Personalized visual communication element recommendation method and system based on Internet big data

Abstract

The invention relates to the field of visual element recommendation, in particular to a personalized visual communication element recommendation method and system based on Internet big data. The method comprises the following steps: acquiring user multi-dimensional internet behavior data, and mining platform user interaction behaviors one by one to obtain a user multi-platform interaction content data pool; performing user interaction intention speculation on the user multi-platform interaction content data pool, and performing personalized interaction interest preference mining to obtain user interaction interest preferences; deep interaction demand mining is carried out according to user interaction interest preference, adaptive portrait evolution is carried out, and a personalized user demand portrait is constructed; and performing visual element analysis and multi-modal visual mining according to the user multi-platform interactive content data pool to obtain multi-modal visual element information of each homepage. According to the method and the device, the accuracy and the quality of visual element recommendation are improved through accurate user aesthetic demand analysis.

Description

Technical Field

[0001] The present invention relates to the field of visual element recommendation, and in particular to a method and system for recommending personalized visual communication elements based on Internet big data. Background Art

[0002] With the continuous development of information technology and digital media, visual communication, as the most direct and influential form of human information interaction, is playing an increasingly important role in a variety of fields, including advertising and marketing, user interface design, and digital content production. With the continuous expansion of the internet user base and the rapid development of big data technology, the increasing amount of user behavior data on internet platforms provides a solid data foundation for more accurate and efficient personalized visual communication.

[0003] In the current internet landscape, users engage in a vast array of interactions daily across multiple channels, including social media, e-commerce platforms, and content platforms. These interactions not only reflect their content preferences and interests, but also reveal diverse individual characteristics in terms of visual aesthetics, cultural preferences, and interaction styles. Traditional visual communication methods often rely on manual experience or generic templates for content design and presentation. These methods lack the ability to deeply explore and precisely adapt to individual user differences, making them unable to meet the growing demand for personalized expression and visual experience among modern users.

[0004] Existing personalized recommendation methods primarily focus on recommending structured data such as text, products, or videos, but still face numerous challenges in recommending visual communication elements. On the one hand, user visual preferences are highly subjective and dynamic, making them difficult to accurately model using single-dimensional data. On the other hand, existing systems are insufficiently capable of integrating and processing multimodal factors such as user aesthetic characteristics, cultural background, and interaction habits, resulting in recommendations that lack specificity and appeal in visual expression. Furthermore, traditional methods, which mostly employ static rules or simple label matching mechanisms, lack a deep understanding of user behavior semantics and the ability to intelligently generate visual elements, making it impossible to achieve truly personalized visual communication. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes a personalized visual communication element recommendation method and system based on Internet big data to solve at least one of the above technical problems.

[0006] To achieve the above objectives, the present invention provides a personalized visual communication element recommendation method based on Internet big data, comprising the following steps: Step S1: Obtain multi-dimensional Internet behavior data of users, and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; Step S2: Inferring user interaction intentions from the user multi-platform interaction content data pool and mining personalized interaction interest preferences to obtain user interaction interest preferences; Step S3: Conduct deep interaction demand mining based on user interaction interest preferences, and perform adaptive profile evolution to build a personalized user demand profile; Step S4: Perform visual element analysis and multimodal visual mining based on the user multi-platform interactive content data pool to obtain multimodal visual element information of each homepage; Step S5: performing user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user dynamic aesthetic situation; Step S6: Based on the user's dynamic aesthetic situation, multi-visual element driven learning is performed on the personalized user demand portrait, and feedback compensation optimization is performed to build an intelligent visual element generation model.

[0007] This invention integrates user behavior data from multiple platforms (such as social media, e-commerce, content platforms, and search engines) to comprehensively analyze user behavior characteristics across different environments. Mining the interactive behavior characteristics of each platform helps to form a clearly structured and data-rich interactive content data pool, providing a solid foundation for subsequent preference inference and recommendation. Through cross-platform comparison and cross-validation, it identifies the differentiated behaviors exhibited by users on different platforms, forming a multidimensional behavioral matrix that facilitates accurate modeling of users' true interests. Intent recognition techniques (such as deep semantic parsing and contextual modeling) are used to infer users' true interaction motivations and goals from their surface behaviors. It identifies changes in user preferences over time and in different contexts, dynamically mapping user interests. By integrating intent and behavior, it avoids preference assumptions based solely on frequency, achieving a deeper understanding of user preferences and laying a precise foundation for subsequent recommendations. It extends from explicit interests and preferences to implicit needs (such as aesthetic trends, potential purchase intentions, and cognitive styles) to uncover users' potential needs. User profiles are not static; they continuously evolve in real time based on new user behaviors, ensuring that the profiles remain relevant to the user's current state and adapt to fluctuations in their interests. This system constructs demand profiles at three levels: semantic, aesthetic, and behavioral, evolving from "user labels" to "user cognitive models." It integrates visual elements such as images, text, typography, and animation to achieve a more realistic and three-dimensional understanding of webpage / platform content. It combines visual elements with user semantic interests to analyze the potential correlations between patterns, colors, and fonts and user semantic preferences. It transcends the limitations of traditional text-based recommendations by incorporating visual communication into the system's analysis dimension, making recommendations more intuitive and aesthetically pleasing. It taps into users' deep aesthetic and cultural foundations based on their cultural background, aesthetic preferences, and regional styles. By adapting to users' aesthetic trends, it effectively matches visual content with their individual aesthetic sense, making recommendations more compelling. It tracks the evolution of users' aesthetic tastes to achieve "aesthetic trend perception" in visual recommendations, satisfying users' expectations for visual freshness and novelty. Models such as GAN (Generative Adversarial Network) are used to generate visual communication elements that align with users' aesthetics and needs, breaking the constraints of a fixed library of content. Recommendations are continuously optimized through user feedback, enabling visual co-creation between users and the system, allowing the system to continuously "learn" to better understand users. Through feedback mechanisms such as click-through rate, user ratings, and dwell time, reverse learning is carried out to continuously optimize the quality of visual element recommendations and form a self-closed loop.

[0008] In this specification, a personalized visual communication element recommendation system based on Internet big data is provided, which is used to implement the personalized visual communication element recommendation method based on Internet big data as described above, including: The behavior mining module is used to obtain multi-dimensional Internet behavior data of users and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; The interest preference module is used to infer user interaction intentions from the user multi-platform interaction content data pool and conduct personalized interaction interest preference mining to obtain user interaction interest preferences; The user portrait module is used to conduct in-depth interaction demand mining based on user interaction interest preferences, and to perform adaptive portrait evolution to build personalized user demand portraits; The visual element parsing module is used to perform visual element parsing and multimodal visual mining based on the user multi-platform interactive content data pool to obtain the multimodal visual element information of each homepage; An aesthetic situation analysis module is used to perform user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user's dynamic aesthetic situation; The feedback compensation optimization module is used to conduct multi-visual element driven learning of personalized user demand portraits based on the user's dynamic aesthetic status, and perform feedback compensation optimization to build an intelligent visual element generation model.

[0009] By monitoring user behavior across multiple internet platforms (such as social media, e-commerce websites, and news platforms), the present invention can obtain rich user interaction data, including clicks, browsing, comments, and sharing, laying a solid foundation for subsequent analysis. Mining user interaction across different platforms helps identify user behavior patterns and habits, understanding their responses in different situations, and thus providing a basis for personalized recommendations. By analyzing the user interaction content data pool, it is possible to infer user interests and intentions, providing basic data for personalized recommendation systems and helping the system more accurately understand user needs. Deeply exploring user interaction interest preferences can help platforms recommend content that better suits their interests, thereby improving user satisfaction and retention. Through deep demand mining, dynamic, personalized user profiles can be generated that reflect users' latest needs and preferences. These profiles can evolve as user behavior changes, maintaining timeliness and accuracy. The construction of personalized user demand profiles helps companies develop more targeted strategies for product promotion and marketing, thereby improving conversion rates. Parsing the visual elements of user interaction content can yield rich multimodal visual information, including images, videos, and graphics. This provides multi-dimensional data support for understanding user aesthetics and preferences. Multimodal visual mining enables recommendation systems to provide more diverse and rich visual content, enhancing the user experience. Analyzing users' aesthetic and cultural backgrounds can help platforms understand the aesthetic tendencies of users from different cultural backgrounds, enabling them to make recommendations more aligned with their cultural identity. Constructing a user's dynamic aesthetic posture can reflect changes in their aesthetic needs in real time, enabling the platform to promptly adjust its recommendation strategy and enhance the user experience. Based on users' dynamic aesthetic postures, through multi-visual element-driven learning, an intelligent visual element generation model can be established. This model can automatically generate visual content that meets user preferences, enhancing the intelligence of recommendations. A feedback compensation optimization mechanism can continuously improve recommendation effectiveness based on actual user feedback, ensuring that the system's adaptability remains consistent with evolving user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flowchart of the steps of a personalized visual communication element recommendation method based on Internet big data of the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION

[0011] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0012] This application example provides a personalized visual communication element recommendation method and system based on internet big data. The execution entities of this personalized visual communication element recommendation method and system based on internet big data include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc. that are equipped with the system, which can be regarded as general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0013] See also Figures 1 to 4 The present invention provides a personalized visual communication element recommendation method based on Internet big data, and the personalized visual communication element recommendation method based on Internet big data includes the following steps: Step S1: Obtain multi-dimensional Internet behavior data of users, and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; Step S2: Inferring user interaction intentions from the user multi-platform interaction content data pool and mining personalized interaction interest preferences to obtain user interaction interest preferences; Step S3: Conduct deep interaction demand mining based on user interaction interest preferences, and perform adaptive profile evolution to build a personalized user demand profile; Step S4: Perform visual element analysis and multimodal visual mining based on the user multi-platform interactive content data pool to obtain multimodal visual element information of each homepage; Step S5: performing user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user dynamic aesthetic situation; Step S6: Based on the user's dynamic aesthetic situation, multi-visual element driven learning is performed on the personalized user demand portrait, and feedback compensation optimization is performed to build an intelligent visual element generation model.

[0014] This invention integrates user behavior data from multiple platforms (such as social media, e-commerce, content platforms, and search engines) to comprehensively analyze user behavior characteristics across different environments. Mining the interactive behavior characteristics of each platform helps to form a clearly structured and data-rich interactive content data pool, providing a solid foundation for subsequent preference inference and recommendation. Through cross-platform comparison and cross-validation, it identifies the differentiated behaviors exhibited by users on different platforms, forming a multidimensional behavioral matrix that facilitates accurate modeling of users' true interests. Intent recognition techniques (such as deep semantic parsing and contextual modeling) are used to infer users' true interaction motivations and goals from their surface behaviors. It identifies changes in user preferences over time and in different contexts, dynamically mapping user interests. By integrating intent and behavior, it avoids preference assumptions based solely on frequency, achieving a deeper understanding of user preferences and laying a precise foundation for subsequent recommendations. It extends from explicit interests and preferences to implicit needs (such as aesthetic trends, potential purchase intentions, and cognitive styles) to uncover users' potential needs. User profiles are not static; they continuously evolve in real time based on new user behaviors, ensuring that the profiles remain relevant to the user's current state and adapt to fluctuations in their interests. This system constructs demand profiles at three levels: semantic, aesthetic, and behavioral, evolving from "user labels" to "user cognitive models." It integrates visual elements such as images, text, typography, and animation to achieve a more realistic and three-dimensional understanding of webpage / platform content. It combines visual elements with user semantic interests to analyze the potential correlations between patterns, colors, and fonts and user semantic preferences. It transcends the limitations of traditional text-based recommendations by incorporating visual communication into the system's analysis dimension, making recommendations more intuitive and aesthetically pleasing. It taps into users' deep aesthetic and cultural foundations based on their cultural background, aesthetic preferences, and regional styles. By adapting to users' aesthetic trends, it effectively matches visual content with their individual aesthetic sense, making recommendations more compelling. It tracks the evolution of users' aesthetic tastes to achieve "aesthetic trend perception" in visual recommendations, satisfying users' expectations for visual freshness and novelty. Models such as GAN (Generative Adversarial Network) are used to generate visual communication elements that align with users' aesthetics and needs, breaking the constraints of a fixed library of content. Recommendations are continuously optimized through user feedback, enabling visual co-creation between users and the system, allowing the system to continuously "learn" to better understand users. Through feedback mechanisms such as click-through rate, user ratings, and dwell time, reverse learning is carried out to continuously optimize the quality of visual element recommendations and form a self-closed loop.

[0015] In the embodiment of the present invention, see Figure 1 , is a flowchart of the steps of a personalized visual communication element recommendation method based on Internet big data of the present invention. In this example, the steps of the personalized visual communication element recommendation method based on Internet big data include: Step S1: Obtain multi-dimensional Internet behavior data of users, and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; In this example, multiple internet platforms are identified for analysis, including social media (such as Facebook and Instagram), e-commerce (such as Amazon and eBay), and content sharing sites (such as YouTube and Pinterest). Data from these platforms will provide a comprehensive perspective for user behavior analysis. Analyze the characteristics of each platform to determine the primary types of user interactions (such as likes, comments, shares, and views). Document each platform's API documentation or data capture strategy. Define the types of user behavior data to be collected, such as user ID, behavior type, timestamp, and content type (images, videos, and text). The goal is to collect a comprehensive record of user behavior on each platform for subsequent analysis. Leverage the APIs provided by each platform to access user behavior data. If a platform does not provide an API, use web crawlers to capture publicly available data. Ensure compliance with each platform's terms of use to avoid data misuse. Design a database structure to effectively store and manage the collected data. The database should contain all relevant information about user behavior records, including user ID, platform name, interaction time, and behavior type. Set a data capture frequency, such as daily, to ensure data timeliness. Plan the amount of data to capture from each platform, for example, capturing 1,000 of the latest user behavior records each time. Clean and preprocess the collected user behavior data to remove duplicate records and outliers. Ensure that all data is consistent and accurate for subsequent analysis. Classify user behavior data, identify different types of user interaction behaviors (such as likes, comments, shares, etc.), and assign weights to each behavior. This step can help understand user preferences. Set classification criteria, such as assigning a weight of 1 to likes, a weight of 2 to comments, and a weight of 3 to shares, to better assess user engagement and preferences in subsequent analysis. Integrate user behavior data collected from all platforms into a unified data pool, ensuring consistent data formats for subsequent analysis. This step is the foundation for building multi-dimensional user data analysis. Index the integrated data pool to improve data query efficiency. Ensure that subsequent data analysis can quickly access data for specific users or specific behavior types.

[0016] Step S2: Inferring user interaction intentions from the user multi-platform interaction content data pool and mining personalized interaction interest preferences to obtain user interaction interest preferences; In this embodiment, a user interaction intent inference model is designed. This model should be able to analyze user interactions across different platforms to identify their underlying intent. Common methods include natural language processing (NLP) techniques and machine learning algorithms (such as decision trees and random forests). User interaction behavior data, including comment content, number of likes, and sharing behavior, is extracted from a pool of multi-platform user interaction content. This data serves as the model's input features. The diversity and representativeness of the feature data ensure that the model can accurately infer user intent. The model's input feature dimensions are set, such as the sentiment score of the comment text, the number of likes, and the number of shares. An appropriate training set and test set ratio (e.g., 70% training set, 30% test set) is selected to ensure model effectiveness. Feature extraction is performed on user interaction behaviors. Comments are processed using sentiment analysis tools to extract sentiment scores and keywords. Furthermore, user intent (e.g., information acquisition, product purchase, social interaction, etc.) is annotated based on the type of user behavior (e.g., comment, like). The annotated dataset is used to train the intent inference model. Model performance is evaluated through cross-validation, and model parameters are optimized to improve inference accuracy. Multiple experimental parameters, such as the learning rate and tree depth, can be set for parameter tuning. Inferred user intent is recorded in a database to ensure that each user's interaction behavior has a corresponding intent label. The model's prediction results are analyzed to observe the distribution of each intent type to identify the user's primary intent. Based on the inferred user interaction intent, a personalized interaction interest preference analysis framework is constructed. This framework should be able to integrate the user's historical interaction behavior and inferred intent to extract user interests. Integrating the inferred user intent with interaction behavior data, the user's preferences for specific content or visual elements under different intents are analyzed. For example, a user with the "information acquisition" intent may prefer articles and infographics, while a user with the "social interaction" intent may prefer images and videos. A threshold for preference extraction is set: if a user's interaction frequency on a certain content type is 1.5 times higher than the average, it is considered to have an interest in that content type. This method can identify users' personalized areas of interest.

[0017] Step S3: Conduct deep interaction demand mining based on user interaction interest preferences, and perform adaptive profile evolution to build a personalized user demand profile; In this embodiment, the goal of deep interaction demand mining is to identify potential user needs based on their interaction interests and preferences. These needs may include desires for specific products, services, content types, or experiences. Data related to user interaction interests and preferences is collected, including historical user behavior data, intent inference results, and sentiment analysis results. The comprehensiveness of the data is ensured so that the model can accurately identify user needs. The dataset is divided into a specific ratio (e.g., 70% training set, 30% test set), and appropriate features are selected, including user interaction frequency, behavior type, and sentiment score, for subsequent model training. Feature extraction is performed on the collected data, and methods such as cluster analysis and association rule mining are used to identify users' deep interaction needs. By analyzing users' high-frequency interactions with specific content types, users are identified as having strong potential for that type of content. Machine learning algorithms (e.g., K-means clustering and Apriori algorithm) are applied to model user needs. Model performance is evaluated through cross-validation, and model parameters are adjusted to improve the accuracy of demand mining. The mined user deep interaction needs are recorded in a database, ensuring that each user need has a corresponding description and label. The demand characteristics of different user groups are analyzed to identify common and individual needs. Design the structure of a personalized user demand portrait, including basic user information, interactive interest preferences, deep needs, etc. Ensure that the portrait can fully reflect the user's needs and behavioral characteristics. Establish an adaptive portrait evolution mechanism to dynamically update according to the user's latest interactive behavior and demand changes. Ensure that the user portrait always reflects the user's latest status. Set the frequency of portrait updates, such as updating the user portrait once a week, or triggering an update when the user's behavior changes significantly. Ensure that the portrait can respond to changes in user needs in a timely manner. Record the mined personalized user demand portrait in the database, ensuring that the data format is consistent for subsequent retrieval and analysis. Each user's portrait should include: user ID, interest preferences, deep needs, emotional tendencies and other information. Combine user feedback and behavioral data to optimize the user demand portrait. If the user has positive feedback on a certain type of content, the weight of this type of content will be increased in the portrait.

[0018] Step S4: Perform visual element analysis and multimodal visual mining based on the user multi-platform interactive content data pool to obtain multimodal visual element information of each homepage; In this embodiment, the goal of the visual element parsing model is to extract visual element information from each homepage from a pool of user multi-platform interactive content. This includes data on multiple dimensions, such as color, pattern, font, and layout, for comprehensive visual analysis. Homepage data for each platform is extracted from the data pool, including HTML structure, CSS styles, and image resources. Data integrity is ensured to facilitate accurate subsequent parsing of visual elements. The types of visual elements to be parsed are defined, such as color, shape, font, and layout. Based on the goal, a parsing algorithm is determined for each element, such as using K-means clustering for color extraction and optical character recognition (OCR) for text content. A color extraction algorithm is used to analyze the visual elements of the homepage. Image processing techniques (such as OpenCV) are used to extract the dominant color of each element and calculate the color distribution ratio on the page. If the homepage is primarily blue and white, these colors and their ratios are recorded. Patterns and shapes on the homepage are analyzed, and the shapes of different elements are identified through edge detection and morphological operations. This helps identify the features of visual elements such as buttons and background patterns. OCR (optical character recognition) is used to extract text from the homepage, and the font style, size, and color are recorded. At the same time, analyze the layout structure of elements and identify their relative position and arrangement on the page. Record the extracted visual element information in a database, ensuring that each element has a corresponding label and feature description. Analyze the distribution of visual elements across different homepages for subsequent comparison and evaluation. Integrate the extracted color, pattern, font, and layout information to form multimodal visual element information for each homepage. This integration can comprehensively reflect the visual style and design characteristics of the homepage. Build a multimodal visual information analysis model and train it on the integrated data using deep learning techniques (such as convolutional neural networks). The model should be able to identify relationships between different visual elements and extract key features. Set the training and test set ratios (e.g., 80% training set, 20% test set) and select appropriate feature dimensions (such as RGB color values, font style, and relative layout position) for model training and evaluation. Record the integrated multimodal visual element information in a database, ensuring consistent data formatting for easy subsequent analysis and retrieval. Each homepage should include detailed information about its visual elements. Use visualization tools to generate a chart displaying multimodal visual elements to help intuitively understand the visual characteristics of each homepage. You can generate color distribution maps, layout structure diagrams, etc. to showcase different homepage design styles. Set the display dimensions of the visualization chart, such as color, font, layout, etc., to ensure the comprehensiveness and effectiveness of the displayed information.

[0019] Step S5: performing user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user dynamic aesthetic situation; In this embodiment, key dimensions of a user's aesthetic and cultural characteristics are defined, including cultural background, aesthetic preferences, and emotional responses. By analyzing a user's multimodal visual element information, cultural characteristics related to the user's aesthetics are identified. Relevant data is extracted from the user's visual element information, including color preferences, design style, and pattern usage frequency. This data will provide a foundation for subsequent aesthetic and cultural analysis. Classification criteria for aesthetic and cultural characteristics are established, such as categorizing user color preferences into "warm," "cool," and "neutral" tones, and analyzing the differences in aesthetic tastes among users from different cultural backgrounds. Multimodal visual element information is integrated with historical user interaction data to analyze user aesthetic preferences across different cultural dimensions. Frequent interaction with elements featuring traditional patterns can be inferred as strong traditional cultural aesthetic tendencies. Cluster analysis is used to categorize users based on their aesthetic and cultural characteristics. Using K-means or hierarchical clustering algorithms, user groups with different aesthetic and cultural types are identified and their characteristics analyzed. The analysis results are recorded in a database to ensure that each user's aesthetic and cultural characteristics are described accordingly. Furthermore, a visual chart displays the distribution of cultural characteristics across different user groups for easy identification and comparison. A dynamic aesthetic situation model is constructed based on the user's aesthetic and cultural characteristics. The model should be able to adapt to changes in users' aesthetic tastes over time and reflect their dynamic aesthetic trends. Regularly collect user interaction behavior data to extract the latest visual element preferences and aesthetic cultural characteristics, which will provide real-time data support for the fitting of dynamic aesthetic trends. Set the frequency of model updates, such as updating the model once a month, to ensure that changes in users' aesthetic tastes can be reflected in a timely manner. At the same time, select appropriate feature dimensions for model training, such as user behavior frequency, interactive emotions, etc. Use historical data to train the dynamic aesthetic trend model and evaluate the model's predictive performance through cross-validation methods. Optimize model parameters to improve fitting accuracy. Record the fitted dynamic aesthetic trend in the database, and establish a feedback mechanism to adjust the model based on the user's new behavior and feedback. If the user recently shows a stronger preference for a certain visual style, the model will automatically adjust the relevant weights.

[0020] Step S6: Based on the user's dynamic aesthetic situation, multi-visual element driven learning is performed on the personalized user demand portrait, and feedback compensation optimization is performed to build an intelligent visual element generation model.

[0021] In this embodiment, dynamic user aesthetic state information is integrated with personalized user demand profiles. This process unifies user aesthetic changes, interactive behaviors, and demand characteristics into a comprehensive data framework for subsequent analysis and model building. Key features are extracted from this integrated data, including the user's color preferences, design style, and emotional response. These features become input variables for the intelligent visual element generation model, helping the model understand user needs. The dimensions for feature extraction are set, such as color type, style, and shape. Methods such as principal component analysis (PCA) can be used to reduce the dimensionality, ensuring that the model's input data is both rich and concise, improving computational efficiency. Based on user demand characteristics, a multi-visual element-driven learning model is designed. Appropriate machine learning algorithms, such as convolutional neural networks (CNNs) or generative adversarial networks (GANs) in deep learning, are selected to generate visual elements that align with the user's aesthetic taste. The model is trained to generate new visual elements using historical user interaction data and feedback. Training data can be augmented, for example, through rotation and scaling, to increase sample diversity and improve the model's generalization ability. Set the ratio of the training set to the test set (e.g., 80% training set, 20% test set), as well as model hyperparameters such as the learning rate and batch size. Optimize hyperparameters using methods such as grid search to ensure optimal model performance. During user interaction, collect real-time user feedback on generated visual elements. This can be done through various means, such as ratings, click-through rates, and comments, to generate user preference data for visual elements. Design a feedback compensation mechanism to integrate user feedback into the intelligent visual element generation model. Based on user feedback, adjust the model generation strategy, for example, by increasing the frequency of high-rated elements and reducing the weight of low-rated elements. Set a feedback update frequency, such as weekly, to ensure the model can respond promptly to changing user needs. Develop specific rules for feedback compensation, such as setting a threshold so that only feedback exceeding this threshold will affect model adjustments. Use the integrated user data and feedback to train the intelligent visual element generation model. Through iterative training, continuously optimize model parameters to improve the quality of generated elements and user satisfaction. After model training is complete, evaluate the model using the test set. A variety of metrics, such as the diversity of generated elements and the average user rating, can be used to comprehensively evaluate the model's generation capabilities. Model training results and user feedback can be recorded, and visualization tools can be used to display the distribution of generated elements and changes in user satisfaction. Radar charts can be generated to display the ratings of different visual elements, helping to understand changes in user preferences.

[0022] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Acquire multi-dimensional Internet behavior data of users; clean the Internet behavior data for abnormalities to obtain cleaned and optimized Internet behavior data; Performing a multi-platform content characteristics analysis based on the cleansed and optimized Internet behavior data to obtain content characteristics for each platform; Performing multi-platform difference identification on the content characteristics to generate differential features of the multiple platforms; Classify the platform types according to the differential features and generate content tags for each platform; The cleaned and optimized Internet behavior data is mined for user interaction behaviors on each platform according to the content tags of each platform to obtain a user multi-platform interaction content data pool.

[0023] In this example, target platforms for data collection are identified, including social media, e-commerce, news websites, and video platforms. Each platform should have rich user behavior data, including clicks, browsing time, shares, and comments. After obtaining authorization from the user and the relevant platform, collect user behavior data on social media platforms (such as Facebook and Instagram) and purchase history on e-commerce platforms (such as Amazon). Use tools such as data crawlers or APIs to extract user behavior data from each platform. Ensure that the collected data includes information such as user ID, behavior timestamp, behavior type, and interaction content. Set up a crawler program on each platform to automatically capture the latest user behavior data every hour and store it in a unified database. After data collection, integrate user behavior data from different platforms into a unified data framework to facilitate subsequent processing. Ensure that the data format is consistent for easy analysis. The integrated data framework should include fields such as user ID, platform, behavior time, behavior type, and content ID. Define criteria for identifying abnormal data, including data integrity, rationality, and consistency. Improper timestamps (such as those in the future) or mismatched user behavior types (such as purchases without product IDs) are considered abnormal data. If a user has behavior records on multiple different platforms within the same timestamp, there may be anomalies. Use data cleaning algorithms to identify and eliminate abnormal data. Rule-based filtering, statistical analysis (such as Z-score), and other methods can be used to identify outliers. If a user is found to have more than 100 behaviors in a day, and the behavior type is mainly "browsing", these data will be marked as abnormal. Record the data results after cleaning to ensure the integrity and accuracy of the data set. Verify the cleaning effect through sampling inspection of the cleaning results to ensure that no abnormal data remains. Record the change in data volume before and after cleaning. If the original data is 10,000 and 8,500 remain after cleaning, record this change and analyze the reasons for elimination.

[0024] Clarify the dimensions for content feature analysis, including content type (video, text, images), theme (technology, fashion, entertainment), and interactivity (number of comments, number of shares). Set the analysis content type to "text and image" or "video," and the theme to "news" or "entertainment." Use text analysis and data mining techniques to process cleansed and optimized internet behavior data to extract the content characteristics of each platform. Natural language processing (NLP) techniques can be used for text analysis. For social media platforms, extract keywords from user comments and identify high-frequency words and themes to understand the main content of user interactions. Record the analyzed content characteristics in a database and use visualization tools to display the distribution of content characteristics across platforms to facilitate understanding. Generate radar charts to display the performance of each platform across multiple dimensions (such as interactivity and content type) for intuitive comparison. Select appropriate statistical analysis methods (such as analysis of variance, t-test, etc.) to identify differences in content characteristics across platforms and determine significant differences in these characteristics. Compare user interaction patterns between social media and e-commerce platforms to determine whether there are significant differences in the number of comments, browsing time, and other aspects.

[0025] Based on the statistical analysis results, generate differential features across multiple platforms. These features will help identify the unique characteristics of each platform and provide a basis for subsequent content labeling. If the analysis results show that social media platforms are primarily focused on images and short videos, while e-commerce platforms are primarily focused on user reviews and purchase behavior, these differential features will be recorded. These differential features will be recorded in the system and displayed using visualization tools to facilitate understanding of the differences between platforms. A bar chart will be generated to illustrate the differences in content type and interactivity across platforms to help identify the characteristics of each platform. Based on the differential features generated in the previous step, platform classification criteria will be established. These classifications can be based on content characteristics and user interaction behavior, for example, categorizing platforms as social, transactional, or informational. Social platforms can be defined as those with highly interactive content, while e-commerce platforms can be defined as those primarily focused on transactions. Each platform will be classified and content labels generated for each platform using machine learning algorithms (such as K-means clustering) or rule-based classification methods. If a platform's content characteristics demonstrate strong interactivity and a predominance of images and text, it will be labeled a "social interaction platform." The platform categories and content labels will be recorded in the database and verified to ensure classification accuracy. Conduct cross-validation to ensure the consistency and reliability of classification results. Clarify the definition of user interaction behavior, including likes, comments, shares, and browsing. Identify user interaction behaviors on different platforms based on content tags. Set interaction behaviors on social platforms to include comments and shares, while interaction behaviors on e-commerce platforms include purchases and reviews. Based on cleaned and optimized Internet behavior data and platform content tags, conduct user interaction behavior mining on each platform. Use data mining techniques (such as association rule analysis) to identify user behavior patterns. Analyze user behavior on social platforms and find that users interact more frequently with content on specific topics within a certain period of time. Integrate the mined multi-platform user interaction content data into a user interaction behavior pool to provide data support for subsequent personalized recommendations. Record each user's interaction behavior on different platforms to form a user interaction data pool to facilitate analysis and mining of user preferences.

[0026] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Extract key interactive fonts from the user multi-platform interactive content data pool to obtain multiple key interactive content fonts; Performing deep semantic meaning analysis on the key interactive content fonts to generate deep semantic features for each font; Inferring user interaction intention based on the deep semantic features to generate user interaction intention; Perform personalized interaction emotion recognition on user interaction intentions to generate personalized interaction emotions; Based on the user's multi-platform interactive content data pool, personalized interactive interest preference mining is carried out to obtain the user's interactive interest preference.

[0027] In this embodiment, user interaction content, including comments, shares, and likes, is extracted from a multi-platform user interaction data pool. Data integrity and consistency are ensured, and duplicate records and outliers are removed for subsequent analysis. If the data pool contains records of comments and likes by a user on a specific social media post, these records must be accurate and free of duplication. Text analysis algorithms (such as TF-IDF and TextRank) are used to extract key interaction fonts from user interaction content. TF-IDF measures the importance of words in a document, identifying the most frequently mentioned keywords. Keywords in user comments are analyzed. If "environmental protection" appears frequently in multiple comments, it is identified as a key interaction font. The extracted key interaction fonts are recorded in a database, and a visualization chart is generated to help intuitively understand the user's interaction focus. Appropriate natural language processing (NLP) tools are selected for in-depth semantic analysis. Models such as Word2Vec and BERT can be used to capture the contextual relationships and semantic information of words. The BERT model is used to analyze the word "environmental protection" and identify its meaning in different contexts, such as "environmentally friendly products" and "environmental protection policies."

[0028] Deep semantic analysis is performed on the extracted key interactive fonts to generate deep semantic features for each font. Semantic similarity and related topics are extracted by considering word vector representations and contextual relevance. The model calculates a similarity of 0.85 between "environmental protection" and "sustainable development," indicating a strong correlation between the two in user interactions. Based on these deep semantic features, a user interaction intent inference model is constructed. Classification models (such as decision trees and support vector machines) can be used to classify user interactions and infer their underlying intent. User comments are matched with known intents (such as "purchase intent" and "information acquisition intent") to train the model for improved accuracy. The inference model analyzes user interaction content to identify each user's interaction intent. If a user frequently comments on "environmental protection," their intent is inferred to be to promote environmentally friendly products. If a user discusses "how to reduce plastic use" on multiple platforms, their intent is inferred to be to advocate for environmental protection. The inferred user interaction intent is recorded in a database and verified to ensure the accuracy of the inferred intent. Cross-validation is used to improve the reliability of the model. The record format is: User ID: 001, Interaction Intent: "Focus on environmentally friendly products." Verification result: High accuracy. Select an appropriate sentiment analysis model (such as a sentiment lexicon or deep learning model) to identify the user's personalized interaction sentiment. Sentiment analysis helps understand users' emotional reactions to specific topics. Use an LSTM model to analyze the sentiment of user comments and identify whether the user's sentiment is positive, negative, or neutral.

[0029] Through sentiment analysis, we extract personalized emotional characteristics from user interactions. We analyze user comments on the topic of "environmental protection" and identify their sentiment as positive, negative, or neutral. For example, if a user comments, "I like this environmentally friendly product; it's great," we can identify their sentiment as positive. We record these personalized interaction sentiment characteristics in the system and generate a visual chart showing the distribution of user sentiment across different topics. We clearly define personalized interaction interest preferences, including user preferences for different topics, content types, and interaction methods. We can identify these interests by analyzing user interaction content. If a user frequently interacts with content related to "technological innovation," their interest preference can be defined as "technology." Based on user interaction behavior and emotional characteristics, we can mine personalized interest preferences. We can use methods such as cluster analysis to segment users into different interest groups. Using K-means clustering, we can categorize user interests into categories such as "environmental protection," "technology," and "fashion." We record these mined user interaction interest preferences in a database and establish a feedback mechanism to dynamically adjust recommendation strategies in subsequent recommendations. The record format is: User ID: 001, Interest Preference: "Environmental Protection," Status: Highly Related, facilitating subsequent personalized recommendations.

[0030] In this embodiment, the specific steps of performing personalized interaction interest preference mining based on the user multi-platform interaction content data pool to obtain the user interaction interest preference are: Calculate the interaction frequency and click volume of each platform based on the user's multi-platform interactive content data pool; Extract the timestamp of each interaction behavior based on the user's multi-platform interaction content data pool; Performing user interaction time distribution analysis based on the timestamp to obtain interaction time distribution features; Perform multi-dimensional interaction behavior feature analysis based on the interaction frequency, click volume, and interaction time distribution characteristics to generate a multi-dimensional interaction behavior pattern of the user; Conduct personalized interactive interest preference mining on users' multi-dimensional interactive behavior patterns to obtain users' interactive interest preferences.

[0031] In this example, user behavior data for each platform is extracted from a data pool of user multi-platform interactive content, including interactions such as comments, likes, and shares. Data integrity is ensured, and duplicate records are removed to improve accuracy. Assuming that the collected data includes user interactions on social media, e-commerce, and video platforms, the record format should include fields such as user ID, platform, behavior type, and timestamp. Interactions on each platform are counted, and the total interaction frequency and total clicks for each platform are calculated. Interaction frequency refers to the number of times a user engages in an action on that platform, while clicks refer to the number of links or content clicked by the user. If a user on a social media platform performed 200 interactions (likes, comments, etc.) and clicked 150 links in the past week, the social media interaction frequency and clicks would be recorded as 200 and 150. The calculated interaction frequency and clicks are recorded in a database, and a visualization chart is generated to facilitate analysis of user interactions across different platforms. A bar chart is generated with the platform name on the X-axis and the interaction frequency and clicks on the Y-axis, clearly displaying user activity on each platform. Timestamp Extraction: Extract the timestamp of each interaction from the user's multi-platform interactive content data pool. This step obtains the temporal information of user interactions for subsequent analysis of their temporal distribution. For example, assume a user posts a comment on social media with a timestamp of 2023-04-01 08:00:00. Record the timestamp of each interaction. Organize the extracted timestamps into a unified format to ensure data temporality and consistency. Use a standard time format (such as ISO8601) to store timestamps. Standardize the timestamp format to "YYYY-MM-DD HH:MM:SS" to ensure that all recorded time information can be directly analyzed. Record the extracted timestamps in a database and visualize them to help analyze the temporal characteristics of user interactions. Based on the extracted timestamp data, calculate the distribution of user interactions over different time periods. Time can be divided into hours, days, or weeks to observe changes in interaction behavior. Divide a day into 24 hours and count the number of user interactions within each hour to form a time distribution dataset. Analyze the time distribution data to extract temporal distribution characteristics of interactions, including peak hours, active periods, and quiet periods. You can calculate the proportion of interaction behaviors in each time period to identify the time periods when users are most active. If you find that the user interaction frequency between 8:00 AM and 9:00 AM is significantly higher than that of other time periods, then record this time period as the peak period.

[0032] Interaction time distribution characteristics are recorded in the system and visualized using charts to help intuitively understand user activity levels in different time periods. Heat maps are generated with time periods on the X-axis and interaction frequency on the Y-axis. Color intensity indicates interaction intensity, clearly revealing the temporal patterns of user interactions. Indicators are identified for multi-dimensional interaction behavior characteristics, including interaction frequency, click volume, temporal distribution, and content type. These indicators are combined to form a multi-dimensional user interaction profile. Interaction frequency is defined as the number of times a user engages on a specific platform, and click volume is defined as the number of content items a user clicks on. Based on interaction frequency, click volume, and temporal distribution characteristics, in-depth analysis of user interaction behavior is conducted. Cluster analysis or correlation analysis can be used to identify user behavior patterns and trends. If a user frequently interacts on a social platform and is active between 8:00 PM and 10:00 PM, they are characterized as an "evening active user." Based on multi-dimensional interaction behavior characteristics, a personalized user interest preference model is constructed. Machine learning algorithms (such as collaborative filtering and content recommendation) can be used to identify interest preferences. By combining user interaction frequency and content type, the model is trained to identify users' interests in specific topics. This model analyzes user interaction behavior and uncovers personalized interest preferences. Identify user preferences across different platforms and content types. If the model analyzes that users interact more frequently with "technology"-related content than with other topics, record the user's interest in "technology." Record these user interaction preferences in the system and establish a feedback mechanism to dynamically adjust recommendation strategies. The record format is: User ID: 001, Interest Preference: "Technology," Status: Highly Related, facilitating personalized recommendations.

[0033] In this embodiment, refer to Figure 4 , is a flowchart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Based on the user's personalized interaction emotions and user interaction interest preferences, the interaction topology correlation quantitative analysis is performed to obtain the interaction emotion-interest preference quantitative relationship; Conducting deep interaction demand mining based on the quantitative relationship to generate users' deep interaction demands; Based on the content tags of each platform, potential multi-scenario demand predictions are conducted for users' deep interaction needs, generating multi-scenario demand prediction features. Adaptively evolve the profiles of multi-scenario demand prediction features to build personalized user demand profiles.

[0034] In this embodiment, users' personalized interactive emotions and interactive interest preference data are integrated into a unified data framework. Ensure that the data includes the user ID, interactive emotion score (e.g., positive, negative, neutral), and interactive interest preference (e.g., technology, entertainment, health, etc.). Construct a data table containing the following fields: user ID, emotion score, and interest preference. Assume that user 001 has an emotion score of 0.8 (positive) and an interest preference of "technology." Use statistical analysis methods (such as correlation analysis or regression analysis) to calculate the quantitative relationship between interactive emotions and interest preferences. By calculating the statistical correlation between emotion scores and interest preferences, the degree of correlation between users' emotional attitudes and their interests can be identified. If the average positive emotion score for "technology"-related content is 0.75, while the negative emotion score is 0.2, it can be concluded that the user has strong positive emotions in this area. Based on the quantitative relationship between interactive emotions and interest preferences, a deep interactive demand mining model is constructed. This model can use cluster analysis or association rule analysis to identify users' potential needs. Use the Apriori algorithm to analyze users' interactive behavior, identify frequently occurring emotion-interest combinations, and thus infer deep needs. Through model analysis, identify users' deep interaction needs. These needs may not be directly reflected in their interactions, but can be inferred through a combination of emotions and interests. If users have high sentiment scores in the "technology" category and frequently interact with related content, it can be inferred that they have a need for new technology products. Record the identified deep interaction needs in a database and verify their accuracy through user feedback. Conduct small-scale tests to observe user responses to recommended content. Based on each platform's content tags, identify potential multi-scenario needs related to these deep interaction needs. Content tags should cover the platform's theme, content type, and user interaction methods. Label "technology"-related content with tags such as "Latest Technology News" and "Smart Home Products" to ensure that content tags align with user needs. Build a demand prediction model to predict potential user needs in different scenarios based on users' deep interaction needs and platform content tags. Machine learning algorithms (such as random forests and support vector machines) can be used for demand prediction. If the model detects a user's deep need for "smart home," it can predict the user's potential interest in related products in the future. Record multi-scenario demand forecast characteristics in the system and generate visual charts to help analyze user demand forecast results in different scenarios. Generate trend charts to show changes in user demand in different scenarios (such as home, office, and entertainment) to facilitate understanding of users' potential interests. Determine the components of a personalized user demand profile, including basic user information, interactive behavior characteristics, interests, preferences, and underlying needs. Ensure that the profile fully reflects user needs and behavioral characteristics. Define a user profile including: user ID, interests (such as technology and health), and underlying needs (such as smart home products).Design an adaptive evolution mechanism for user profiles, enabling them to dynamically update based on the user's latest interactions and changing needs. Ensure that the user profile always reflects the user's latest status. If a user's recent frequency of interaction with "smart home" increases, their underlying need will be automatically updated to "interest in smart home products." Record this evolved, personalized user need profile in the system and apply it to the personalized recommendation system to optimize recommendation effectiveness. User ID: 001, Personalized Need Profile: {Interest Preference: "Technology", Underlying Need: "Smart Home Products"}, ensuring that recommendations are dynamically adjusted based on this profile.

[0035] In this embodiment, step S4 includes the following steps: Locate multiple platform homepages based on the user multi-platform interactive content data pool; Identifying visual elements of the plurality of platform homepages and marking the visual communication elements of each homepage; Perform visual element analysis based on the visual communication elements to extract the color, pattern, font and layout information of the homepage visual elements; Performing overall homepage color tone mining on the colors and patterns to generate a homepage visual tone style; Performing element morphological feature analysis based on the font and layout information to generate element morphological characteristics of the homepage; Multimodal visual mining is performed on the visual tone style of the homepage and the morphological characteristics of the homepage elements to obtain multimodal visual element information of each homepage.

[0036] In this example, multiple platforms requiring analysis are identified, such as social media, e-commerce, and news websites. Access to the homepages of these platforms is ensured for subsequent visual element identification and analysis. Target platforms may include Facebook, Instagram, Amazon, and CNN, and the homepage URL for each platform is recorded. A web scraping tool is configured to automatically access and retrieve the homepage content of the selected platforms. This can be achieved using crawler technology or an API interface, ensuring that the complete homepage HTML structure is extracted. Python libraries such as Scrapy or Beautiful Soup are used to crawl the homepage HTML and extract the required elements. The captured homepage data is organized and stored in a unified database. Ensure consistent data formatting for subsequent analysis. The homepage HTML, CSS, and related metadata for each platform are recorded. The recording format includes: platform name, homepage URL, and crawl time, so that it can be traced back for subsequent analysis. Appropriate visual element recognition tools, such as computer vision libraries (OpenCV, Tesseract, etc.), are selected to extract visual elements from the homepages. The tool should be able to recognize information such as images, colors, fonts, and layout. Image processing is performed using OpenCV to identify visual elements such as buttons, images, and backgrounds on the homepages. Label and categorize identified visual elements, including images, buttons, titles, text boxes, and other different types of elements. Ensure each element is labeled for subsequent analysis. If a button is detected, record its type as "button," its location as "top left corner," and label its color and text. Record the labeled visual elements in a database and generate a visual chart to illustrate the distribution of visual elements on each homepage, helping to understand the visual design styles of different platforms. Generate a mind map or flowchart to illustrate the visual element structure of each platform, facilitating analysis of its design characteristics. Use color extraction algorithms (such as K-means clustering) to extract the primary colors from visual elements and calculate the proportion of each color on the homepage. This will help identify the homepage's primary color and color scheme. Analyze the homepage's background color, button color, and text color. Using a clustering algorithm, identify the primary color as blue and the secondary color as white. Analyze patterns (such as textures and background patterns) and layout information (such as element arrangement and alignment) on the homepage. Image processing techniques can be used to identify recurring patterns and layout structures. If the homepage uses a grid layout, the position and size of each element within the grid will be recorded, analyzing the layout for symmetry and uniformity. The analyzed color, pattern, and layout information will be recorded in a database, and a visualization chart will be generated to illustrate the visual characteristics of the homepage's elements. A color distribution map will be generated to show the proportions of different colors on the homepage, and a layout diagram will be drawn to show the spatial distribution of elements. Based on the extracted color information, an overall tonal analysis will be performed to identify the homepage's visual tonal style.Color theory (e.g., color wheel, color schemes) can be used to classify colors into warm, cool, or neutral tones. For example, if a homepage is primarily blue and green, it can be categorized as "cool." Identified visual tonal styles are categorized, and descriptions are generated for each style for subsequent use in personalized recommendation systems. The record format is "Homepage ID: 001, Tonal Style: Cool, Description: Mainly blue and green, giving a refreshing and tranquil feeling." Element morphological characteristics are analyzed based on factors such as shape (e.g., rounded, square), corners (e.g., rounded, right-angled), and aspect ratios (e.g., aspect-to-height ratio). If a button is analyzed as a rounded rectangle, its dimensions are recorded as 100px wide and 50px high. Based on the extracted layout information, the element morphology of the homepage is analyzed, and the morphological characteristics of each element are recorded and classified. If multiple circular buttons are identified and evenly distributed in the layout, these are recorded as "round buttons." The morphological feature analysis results are recorded in a database, and a visualization chart is generated to help understand the morphological characteristics of elements on different homepages. Generate a bar chart showing the number of elements with different morphological characteristics, visually reflecting the design style of each platform. Integrate the visual tonal style and element morphological characteristics of the homepages to construct multimodal visual element information. Ensure that the data includes information on multiple dimensions, such as tonal, morphological, and layout. The record format is "Homepage ID: 001, Tonal Style: Cool Tones, Element Form: Round Button, Layout: Grid." Analyze this integrated multimodal visual element information to identify the design characteristics of the homepages across different dimensions. Methods such as cluster analysis can be used to identify homepages with similar design styles. Analysis reveals that certain homepages share commonalities in the use of cool tones and circular elements, allowing them to be grouped together. This multimodal visual element information is recorded in a database, and a visual chart displays the comprehensive visual characteristics of different homepages. Generate a radar chart that comprehensively displays the performance of each homepage across multiple dimensions, such as tonal, morphological, and layout, to facilitate subsequent personalized recommendations.

[0037] In this embodiment, step S5 includes the following steps: Inferring positive and negative elements of the multimodal visual element information based on the user's personalized interactive emotions, and identifying the positive and negative values of each visual element information; Conducting user aesthetic cultural analysis based on the positive and negative values to generate aesthetic cultural characteristics of positive and negative elements; Conduct differential preference analysis of positive and negative elements based on their aesthetic and cultural characteristics to obtain differential preference characteristics of positive and negative elements; Dynamic aesthetic trends are fitted based on the differentiated preference characteristics of positive and negative elements to construct users' dynamic aesthetic trends.

[0038] In this embodiment, sentiment data is extracted from user interactions, recording user feedback on each visual element, including likes, comments, and shares. Data integrity and accuracy are ensured for subsequent analysis. If a user comments on a visual element, "I really like this design," the sentiment score is recorded as positive; if a comment is, "I don't like this color," it is recorded as negative. A sentiment analysis model (such as a sentiment lexicon or deep learning model) is used to analyze user feedback and infer the positive and negative values of each visual element. This model should be able to identify and quantify the emotional tendencies in the text. The BERT model is used to analyze user comments. If the model outputs a sentiment score above 0.5, the element is labeled positive; if it is below 0.5, it is labeled negative. Based on the inferred positive and negative elements, analysis indicators for aesthetic cultural characteristics are defined, including cultural background, aesthetic preferences, and emotional reactions. These characteristics help identify the user's aesthetic culture. Cultural characteristic dimensions are set as follows: traditional culture, modern design, color preference, etc. Statistical analysis is performed on the positive and negative values of each visual element to identify user preferences for different cultural characteristics. Cluster analysis can be used to group users into different aesthetic cultural groups. If a user is found to prefer positive elements associated with traditional culture, their aesthetic cultural characteristics are recorded as "traditional aesthetics." These aesthetic cultural characteristics are recorded in the database, and the distribution of these characteristics across users is displayed graphically to help understand the aesthetic tendencies of the user group. A radar chart is generated to display different users' preferences for traditional and modern design, clearly presenting their aesthetic cultural characteristics. A method for differential preference analysis is developed to identify differences in user preferences for positive and negative elements. Statistical methods such as t-tests or analysis of variance can be used for comparative analysis. User preference scores for positive and negative elements in areas such as color, shape, and layout are compared to analyze the differences. Based on the analysis results, differential preference characteristics for positive and negative elements are extracted, including preferred visual features and design styles. This helps identify subtle differences in user aesthetics. For example, if a user prefers bright colors for positive elements and dark colors for negative elements, these differential characteristics are recorded. These differential preference characteristics are recorded in the database, and the analysis results are displayed visually to help understand the differences in user preferences for positive and negative elements. A two-column chart is generated to compare the scores of positive and negative elements across various characteristic dimensions, visually demonstrating differences in user preferences. Based on the differentiated preference characteristics of positive and negative elements, a dynamic user aesthetic attitude model is constructed. This model should be able to reflect changes in user aesthetics in real time and adapt to user feedback and needs. Time series analysis methods are used to fit user aesthetic attitudes and identify trends in their aesthetics over different time periods. The dynamic aesthetic attitude model is trained using user interaction data and aesthetic feedback to optimize its predictive capabilities. The model should be able to predict future aesthetic tendencies based on users' historical behavior.If user feedback over the past month indicates an increased preference for bright colors, the model will adjust its color strategy for future recommendations. The fitted dynamic aesthetic trends are recorded in a database, and the model's effectiveness is verified by evaluating its predictive accuracy. Model performance can be observed through methods such as A / B testing. User satisfaction ratings after using the new model are recorded; if there is a significant improvement, the model's effectiveness is confirmed.

[0039] In this embodiment, step S6 includes the following steps: Based on users' dynamic aesthetic trends, multi-visual element-driven learning is conducted on personalized user demand portraits to build a dynamic visual element recommendation engine; Generate multiple visual communication simulation elements based on the dynamic visual element recommendation engine, present them to users visually, and collect real-time user feedback information; Analyze user real-time feedback information to obtain user feedback needs; Based on user feedback requirements, the dynamic visual element recommendation engine is optimized through feedback compensation to build an intelligent visual element generation model.

[0040] In this embodiment, historical user interaction data, emotional feedback, and preference information are collected to analyze users' dynamic aesthetic trends. This data may include user behavior records, interaction emotions, and interest preferences across different platforms. User activity over the past month is analyzed, recording the frequency of interaction and emotional ratings for different visual elements (such as color, shape, and layout) to ensure data integrity and accuracy. Based on the collected dynamic user aesthetic trend data, a multi-visual element-driven learning model is constructed. Deep learning algorithms (such as convolutional neural networks) can be used to identify and learn users' visual preferences. The model is trained using user interaction data to identify patterns in user preferences for visual elements, such as a preference for warm colors and minimalist layouts. The trained model is integrated into a recommendation engine to implement dynamic visual element recommendations. The engine should be able to analyze users' aesthetic trends in real time and adjust recommendations based on changes. If the model identifies a recent increase in a user's preference for "natural-style" elements, the recommendation engine will prioritize these visual elements. A visual element generation mechanism is designed to automatically generate multiple visual communication simulation elements based on the recommendation engine's output. These elements should cover different styles, tones, and shapes to meet users' diverse needs. Generated simulation elements can include buttons with different color combinations, background patterns, and font styles to provide a variety of options. The generated visual communication simulation elements are presented to users through a visual interface. Design a simple and intuitive user interface that allows users to easily browse and select different visual elements. Create a visual display page containing thumbnails of various simulation elements, allowing users to click to view detailed information. As users browse and select visual elements, collect real-time user feedback. This can be done through various methods, such as questionnaires, rating systems, or click-through rates. After viewing the simulation elements, users can be asked to rate their satisfaction (e.g., on a scale of 1 to 5), and their choices and preferences are recorded. Organize and categorize the collected real-time user feedback. Ensure a uniform data format for subsequent analysis. Record each user's rating, selection, and feedback on different simulation elements. Create a data table containing fields such as user ID, element ID, rating, and comment to ensure data integrity. Build a feedback demand analysis model based on user feedback. Natural language processing techniques can be used to analyze user text comments to extract keywords and sentiment. If a user comments, "This color looks good," this is considered positive feedback, and the color preference is recorded. The parsed user feedback requirements are recorded in a database and visualized to help understand overall user feedback trends. A bar chart is generated to display the average ratings of different visual elements and the sentiment distribution of user feedback to help optimize recommendation strategies. Based on user feedback requirements, feedback compensation is optimized for the dynamic visual element recommendation engine. A mechanism is designed to enable the engine to dynamically adjust based on negative user feedback. If a user gives a low rating to an element, its frequency of appearance is reduced in subsequent recommendations.Based on a feedback compensation mechanism, an intelligent visual element generation model is constructed. This model should be able to adapt to changing user needs and dynamically adjust the style and type of generated visual elements. If users have expressed a preference for "minimalist design" in past feedback, the model will prioritize generating visual elements that match this style. The effectiveness of the optimized recommendation engine and generation model is recorded and evaluated in the system. Through methods such as A / B testing, the impact of the new model on user satisfaction and interaction behavior is observed. Changes in user satisfaction scores after using the new model are recorded. If there is a significant improvement, the optimization is considered successful.

[0041] In this embodiment, a personalized visual communication element recommendation system based on Internet big data is provided, which is used to execute the personalized visual communication element recommendation method based on Internet big data as described above, including: The behavior mining module is used to obtain multi-dimensional Internet behavior data of users and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; The interest preference module is used to infer user interaction intentions from the user multi-platform interaction content data pool and conduct personalized interaction interest preference mining to obtain user interaction interest preferences; The user portrait module is used to conduct in-depth interaction demand mining based on user interaction interest preferences, and to perform adaptive portrait evolution to build personalized user demand portraits; The visual element parsing module is used to perform visual element parsing and multimodal visual mining based on the user multi-platform interactive content data pool to obtain the multimodal visual element information of each homepage; An aesthetic situation analysis module is used to perform user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user's dynamic aesthetic situation; The feedback compensation optimization module is used to conduct multi-visual element driven learning of personalized user demand portraits based on the user's dynamic aesthetic status, and perform feedback compensation optimization to build an intelligent visual element generation model.

[0042] By monitoring user behavior across multiple internet platforms (such as social media, e-commerce websites, and news platforms), the present invention can obtain rich user interaction data, including clicks, browsing, comments, and sharing, laying a solid foundation for subsequent analysis. Mining user interaction across different platforms helps identify user behavior patterns and habits, understanding their responses in different situations, and thus providing a basis for personalized recommendations. By analyzing the user interaction content data pool, it is possible to infer user interests and intentions, providing basic data for personalized recommendation systems and helping the system more accurately understand user needs. Deeply exploring user interaction interest preferences can help platforms recommend content that better suits their interests, thereby improving user satisfaction and retention. Through deep demand mining, dynamic, personalized user profiles can be generated that reflect users' latest needs and preferences. These profiles can evolve as user behavior changes, maintaining timeliness and accuracy. The construction of personalized user demand profiles helps companies develop more targeted strategies for product promotion and marketing, thereby improving conversion rates. Parsing the visual elements of user interaction content can yield rich multimodal visual information, including images, videos, and graphics. This provides multi-dimensional data support for understanding user aesthetics and preferences. Multimodal visual mining enables recommendation systems to provide more diverse and rich visual content, enhancing the user experience. Analyzing users' aesthetic and cultural backgrounds can help platforms understand the aesthetic tendencies of users from different cultural backgrounds, enabling them to make recommendations more aligned with their cultural identity. Constructing a user's dynamic aesthetic posture can reflect changes in their aesthetic needs in real time, enabling the platform to promptly adjust its recommendation strategy and enhance the user experience. Based on users' dynamic aesthetic postures, through multi-visual element-driven learning, an intelligent visual element generation model can be established. This model can automatically generate visual content that meets user preferences, enhancing the intelligence of recommendations. A feedback compensation optimization mechanism can continuously improve recommendation effectiveness based on actual user feedback, ensuring that the system's adaptability remains consistent with evolving user needs.

[0043] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0044] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A personalized visual communication element recommendation method based on Internet big data, characterized by: The following steps are involved: Step S1: Obtain multi-dimensional Internet behavior data of users, and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; Step S2: Inferring user interaction intentions from the user multi-platform interaction content data pool and mining personalized interaction interest preferences to obtain user interaction interest preferences; Step S3: Conduct deep interaction demand mining based on user interaction interest preferences, and perform adaptive profile evolution to build a personalized user demand profile; Step S4: Perform visual element analysis and multimodal visual mining based on the user multi-platform interactive content data pool to obtain multimodal visual element information of each homepage; Step S5: performing user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user dynamic aesthetic situation; Step S6: Based on the user's dynamic aesthetic situation, multi-visual element driven learning is performed on the personalized user demand portrait, and feedback compensation optimization is performed to build an intelligent visual element generation model.

2. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S1 are: Acquire multi-dimensional Internet behavior data of users; clean the Internet behavior data for abnormalities to obtain cleaned and optimized Internet behavior data; Performing a multi-platform content characteristics analysis based on the cleansed and optimized Internet behavior data to obtain content characteristics for each platform; Performing multi-platform difference identification on the content characteristics to generate differential features of the multiple platforms; Classify the platform types according to the differential features and generate content tags for each platform; The cleaned and optimized Internet behavior data is mined for user interaction behaviors on each platform according to the content tags of each platform to obtain a user multi-platform interaction content data pool.

3. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S2 are: Extract key interactive fonts from the user multi-platform interactive content data pool to obtain multiple key interactive content fonts; Performing deep semantic meaning analysis on the key interactive content fonts to generate deep semantic features for each font; Inferring user interaction intention based on the deep semantic features to generate user interaction intention; Perform personalized interaction emotion recognition on user interaction intentions to generate personalized interaction emotions; Based on the user's multi-platform interactive content data pool, personalized interactive interest preference mining is carried out to obtain the user's interactive interest preference.

4. The personalized visual communication element recommendation method based on Internet big data according to claim 3 is characterized in that: The specific steps of mining personalized interaction interest preferences based on the user multi-platform interaction content data pool to obtain the user interaction interest preferences are as follows: Calculate the interaction frequency and click volume of each platform based on the user's multi-platform interactive content data pool; Extract the timestamp of each interaction behavior based on the user's multi-platform interaction content data pool; Performing user interaction time distribution analysis based on the timestamp to obtain interaction time distribution features; Perform multi-dimensional interaction behavior feature analysis based on the interaction frequency, click volume, and interaction time distribution characteristics to generate a multi-dimensional interaction behavior pattern of the user; Conduct personalized interactive interest preference mining on users' multi-dimensional interactive behavior patterns to obtain users' interactive interest preferences.

5. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S3 are: Based on the user's personalized interaction emotions and user interaction interest preferences, the interaction topology correlation quantitative analysis is performed to obtain the interaction emotion-interest preference quantitative relationship; Conducting deep interaction demand mining based on the quantitative relationship to generate users' deep interaction demands; Based on the content tags of each platform, potential multi-scenario demand predictions are conducted for users' deep interaction needs, generating multi-scenario demand prediction features. Adaptively evolve the profiles of multi-scenario demand prediction features to build personalized user demand profiles.

6. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S4 are: Locate multiple platform homepages based on the user multi-platform interactive content data pool; Identifying visual elements of the plurality of platform homepages and marking the visual communication elements of each homepage; Perform visual element analysis based on the visual communication elements to extract the color, pattern, font and layout information of the homepage visual elements; Performing overall homepage color tone mining on the colors and patterns to generate a homepage visual tone style; Performing element morphological feature analysis based on the font and layout information to generate element morphological characteristics of the homepage; Multimodal visual mining is performed on the visual tone style of the homepage and the morphological characteristics of the homepage elements to obtain multimodal visual element information of each homepage.

7. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S5 are: Inferring positive and negative elements of the multimodal visual element information based on the user's personalized interactive emotions, and identifying the positive and negative values of each visual element information; Conducting user aesthetic cultural analysis based on the positive and negative values to generate aesthetic cultural characteristics of positive and negative elements; Conduct differential preference analysis of positive and negative elements based on their aesthetic and cultural characteristics to obtain differential preference characteristics of positive and negative elements; Dynamic aesthetic trends are fitted based on the differentiated preference characteristics of positive and negative elements to construct users' dynamic aesthetic trends.

8. The personalized visual communication element recommendation method based on Internet big data according to claim 1 is characterized in that: The specific steps of step S6 are: Based on users' dynamic aesthetic trends, multi-visual element-driven learning is conducted on personalized user demand portraits to build a dynamic visual element recommendation engine; Generate multiple visual communication simulation elements based on the dynamic visual element recommendation engine, present them to users visually, and collect real-time user feedback information; Analyze user real-time feedback information to obtain user feedback needs; Based on user feedback requirements, the dynamic visual element recommendation engine is optimized through feedback compensation to build an intelligent visual element generation model.

9. A personalized visual communication element recommendation system based on Internet big data, characterized by: The method for recommending personalized visual communication elements based on Internet big data according to claim 1 comprises: The behavior mining module is used to obtain multi-dimensional Internet behavior data of users and conduct platform-by-platform user interaction behavior mining to obtain a user multi-platform interaction content data pool; The interest preference module is used to infer user interaction intentions from the user multi-platform interaction content data pool and conduct personalized interaction interest preference mining to obtain user interaction interest preferences; The user portrait module is used to conduct in-depth interaction demand mining based on user interaction interest preferences, and to perform adaptive portrait evolution to build personalized user demand portraits; The visual element parsing module is used to perform visual element parsing and multimodal visual mining based on the user multi-platform interactive content data pool to obtain the multimodal visual element information of each homepage; An aesthetic situation analysis module is used to perform user aesthetic culture analysis and dynamic aesthetic situation fitting on the multimodal visual element information to construct a user's dynamic aesthetic situation; The feedback compensation optimization module is used to conduct multi-visual element driven learning of personalized user demand portraits based on the user's dynamic aesthetic status, and perform feedback compensation optimization to build an intelligent visual element generation model.

Images