed.

[0032] Continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability during closed-loop optimization, and determine whether fault tolerance mechanisms and data cleaning strategies are triggered.

[0033] In this embodiment of the invention, the high-concurrency data processing load includes server CPU utilization, memory utilization, I / O wait time, etc., reflecting the data processing load. The feedback loop delay includes network delay, data transmission delay, and processing delay, etc. For each monitoring period, a weighted sum of the high-concurrency data processing load and feedback loop latency is calculated based on the high-concurrency data processing load and feedback loop latency, and defined as the total load latency index. The specific calculation formula is as follows: ,in The total load delay index, For high-concurrency data processing loads, the data is represented by a weighted average of server CPU utilization, memory utilization, and I / O latency. The feedback loop delay is represented by a weighted average of network latency, data transmission latency, and processing latency. These are preset proportional coefficients for high-concurrency data processing load and feedback loop latency, respectively. All are greater than 0; It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. The initial value can be 0.5, 0.5; The stability of closed-loop optimization is analyzed based on the total load latency index, specifically as follows: if the total load latency index exceeds a preset threshold, the closed-loop optimization of the system is considered to be in an unstable state. The total load latency index threshold is determined through historical testing and performance evaluation, and is used to judge whether the system can operate stably under the current load.

[0034] If the closed-loop optimization becomes unstable, a fault tolerance mechanism and data cleaning strategy are triggered. The implementation of the fault tolerance mechanism typically includes: Dynamically adjust server resources, such as adding compute nodes or optimizing data allocation.

[0035] Transfer some high-priority requests to a standby server or cold backup system.

[0036] Degrade service quality, such as reducing the load on certain non-critical functions to ensure the stability of core functions.

[0037] Data cleaning strategies typically include: Filter out abnormal data, such as deleting data points that do not match actual behavior or are obviously wrong.

[0038] Fill in missing values ​​by using interpolation or inferring missing data based on similar user behavior.

[0039] Anomaly detection uses machine learning algorithms (such as Isolation Forest, K-means clustering, etc.) to detect and remove data with abnormal behavior.

[0040] This invention presents a modular website generation method based on multimodal AI and closed-loop optimization. By collecting and processing multimodal data in real time, it can accurately grasp user behavior characteristics and automatically generate personalized content and dynamically adjust the website structure based on these characteristics. Through deep learning models, it intelligently selects functional components and constructs a responsive layout, achieving a high degree of matching between website content and user needs. In high-concurrency scenarios, it utilizes load balancing and distributed data processing frameworks to efficiently allocate requests and process large-scale data, effectively reducing system response latency. Through caching optimization and continuous monitoring of system load and feedback latency, it ensures the stability of the feedback loop, promptly identifies and addresses situations of excessive system load or delayed feedback. If the closed-loop optimization is found to be unstable, it automatically triggers fault tolerance mechanisms and data cleaning strategies to ensure the continuous and efficient operation of the system.

[0041] This invention significantly improves the responsiveness and accuracy of the system, effectively solving the problems of data processing latency and closed-loop optimization distortion under high concurrency. The user experience is no longer affected by excessive system load. Website content, layout, and recommendations can quickly adapt to user needs based on real-time feedback, avoiding delays in content and function updates. Through adaptive optimization, it can dynamically adjust according to users' real-time behavior and needs, improving user satisfaction and engagement. At the same time, resource utilization is optimized, avoiding traffic waste and ensuring stability and efficiency under high concurrency.

[0042] Example 2: This example introduces a modular website integrated generation system based on multimodal AI and closed-loop optimization, such as... Figure 2 As shown, it includes a data acquisition and preprocessing module, a content generation and structure building module, a personalized content generation and recommendation module, a feedback and optimization adjustment module, a high-concurrency request processing and optimization module, and a performance monitoring and fault tolerance module. The data acquisition and preprocessing module is used to collect users' text, voice, and image multimodal data in real time and perform noise reduction, time alignment and synchronization processing, extract user behavior features, and generate a unified data stream; The content generation and structure building module is used to input data streams into the content generation engine, select functional components based on deep learning models and build the website structure, and dynamically adjust the page layout using responsive design. The personalized content generation and recommendation module is used by the content generation engine to generate personalized content based on user behavior characteristics, and then pass the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. The feedback and optimization module is used to transmit user behavior feedback data to the optimization engine and adjust the content, recommendations and layout based on real-time feedback; The high-concurrency request processing and optimization module is used to distribute requests through load balancing under high-concurrency feedback adjustment, process large-scale data using a distributed data processing framework, and reduce response latency through caching optimization. The performance monitoring and fault tolerance module is used to continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability in closed-loop optimization, and determine whether to trigger fault tolerance mechanisms and data cleaning strategies.

[0043] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0044] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0045] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0046] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and method described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0047] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways.

[0048] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A modular website generation method based on multimodal AI and closed-loop optimization, characterized by: Includes the following steps: Real-time acquisition of users' text, voice, and image multimodal data, followed by denoising, temporal alignment, and synchronization processing; extraction of user behavior features; and generation of a unified data stream. The data stream is input into the content generation engine, which selects functional components and builds the website structure based on a deep learning model, and dynamically adjusts the page layout using responsive design. The content generation engine generates personalized content based on user behavior characteristics and delivers the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. User behavior feedback data is transmitted to the optimization engine, which adjusts content, recommendations, and layout based on real-time feedback. Under high-concurrency feedback adjustment, requests are distributed through load balancing, large-scale data is processed using a distributed data processing framework, and response latency is reduced through caching optimization. Continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability during closed-loop optimization, and determine whether fault tolerance mechanisms and data cleaning strategies are triggered.

2. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 1, characterized in that: The denoising includes text denoising, speech denoising, and image denoising; The time alignment includes assigning a uniform timestamp to each modality of data to ensure consistency of data from different modalities on the timeline; and aligning multimodal data using a time alignment algorithm. The extraction of user behavior features includes text feature extraction, voice feature extraction, and image feature extraction.

3. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 1, characterized in that: The process of inputting the data stream into the content generation engine includes inputting the generated multimodal feature stream into the content generation engine; the content generation engine parses the input multimodal feature stream, extracts key information, and forms a structured data representation; The process of selecting functional components and constructing the website structure based on a deep learning model includes using a deep learning model to analyze the input multimodal feature stream and predict user needs and preferences; based on the prediction results, the content generation engine selects functional components and content display formats; based on the selected functional components, the content generation engine automatically generates the initial structural layout of the website; and binding the selected functional components with user data to ensure that the website content is updated in real time according to user characteristics. The method of dynamically adjusting the page layout using responsive design includes automatically identifying device information based on the user's device type and selecting a responsive design strategy; and automatically adjusting the page layout based on the responsive design framework.

4. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 1, characterized in that: The content generation engine generates personalized content based on user behavior characteristics, including receiving user behavior characteristic data from the multimodal data processing stage and forming a personalized user behavior profile. The content generation engine uses natural language generation models to analyze user behavior characteristics and generate personalized content that matches user interests and needs; the generated personalized content is then bound to user behavior characteristics and needs. The process of delivering personalized content to the recommendation system to push relevant content and functions includes delivering the generated personalized content to the recommendation system as input data for the recommendation engine. The recommendation system uses collaborative filtering and content-based deep learning recommendation models to analyze the correlation between the content and other user behaviors, and generates a recommendation list of relevant content. The recommendation system generates a recommendation list based on personalized content, pushing content and functions related to user needs; The monitoring and collection of user behavior feedback data includes real-time monitoring of user behavior after the user interacts with the recommended content, and collecting the user behavior feedback data into a behavior database.

5. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 1, characterized in that: The process involves load balancing to distribute requests under high-concurrency feedback adjustment conditions and using a distributed data processing framework to process large-scale data, including triggering a high-concurrency state warning when the number of user requests exceeds a preset request threshold. A load balancer is used to intelligently distribute high-concurrency requests. The load balancer dynamically selects the most suitable server node for processing based on the source of the request, the type of request, and the current load of the server. The user requests allocated to each server node are passed to the distributed data processing framework; The reduction of response latency through caching optimization includes accelerating access through caching mechanisms, storing hot data, recommendation results, and user historical behavior data in the cache.

6. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 1, characterized in that: The high-concurrency data processing load includes server CPU utilization, memory usage, and I / O latency. The feedback loop delay includes network latency, data transmission latency, and processing latency.

7. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 6, characterized in that: For each monitoring period, a weighted sum of the high-concurrency data processing load and feedback loop latency is calculated based on the high-concurrency data processing load and feedback loop latency, and defined as the total load latency index. The specific calculation formula is as follows: ,in The total load delay index, For high-concurrency data processing loads, For feedback loop delay, These are preset proportional coefficients for high-concurrency data processing load and feedback loop latency, respectively. All are greater than 0.

8. The modular website integrated generation method based on multimodal AI and closed-loop optimization according to claim 7, characterized in that: The stability of closed-loop optimization is analyzed based on the total load delay index, as follows: if the total load delay index exceeds the preset total load delay index threshold, the closed-loop optimization of the system is considered to be in an unstable state. If the closed-loop optimization is in an unstable state, the fault tolerance mechanism and data cleaning strategy will be triggered.

9. A modular website integrated generation system based on multimodal AI and closed-loop optimization, used to implement the modular website integrated generation method based on multimodal AI and closed-loop optimization as described in any one of claims 1-8, characterized in that: It includes modules for data acquisition and preprocessing, content generation and structure building, personalized content generation and recommendation, feedback and optimization, high-concurrency request processing and optimization, and performance monitoring and fault tolerance. The data acquisition and preprocessing module is used to collect users' text, voice, and image multimodal data in real time and perform noise reduction, time alignment and synchronization processing, extract user behavior features, and generate a unified data stream; The content generation and structure building module is used to input data streams into the content generation engine, select functional components based on deep learning models and build the website structure, and dynamically adjust the page layout using responsive design. The personalized content generation and recommendation module is used by the content generation engine to generate personalized content based on user behavior characteristics, and then pass the personalized content to the recommendation system to push relevant content and functions, while monitoring and collecting user behavior feedback data. The feedback and optimization module is used to transmit user behavior feedback data to the optimization engine and adjust the content, recommendations and layout based on real-time feedback; The high-concurrency request processing and optimization module is used to distribute requests through load balancing under high-concurrency feedback adjustment, process large-scale data using a distributed data processing framework, and reduce response latency through caching optimization. The performance monitoring and fault tolerance module is used to continuously monitor the system's high-concurrency data processing load and feedback loop latency, analyze the stability in closed-loop optimization, and determine whether to trigger fault tolerance mechanisms and data cleaning strategies.

Images