An innovative display solution system based on big data

Abstract

The present invention relates to the field of big data, and specifically refers to an innovative display solution system based on big data, comprising a data acquisition module, a data storage module, an interaction module, a data analysis module, a communication module and a display module. The solution performs labeled classification in the data acquisition module, optimizes the classification of labels by using a residual diversified integration method, forms a diversified model structure by combining a deep classifier and a shallow classifier, and introduces a generative adversarial network to generate tail label samples; introduces a C-WOA optimization algorithm in the training of the diversified integration model, adopts an adaptive parameter adjustment mechanism, and dynamically optimizes the model parameter search range and step size; constructs a graph convolutional network model in the data analysis module to analyze user behavior data and data packets, introduces a dynamic graph structure, forms a high-dimensional graph structure distribution by a variational inference method, and further optimizes the model's processing capability for time series data.

Description

Technical Field

[0001] The present invention relates to the field of big data display technology, and in particular to an innovative display solution system based on big data. Background Art

[0002] With the continuous development of big data and artificial intelligence technology, the application scope of data analysis and display systems has gradually expanded. The innovative display solution system is a comprehensive solution that integrates big data analysis, dynamic display technology and user interaction experience. When processing data with tail label characteristics, traditional classification methods have problems of sample scarcity and prediction bias when processing tail label samples. The learning ability of tail label samples is weak, which easily leads to model bias and difficulty in capturing key but scarce information; in the training process of multi-classifier models in traditional systems, the parameter adjustment method is inefficient, the training time is long, and it is difficult to quickly find the global optimal solution. The fixed parameter range makes the optimization results easy to fall into the local optimum; the traditional innovative display system is inaccurate in user behavior analysis and has the problem of inaccurate analysis results. Because static models are mostly used for analysis, the dynamic correlation between data over time is ignored. Especially when processing time series data and complex data associations, traditional analysis methods are difficult to capture dynamic changes and complex internal relationships in data. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an innovative display solution system based on big data. In view of the problem that the traditional classification method has a weak learning ability for tail label samples, labeled classification is performed in the data acquisition module, and the classification of labels is optimized by using the residual diversified integration method. A diversified model structure is formed by combining deep classifiers and shallow classifiers, and a generative adversarial network is introduced to generate tail label samples, thereby enhancing the model learning ability of tail labels, improving the prediction effect of tail labels, and solving the model bias problem caused by the scarcity of tail label samples in traditional machine learning. In view of the low efficiency of general parameter adjustment methods, a C-WOA optimization algorithm (Coupled Whale Optimization) is introduced into the training of the diversified integration model. Algorithm), adopts an adaptive parameter adjustment mechanism to dynamically optimize the model parameter search range and step size, improves the efficiency and accuracy of model parameter optimization, and enables diversified integrated models to converge to the global optimal solution more quickly; to address the problem of inaccurate analysis results, a graph convolutional network model is built in the data analysis module to analyze user behavior data and data packets, introduces a dynamic graph structure, integrates user behavior data and data packets, constructs time series features, combines graph convolutional networks and attention mechanisms on dynamic graphs, captures complex associations between nodes, and forms a high-dimensional graph structure distribution through variational inference methods, further optimizing the model's ability to process time series data.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an innovative display solution system based on big data, including a data acquisition module, a data storage module, an interaction module, a data analysis module, a communication module and a display module, specifically including the following contents:

[0005] The data acquisition module collects data related to internal publicity and its metadata and classifies them by labels, obtains data packets and transmits them to the data storage module;

[0006] The data storage module stores the data packets in a relational database and a NoSQL database;

[0007] The interaction module collects user behavior data, including click times and dwell time, and transmits it to the data analysis module;

[0008] The data analysis module analyzes the user's behavior data and data packets by constructing a graph convolutional network model to generate a display plan;

[0009] The communication module uses HTTP protocol for inter-module transmission;

[0010] The display module presents the promotional content through the display screen according to the display plan.

[0011] Furthermore, label classification is performed in the data collection module, which specifically includes the following steps:

[0012] Step S1: Data cleaning: cleaning and organizing the data related to internal publicity;

[0013] Step S2: Feature extraction, using word segmentation technology and TF-IDF algorithm to extract tags, all tags are recorded as tag sets;

[0014] Step S3: Tag classification. Set a frequency threshold. Tags with a frequency threshold greater than the frequency threshold are head tags, and tags with a frequency threshold less than the frequency threshold are tail tags. The tag set obtained in step S2 is the actual tag. Calculate the frequency of occurrence of each tag in the tag set to obtain the true tag distribution.

[0015] Step S4: Use the residual diversity integration method to optimize the classification of labels, balance the distribution of head labels and tail labels, and obtain an optimized label set;

[0016] Step S5: Generate a data package, combining the optimized label set and the cleaned data related to internal publicity into a data package.

[0017] Further, step S4 specifically includes the following steps:

[0018] Step S41: Model design, designing a diversified integrated model, including a shallow classifier, a deep classifier, a discriminator, and a generator;

[0019] Step S42: Error correction, construct a basic classifier, use the deep neural network model as the basic classifier, predict the head label, obtain the predicted label, and calculate the prediction residual of the basic classifier for the tail label. The formula used is as follows:

[0020] ;

[0021] in, is the prediction residual, is the actual label obtained in step S3, is the predicted label;

[0022] Step S43: model training, inputting prediction residuals, setting a shallow classifier to optimize the prediction of tail labels, training the diversified ensemble model in stages, introducing the C-WOA optimization algorithm, adjusting the parameters of the diversified ensemble model, and obtaining the trained diversified ensemble model;

[0023] Step S44: fusion of results, fusing the predicted labels in step S42 and the prediction of the tail labels optimized by the shallow classifier in step S43 to obtain an optimized label set.

[0024] Furthermore, in step S41, the diversified integration model specifically includes the following contents:

[0025] Building shallow classifiers: Six shallow classifiers were set up, including a random forest model, a SVM model, two single-layer neural networks, an XGBoost model, and a generalized linear model. A MobileNet model was introduced as an additional shallow classifier, and Dropout and Boosting techniques were added to increase the randomness between shallow classifiers.

[0026] Build a deep classifier: Set up a deep classifier and use a pre-trained BERT model as a deep classifier;

[0027] Constructing a generator: Use a generative adversarial network model as a generator to generate false tail label samples, calculate the frequency of occurrence of each label in the false tail label samples, and obtain the generated label distribution;

[0028] Construct a discriminator: Use a convolutional neural network model as the discriminator, and make a judgment based on the tail label obtained in step S3 and the false tail label sample output by the generator to obtain a judgment result.

[0029] Further, in step S43, the C-WOA optimization algorithm specifically includes the following steps:

[0030] Step S431: Initialize the population, randomly generate a set of initial solutions as the population, define the individuals as parameters of the diversified ensemble model, and set the parameter range;

[0031] Step S432: Calculate the fitness, use the fitness function to evaluate the fitness of each individual, and obtain the fitness value of each individual;

[0032] Step S433: Update the optimal solution, compare all fitness values, and select the individual with the highest fitness value as the current optimal solution;

[0033] Step S434: update parameters, dynamically adjust algorithm parameters according to the adaptive mechanism, including update range and search step size;

[0034] Step S435: Update the position of each individual, using the following formula:

[0035] ;

[0036] in, is the updated individual position, is the current individual position, is a random number function, is the parameter for adaptive adjustment, is the difference between the current optimal solution and the current position;

[0037] Step S436: Correct the position. Correct the position beyond the boundary according to the parameter range. The formula used is as follows:

[0038] ;

[0039] ;

[0040] in, To correct the position, is the individual local search parameter, is the step length;

[0041] Step S437: Update the fitness and calculate the fitness value of each individual according to the new position;

[0042] Step S438: repeat steps S434 to S437, and when the current individual fitness is better than the current optimal solution, update the optimal solution;

[0043] Step S439: Output the result, the optimal solution is the optimal parameters of the diversified integration model.

[0044] Furthermore, in the data analysis module, a graph convolutional network model is constructed to analyze the user's behavior data and data packets, which specifically includes the following steps:

[0045] Step Q1: Construct a dynamic graph structure. Construct a dynamic graph based on the user's behavior data and data packets. Set three types of nodes, including user behavior nodes, data packet nodes, and label nodes. The three types of edges are the edge between the user behavior node and the data packet node, the edge between the user behavior node and the label node, and the edge between the data packet node and the label node.

[0046] Step Q2: Feature initialization: Initialize the features of each node, use the BERT model to pre-train and embed all nodes, obtain the features of all nodes and combine them to obtain the initial feature matrix;

[0047] Step Q3: VGCN model construction, applying graph convolution operations on dynamic graphs, introducing attention mechanisms, using RNN to integrate the features of all nodes, and generating time series features on dynamic graphs;

[0048] Step Q4: Introduce the variational graph convolution structure and the variational inference method to form a continuous high-dimensional graph structure distribution based on the time series characteristics on the dynamic graph;

[0049] Step Q5: Model training, training the graph convolutional network model according to the continuous high-dimensional graph structure distribution.

[0050] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0051] (1) To address the problem that traditional classification methods have a weak learning ability for tail label samples, labeled classification is performed in the data acquisition module. The classification of labels is optimized by using the residual diversification ensemble method. A diversified model structure is formed by combining deep classifiers and shallow classifiers. A generative adversarial network is introduced to generate tail label samples, which enhances the model learning ability of tail labels and improves the prediction effect of tail labels. This solves the model bias problem caused by the scarcity of tail label samples in traditional machine learning.

[0052] (2) In view of the low efficiency of general parameter adjustment methods, the C-WOA optimization algorithm is introduced into the training of the diversified integrated model. The adaptive parameter adjustment mechanism is used to dynamically optimize the model parameter search range and step size, which improves the efficiency and accuracy of model parameter optimization and enables the diversified integrated model to converge to the global optimal solution faster.

[0053] (3) To address the problem of inaccurate analysis results, a graph convolutional network model is constructed in the data analysis module to analyze user behavior data and data packets. A dynamic graph structure is introduced to integrate user behavior data and data packets, and time series features are constructed. The graph convolutional network and attention mechanism are combined on the dynamic graph to capture the complex associations between nodes. Finally, a high-dimensional graph structure distribution is formed through the variational inference method, further optimizing the model's ability to process time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of an innovative display solution system based on big data proposed by the present invention;

[0055] Figure 2 A flowchart of the steps of labeling classification proposed by the present invention;

[0056] Figure 3 This is a flow chart of the C-WOA optimization algorithm proposed in the present invention.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] Example 1, see Figure 1 The technical solution adopted by the present invention is as follows: The present invention provides an innovative display solution system based on big data, including a data acquisition module, a data storage module, an interaction module, a data analysis module, a communication module and a display module, which specifically includes the following contents:

[0060] The data acquisition module collects data related to internal publicity and its metadata and classifies them by labels, obtains data packets and transmits them to the data storage module;

[0061] The data storage module stores the data packets in a relational database and a NoSQL database;

[0062] The interaction module collects user behavior data, including click times and dwell time, and transmits it to the data analysis module;

[0063] The data analysis module analyzes the user's behavior data and data packets by constructing a graph convolutional network model to generate a display plan;

[0064] The communication module uses HTTP protocol for inter-module transmission;

[0065] The display module presents the promotional content through the display screen according to the display plan.

[0066] In this embodiment, the data collection module collects data related to internal propaganda through sensors, monitoring systems, and online surveys. When an employee clicks on an internal news link, the data collection module records the behavior and marks the data as "news" or "internal propaganda". These labeled behavior data are converted into data packets and transmitted to the data storage module. When employees watch and interact in front of the display screen, the interaction module records Li Si's behavior, such as watching a video for 3 minutes and clicking the "participate in the event" button. The data is then transmitted to the data analysis module.

[0067] Example 2, see Figure 2 Based on the above embodiment, this embodiment performs label classification in the data acquisition module, and specifically includes the following steps:

[0068] Step S1: Data cleaning: cleaning and organizing the data related to internal publicity;

[0069] Step S2: Feature extraction, using word segmentation technology and TF-IDF algorithm to extract tags, all tags are recorded as tag sets;

[0070] Step S3: Tag classification. Set a frequency threshold. Tags with a frequency threshold greater than the frequency threshold are head tags, and tags with a frequency threshold less than the frequency threshold are tail tags. The tag set obtained in step S2 is the actual tag. Calculate the frequency of occurrence of each tag in the tag set to obtain the true tag distribution.

[0071] Step S4: Use the residual diversity integration method to optimize the classification of labels, balance the distribution of head labels and tail labels, and obtain an optimized label set;

[0072] Step S5: Generate a data package, combining the optimized label set and the cleaned data related to internal publicity into a data package.

[0073] In this embodiment, in step S2, for example, for an article about "company annual meeting", the system will segment its content and extract keywords such as "company annual meeting", "activity arrangement", "employee reward", etc., and then use the TF-IDF algorithm to evaluate the keyword weights of these words to form a tag set, including tags such as "annual meeting", "activity", "employee benefits", etc.;

[0074] In step S3, the label "annual meeting" appears 80% of the time in the entire data set, so it is classified as a head label; while the label "employee reward" appears only 10% of the time, so it is classified as a tail label. By counting the frequencies of all labels, the system generates the frequency of occurrence of each label and obtains the true label distribution of the label set;

[0075] In step S4, the head label "annual meeting" is adjusted by the residual diversification ensemble method to reduce its over-distribution, while increasing the weight of tail labels such as "employee rewards", and finally generating a balanced label set;

[0076] In step S5, the optimized tag set includes tags such as "annual meeting", "employee benefits", and "activity schedule". These tags are combined with corresponding articles, videos, and activity information (such as "2024 employee annual meeting schedule") into a data package, which includes content, tags, time information, and other relevant information (such as participants, event venue, etc.), and is transmitted to subsequent modules for processing.

[0077] Embodiment 3, based on the above embodiment, step S4 specifically includes the following steps:

[0078] Step S41: Model design, designing a diversified integrated model, including a shallow classifier, a deep classifier, a discriminator, and a generator;

[0079] Step S42: Error correction, construct a basic classifier, use the deep neural network model as the basic classifier, predict the head label, obtain the predicted label, and calculate the prediction residual of the basic classifier for the tail label. The formula used is as follows:

[0080] ;

[0081] in, is the prediction residual, is the actual label obtained in step S3, is the predicted label;

[0082] Step S43: model training, inputting prediction residuals, setting a shallow classifier to optimize the prediction of tail labels, training the diversified ensemble model in stages, introducing the C-WOA optimization algorithm, adjusting the parameters of the diversified ensemble model, and obtaining the trained diversified ensemble model;

[0083] Step S44: fusion of results, fusing the predicted labels in step S42 and the prediction of the tail labels optimized by the shallow classifier in step S43 to obtain an optimized label set.

[0084] Embodiment 4: This embodiment is based on the above embodiment. In step S41, the diversified integration model specifically includes the following contents:

[0085] Building shallow classifiers: Six shallow classifiers were set up, including a random forest model, a SVM model, two single-layer neural networks, an XGBoost model, and a generalized linear model. A MobileNet model was introduced as an additional shallow classifier, and Dropout and Boosting techniques were added to increase the randomness between shallow classifiers.

[0086] Build a deep classifier: Set up a deep classifier and use a pre-trained BERT model as a deep classifier;

[0087] Constructing a generator: Use a generative adversarial network model as a generator to generate false tail label samples, calculate the frequency of occurrence of each label in the false tail label samples, and obtain the generated label distribution;

[0088] Construct a discriminator: Use a convolutional neural network model as the discriminator, and make a judgment based on the tail label obtained in step S3 and the false tail label sample output by the generator to obtain a judgment result.

[0089] Example 5, see Figure 3 This embodiment is based on the above embodiment. In step S43, the C-WOA optimization algorithm specifically includes the following steps:

[0090] Step S431: Initialize the population, randomly generate a set of initial solutions as the population, define the individuals as parameters of the diversified ensemble model, and set the parameter range;

[0091] Step S432: Calculate the fitness, use the fitness function to evaluate the fitness of each individual, and obtain the fitness value of each individual;

[0092] Step S433: Update the optimal solution, compare all fitness values, and select the individual with the highest fitness value as the current optimal solution;

[0093] Step S434: update parameters, dynamically adjust algorithm parameters according to the adaptive mechanism, including update range and search step size;

[0094] Step S435: Update the position of each individual, using the following formula:

[0095] ;

[0096] in, is the updated individual position, is the current individual position, is a random number function, is the parameter for adaptive adjustment, is the difference between the current optimal solution and the current position;

[0097] Step S436: Correct the position. Correct the position beyond the boundary according to the parameter range. The formula used is as follows:

[0098] ;

[0099] ;

[0100] in, To correct the position, is the individual local search parameter, is the step length;

[0101] Step S437: Update the fitness and calculate the fitness value of each individual according to the new position;

[0102] Step S438: repeat steps S434 to S437, and when the current individual fitness is better than the current optimal solution, update the optimal solution;

[0103] Step S439: Output the result, the optimal solution is the optimal parameters of the diversified integration model.

[0104] In this example, an initial population is randomly generated, including 5 individuals, each of which contains hyperparameters of a diversified ensemble model. The following parameters need to be optimized:

[0105] The maximum depth of the random forest;

[0106] Regularization parameter of SVM;

[0107] XGBoost learning rate;

[0108] MobileNet’s dropout rate;

[0109] lambda value for GLM;

[0110] Set the parameter range as follows:

[0111] Random Forest max_depth: [5, 20]

[0112] SVM C: [0.1, 10]

[0113] XGBoost learning_rate:[0.001, 0.1]

[0114] MobileNet dropout_rate:[0.1, 0.7]

[0115] GLM lambda: [0.001, 0.1]

[0116] Embodiment 6: Based on the above embodiment, in the data analysis module, a graph convolutional network model is constructed to analyze the user's behavior data and data packets, and specifically includes the following steps:

[0117] Step Q1: Construct a dynamic graph structure. Construct a dynamic graph based on the user's behavior data and data packets. Set three types of nodes, including user behavior nodes, data packet nodes, and label nodes. The three types of edges are the edge between the user behavior node and the data packet node, the edge between the user behavior node and the label node, and the edge between the data packet node and the label node.

[0118] Step Q2: Feature initialization: Initialize the features of each node, use the BERT model to pre-train and embed all nodes, obtain the features of all nodes and combine them to obtain the initial feature matrix;

[0119] Step Q3: VGCN model construction, applying graph convolution operations on dynamic graphs, introducing attention mechanisms, using RNN to integrate the features of all nodes, and generating time series features on dynamic graphs;

[0120] Step Q4: Introduce the variational graph convolution structure and the variational inference method to form a continuous high-dimensional graph structure distribution based on the time series characteristics on the dynamic graph;

[0121] Step Q5: Model training, training the graph convolutional network model according to the continuous high-dimensional graph structure distribution.

[0122] In this embodiment, in step Q4, the continuous high-dimensional graph structure distribution includes node feature distribution parameters and overall distribution features;

[0123] In step Q5, the training process includes the following steps:

[0124] Step Q51: feature sampling, sampling from the node feature distribution to generate a node feature matrix;

[0125] Step Q52: edge weight optimization, using the overall distribution features as supervision signals to update the edge weight matrix of the dynamic graph;

[0126] Step Q53: Convolution training, apply the graph convolutional network on the dynamic graph, combine the node feature matrix and the updated edge weight matrix, and optimize the model parameters through back propagation;

[0127] Step Q54: Output the trained graph convolutional network model.

[0128] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0129] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0130] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An innovative display solution system based on big data, characterized by: An innovative display solution system based on big data, including a data acquisition module, a data storage module, an interaction module, a data analysis module, a communication module and a display module, specifically including the following contents: The data acquisition module collects data related to internal publicity and its metadata and classifies them by labels, obtains data packets and transmits them to the data storage module; The data storage module stores the data packets in a relational database and a NoSQL database; The interaction module collects user behavior data, including click times and dwell time, and transmits it to the data analysis module; The data analysis module analyzes the user's behavior data and data packets by constructing a graph convolutional network model to generate a display plan; The communication module uses HTTP protocol for inter-module transmission; The display module presents the promotional content through the display screen according to the display plan; Label classification is performed in the data collection module, which specifically includes the following steps: Step S1: Data cleaning: cleaning and organizing the data related to internal publicity; Step S2: Feature extraction, using word segmentation technology and TF-IDF algorithm to extract tags, all tags are recorded as tag sets; Step S3: Tag classification. Set a frequency threshold. Tags with a frequency threshold greater than the frequency threshold are head tags, and tags with a frequency threshold less than the frequency threshold are tail tags. The tag set obtained in step S2 is the actual tag. Calculate the frequency of occurrence of each tag in the tag set to obtain the true tag distribution. Step S4: Use the residual diversity integration method to optimize the classification of labels, balance the distribution of head labels and tail labels, and obtain an optimized label set; Step S5: Generate a data package, combining the optimized label set and the cleaned data related to internal publicity into a data package; Step S4 specifically includes the following steps: Step S41: Model design, designing a diversified integrated model, including a shallow classifier, a deep classifier, a discriminator, and a generator; Step S42: Error correction, construct a basic classifier, use the deep neural network model as the basic classifier, predict the head label, obtain the predicted label, and calculate the prediction residual of the basic classifier for the tail label. The formula used is as follows: ； in, is the prediction residual, is the actual label obtained in step S3, is the predicted label; Step S43: model training, inputting prediction residuals, setting a shallow classifier to optimize the prediction of tail labels, training the diversified ensemble model in stages, introducing the C-WOA optimization algorithm, adjusting the parameters of the diversified ensemble model, and obtaining the trained diversified ensemble model; Step S44: fusion of results, fusing the predicted labels in step S42 and the prediction of the tail labels optimized by the shallow classifier in step S43 to obtain an optimized label set.

2. The innovative display solution system based on big data according to claim 1 is characterized by: In step S41, the diversified integration model specifically includes the following contents: Building shallow classifiers: Six shallow classifiers were set up, including a random forest model, a SVM model, two single-layer neural networks, an XGBoost model, and a generalized linear model. A MobileNet model was introduced as an additional shallow classifier, and Dropout and Boosting techniques were added to increase the randomness between shallow classifiers. Build a deep classifier: Set up a deep classifier and use a pre-trained BERT model as a deep classifier; Constructing a generator: Use a generative adversarial network model as a generator to generate false tail label samples, calculate the frequency of occurrence of each label in the false tail label samples, and obtain the generated label distribution; Construct a discriminator: Use a convolutional neural network model as the discriminator, and make a judgment based on the tail label obtained in step S3 and the false tail label sample output by the generator to obtain a judgment result.

3. The innovative display solution system based on big data according to claim 1 is characterized by: Step S43, the C-WOA optimization algorithm specifically includes the following steps: Step S431: Initialize the population, randomly generate a set of initial solutions as the population, define the individuals as parameters of the diversified ensemble model, and set the parameter range; Step S432: Calculate the fitness, use the fitness function to evaluate the fitness of each individual, and obtain the fitness value of each individual; Step S433: Update the optimal solution, compare all fitness values, and select the individual with the highest fitness value as the current optimal solution; Step S434: update parameters, dynamically adjust algorithm parameters according to the adaptive mechanism, including update range and search step size; Step S435: Update the position of each individual, using the following formula: ； in, is the updated individual position, is the current individual position, is a random number function, is the parameter for adaptive adjustment, is the difference between the current optimal solution and the current position; Step S436: Correct the position. Correct the position beyond the boundary according to the parameter range. The formula used is as follows: ； ； in, To correct the position, is the individual local search parameter, is the step length; Step S437: Update the fitness and calculate the fitness value of each individual according to the new position; Step S438: repeat steps S434 to S437, and when the current individual fitness is better than the current optimal solution, update the optimal solution; Step S439: Output the result, the optimal solution is the optimal parameters of the diversified integration model.

4. The innovative display solution system based on big data according to claim 1 is characterized by: In the data analysis module, a graph convolutional network model is built to analyze user behavior data and data packets, which specifically includes the following steps: Step Q1: Construct a dynamic graph structure. Construct a dynamic graph based on the user's behavior data and data packets. Set three types of nodes, including user behavior nodes, data packet nodes, and label nodes. The three types of edges are the edge between the user behavior node and the data packet node, the edge between the user behavior node and the label node, and the edge between the data packet node and the label node. Step Q2: Feature initialization: Initialize the features of each node, use the BERT model to pre-train and embed all nodes, obtain the features of all nodes and combine them to obtain the initial feature matrix; Step Q3: VGCN model construction, applying graph convolution operations on dynamic graphs, introducing attention mechanisms, using RNN to integrate the features of all nodes, and generating time series features on dynamic graphs; Step Q4: Introduce the variational graph convolution structure and the variational inference method to form a continuous high-dimensional graph structure distribution based on the time series characteristics on the dynamic graph; Step Q5: Model training, training the graph convolutional network model according to the continuous high-dimensional graph structure distribution.

Images