A method for automatically identifying design elements to generate activities through deep learning

Abstract

The application discloses a kind of through deep learning automatically identifying design draft element generation activity method, it is related to artificial intelligence technical field, the method includes the following steps: based on the marketing activity design drawing and user information obtained in advance, and by customer group identification and marketing scene identification mode, obtain marketing activity scene;Utilize convolutional neural network model, identify marketing element in marketing activity design drawing, obtain activity component element feature, and according to the position information of activity component element feature extraction activity component, obtain descriptive data;Based on the component library obtained in advance, combine descriptive data and marketing activity scene to obtain activity component template, and according to the activity component template obtained to carry out page rendering, generate marketing activity page.The application generates marketing activity page with complete interaction automatically by convolutional neural network model, to greatly improve marketing activity page development efficiency and customer satisfaction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method for automatically identifying design element generation activities through deep learning. Background Technology

[0002] In terms of digitalization, the widespread adoption of the internet has made online marketing mainstream; for example, the use of social media platforms; big data technology helps collect and analyze massive amounts of user data to achieve precision marketing and personalized recommendations; the development of mobile technology has driven marketing activities to shift towards mobile devices, such as in-app marketing and location-based services (LBS) marketing; artificial intelligence and machine learning are used in marketing to predict user behavior and optimize advertising strategies; marketing automation tools can automate processes, such as automated SMS marketing, automated email marketing, and automated social media management; cloud computing provides powerful support for marketing data storage and processing; and blockchain technology also plays a role in ensuring data security and trust.

[0003] The continuous integration and innovation of the aforementioned technologies have been driving the development and transformation of marketing activities. However, existing technical solutions, which rely on deep learning to identify elements in images, typically generate simple user interfaces (UIs). These UIs cannot generate activity pages with complete interactions and business logic based on different customer groups and business scenarios. Secondary development is still required on the generated simple UI pages to improve related business interactions. This greatly reduces the efficiency of developing marketing activities, thereby significantly increasing manpower and time costs. Consequently, it greatly affects the speed of adapting to market changes and marketing needs.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In response to the problems in related technologies, this invention proposes a method for automatically identifying design draft element generation activities through deep learning, so as to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] A method for automatically identifying design element generation activities using deep learning, the method comprising the following steps:

[0008] S1. Based on the pre-acquired marketing campaign design and user information, and through customer group identification and marketing scenario identification, obtain the marketing campaign scenario;

[0009] S2. Using a convolutional neural network model, identify marketing elements in the marketing campaign design diagram, obtain the features of campaign component elements, and extract the location information of campaign components based on the features of campaign component elements to obtain descriptive data.

[0010] S3. Based on the pre-acquired component library, combine descriptive data and marketing campaign scenarios to obtain campaign component templates, and then render the page according to the obtained campaign component templates to generate the marketing campaign page.

[0011] Furthermore, based on pre-acquired marketing campaign design diagrams and user information, and through customer group identification and marketing scenario identification methods, the marketing campaign scenario is obtained through the following steps:

[0012] S11. Based on the pre-acquired user information, customer groups are segmented using customer group identification methods to obtain customer group tag data;

[0013] S12. Based on the pre-acquired marketing campaign design diagram, marketing scenarios are bound through marketing scenario recognition to obtain the marketing scenario;

[0014] S13. Link customer group tag data with marketing scenarios to obtain marketing campaign scenarios.

[0015] Furthermore, based on pre-acquired user information, customer groups are segmented using customer group identification methods to obtain customer group tag data, including the following steps:

[0016] S111. Based on the pre-acquired user information, create user profiles and analyze user behavior according to user images to extract user features;

[0017] S112. Based on the extracted user characteristics, the users are divided into customer groups to obtain the customer group segmentation results;

[0018] S113. The customer segmentation results are managed by labeling to obtain customer segment label data.

[0019] Furthermore, using a convolutional neural network model, marketing elements in the marketing campaign design diagram are identified to obtain the features of campaign component elements. Based on these features, the location information of the campaign components is extracted to obtain descriptive data, including the following steps:

[0020] S21. Annotate the marketing campaign design diagram to obtain marketing campaign data. Train the convolutional neural network model using the marketing campaign data to obtain the campaign component element recognition model.

[0021] S22. Input the marketing campaign design diagram into the campaign component element recognition model to identify the marketing elements in the marketing campaign design diagram and obtain the characteristics of the campaign component elements.

[0022] S23. Using the object detection algorithm of the convolutional neural network model and combining it with the features of active component elements, identify the bounding box position of the active component elements and obtain the position information of the active component.

[0023] S24. Extract key element information of the marketing campaign based on the characteristics and location information of the campaign components, and convert the key element information into descriptive data through data transformation.

[0024] Furthermore, the marketing campaign design is annotated to obtain marketing campaign data. The convolutional neural network model is then trained using this data to obtain a campaign component element recognition model, including the following steps:

[0025] S211. Annotate the marketing campaign design diagram, obtain marketing campaign data, and initialize the parameters of the convolutional neural network model;

[0026] S212. Input the marketing campaign data into the initialized convolutional neural network model, optimize the convolutional neural network model through the loss function, and use the gradient descent method to calculate the gradient of the loss function with respect to the parameters of the convolutional neural network model.

[0027] S213. Based on the gradient of the convolutional neural network model parameters, update the parameters of the convolutional neural network model using an optimization algorithm to obtain the active component element recognition model, and evaluate the performance of the active component element recognition model through accuracy, recall, and precision.

[0028] Furthermore, the marketing campaign design is input into the campaign component element recognition model to identify the marketing elements in the campaign design and obtain the campaign component element features, including the following steps:

[0029] S221. Input the marketing campaign design diagram into the campaign component element recognition model, and recognize the marketing elements through convolution operation to obtain the local features of the campaign component elements.

[0030] S222. Reduce the feature dimension of local features of active component elements through pooling operations to obtain the main features of active component elements;

[0031] S223. Based on the local features and main features of the active component elements, feature synthesis is performed using a fully connected operation to obtain the features of the active component elements.

[0032] Furthermore, using a convolutional neural network model for object detection, and combining the features of active component elements, the bounding box positions of active component elements are identified, and the extraction of the active component's position information includes the following steps:

[0033] S231. Based on the features of the active component elements, a region proposal network is used to generate candidate regions to obtain a candidate region dataset.

[0034] S232. Convert the candidate region dataset into a fixed-size feature map set, and extract fixed-size features through pooling operations;

[0035] S233. Classify fixed-size features using object classification, and adjust the bounding box positions of fixed-size features using bounding box regression to obtain the position information of the extracted active components.

[0036] Furthermore, key element information includes the name, type, width, height, position, background, and text of the active component.

[0037] Furthermore, based on the pre-acquired component library, combined with descriptive data and marketing campaign scenarios, activity component templates are obtained, and page rendering is performed according to the obtained activity component templates to generate the marketing campaign page, including the following steps:

[0038] S31. Perform data parsing on the descriptive data to obtain the attributes of each activity component, and extract the component type based on the attributes of each activity component;

[0039] S32. Based on the component type and marketing campaign scenario, retrieve the campaign component template from the pre-acquired component library, and convert the campaign component template into campaign message data by combining it with descriptive data;

[0040] S33. Parse the activity message data and render the page based on the parsing results to generate the marketing activity page.

[0041] Furthermore, page rendering based on the parsing results includes registering and finding activity components, laying out activity components on the page, setting the attributes and styles of activity components, and setting the interaction logic of activity components.

[0042] The beneficial effects of this invention are as follows:

[0043] 1. This invention uses a convolutional neural network model to identify marketing elements in marketing campaign design diagrams and automatically generates marketing campaign pages with complete interaction that are suitable for the current customer base and marketing scenarios based on a pre-acquired component library. This greatly improves the development efficiency of marketing campaign pages and customer satisfaction, thereby increasing the rate of adaptation to market changes and marketing needs, and reducing manpower and time costs.

[0044] 2. This invention obtains activity component templates through descriptive data and marketing activity scenarios, which greatly shortens the time from design to activity planning, enables rapid response to market demands, and seizes marketing opportunities; it also reduces the cost of manual planning and design communication, and avoids the waste of resources caused by repeated modifications and adjustments.

[0045] 3. This invention, through customer group identification and marketing scenario identification, can quickly adjust marketing campaign pages according to different marketing campaign design drafts, enhancing adaptability to marketing scenarios and target audiences; it is also suitable for simultaneously processing the generation of multiple marketing campaign design drafts and marketing campaign pages, helping to maintain the efficiency and high quality of marketing campaign page generation in large-scale marketing campaigns. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of a method for automatically identifying design elements and generating activities using deep learning, according to an embodiment of the present invention.

[0048] Figure 2 This is an architecture diagram of a method for automatically identifying design elements and generating activities using deep learning, according to an embodiment of the present invention.

[0049] Figure 3 This is a flowchart illustrating the generation of activity message data in a method for automatically identifying design elements and generating activities using deep learning, according to an embodiment of the present invention.

[0050] Figure 4 This is a flowchart of page rendering in a method for automatically recognizing design elements and generating activities using deep learning, according to an embodiment of the present invention. Detailed Implementation

[0051] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0052] According to an embodiment of the present invention, a method for automatically identifying design element generation activities through deep learning is provided.

[0053] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for automatically identifying design element generation activities through deep learning includes the following steps:

[0054] S1. Based on the pre-acquired marketing campaign design and user information, and through customer group identification and marketing scenario identification, obtain the marketing campaign scenario.

[0055] Specifically, based on pre-acquired marketing campaign design diagrams and user information, and through customer group identification and marketing scenario identification, the process of obtaining marketing campaign scenarios includes the following steps:

[0056] S11. Based on the pre-acquired user information, customer groups are segmented through customer group identification methods to obtain customer group tag data.

[0057] Specifically, based on pre-acquired user information, customer segmentation is performed through customer group identification methods to obtain customer group tag data, including the following steps:

[0058] S111. Based on the pre-acquired user information, create user profiles and analyze user behavior according to user images to extract user features;

[0059] S112. Based on the extracted user characteristics, the users are divided into customer groups to obtain the customer group segmentation results;

[0060] S113. The customer segmentation results are managed by labeling to obtain customer segment label data.

[0061] Specifically, customer segmentation data refers to the characteristic identifiers or attributes used to classify and describe user groups. These tags can be set based on various user information and behaviors, such as age, gender, region, consumption habits, purchase frequency, purchase amount, and preferred product types. By assigning different customer segmentation tags to users, more targeted market segmentation, precision marketing, personalized recommendations, and user services can be carried out to improve operational efficiency and user satisfaction.

[0062] Specifically, the obtained customer group tag data (customerGroupId) will be persistently saved to the database.

[0063] Specifically, analyzing user behavior includes analyzing factors such as the user's gender, age, occupation, investment risk tolerance, historical behavioral data, and information about the transactions they have processed.

[0064] Specifically, based on the extracted user characteristics, users are grouped into customer groups, that is, users with the same user characteristics are grouped into one customer group.

[0065] S12. Based on the pre-acquired marketing campaign design diagram, marketing scenarios are bound through marketing scenario recognition to obtain the marketing scenario;

[0066] S13. Link customer group tag data with marketing scenarios to obtain marketing campaign scenarios.

[0067] Specifically, customer tag data is associated with marketing scenarios, that is, customer tag data is matched with marketing scenarios. Each marketing scenario has corresponding customer tag data. By comparing the consistency of the current user's customer tag data with the customer tag data of the marketing scenario, it is determined whether the current user can participate in the marketing scenario. Users who can participate in the marketing scenario and the marketing scenario together constitute the current marketing activity scenario. At the same time, the marketing scenario and the corresponding customer tag data are persisted to the storage to facilitate the subsequent generation of activity component templates.

[0068] Specifically, marketing scenarios include wealth management scenarios, loan scenarios, etc. Marketing scenarios include marketing campaign names, descriptions, validity periods, etc. They can also be configured to set participation thresholds, whitelist users, or users with specific customer group tags.

[0069] Specifically, the data format generated in marketing scenarios is as follows:

[0070]

[0071] S2. Using a convolutional neural network model, identify marketing elements in the marketing campaign design diagram, obtain the features of campaign component elements, and extract the location information of campaign components based on the features of campaign component elements to obtain descriptive data.

[0072] Specifically, marketing elements include button elements, text fields, product promotional images, and lottery component information in marketing campaign design drawings.

[0073] Specifically, using a convolutional neural network model, marketing elements in the marketing campaign design are identified, the features of campaign component elements are obtained, and the location information of campaign components is extracted based on these features to obtain descriptive data. This process includes the following steps:

[0074] S21. Annotate the marketing campaign design diagram to obtain marketing campaign data. Train the convolutional neural network model using the marketing campaign data to obtain the campaign component element recognition model.

[0075] Specifically, the process of annotating the marketing campaign design to obtain marketing campaign data, and then training a convolutional neural network model using this data to obtain a campaign component element recognition model includes the following steps:

[0076] S211. Annotate the marketing campaign design diagram, obtain marketing campaign data, and initialize the parameters of the convolutional neural network model;

[0077] S212. Input the marketing campaign data into the initialized convolutional neural network model, optimize the convolutional neural network model through the loss function, and use the gradient descent method to calculate the gradient of the loss function with respect to the parameters of the convolutional neural network model.

[0078] S213. Based on the gradient of the convolutional neural network model parameters, update the parameters of the convolutional neural network model using an optimization algorithm to obtain the active component element recognition model, and evaluate the performance of the active component element recognition model through accuracy, recall, and precision.

[0079] Specifically, a convolutional neural network model mainly consists of convolutional layers, pooling layers, and fully connected layers.

[0080] Specifically, the marketing campaign design is annotated to obtain marketing campaign data, which includes marking features related to the marketing campaign, such as text fields, buttons, tables (TAB), product promotional images, etc. The marketing campaign data is then divided into training set, validation set, and test set.

[0081] Specifically, initializing the parameters of a convolutional neural network model involves determining the number of convolutional layers, the size and number of convolutional kernels, the type and size of pooling layers, and the number of nodes in fully connected layers.

[0082] Specifically, the marketing campaign data is input into the initialized convolutional neural network model, i.e., the marketing campaign data passes through convolutional layers, pooling layers, and fully connected layers, and the output value of each neuron is calculated. For the convolutional layers, the value of each position in the output feature map is calculated using the convolution formula. For the fully connected layers, the output value can be calculated using a weighted summation activation function of the neurons, such as using the Rectified Linear Unit (ReLU) activation function, which is a commonly used activation function widely applied in neural networks.

[0083] The expression for calculating the ReLU activation function is:

[0084] f(x) = max(0,x);

[0085] In the formula, f(x) represents the ReLU activation function; x represents the input feature vector.

[0086] When the input x is less than 0, the output is 0; when the input x is greater than or equal to 0, the output is equal to the input value x. The main advantages of the ReLU activation function include simple computation and fast computation speed, which improves the training efficiency of convolutional neural network models; it effectively alleviates the gradient vanishing problem and helps to train deeper convolutional neural network models.

[0087] Assume the input feature vector is x with dimension n, the number of neurons in the fully connected layer is m, the connection weight matrix is ​​W (with dimension m×n), and the bias vector is b (with dimension m); the linear combination is calculated as z = Wx + b, where the multiplication is matrix multiplication and the addition is vector addition.

[0088] Activation functions such as ReLU, sigmoid, and hyperbolic tangent (Tanh) are typically used to process z to obtain the output a. For example, for the ReLU function, a = max(0, z).

[0089] Specifically, optimizing a convolutional neural network model using a loss function involves selecting an appropriate loss function, such as the cross-entropy loss function, to measure the difference between the model's predicted values ​​and the true labels.

[0090] For binary classification problems, the cross-entropy loss function is calculated as follows:

[0091]

[0092] In the formula, L represents the loss value, which is a measure of the difference between the prediction result and the actual result of the convolutional neural network model; N represents the number of samples; i represents the index of the sample, traversing all samples from 1 to N; y i This represents the true label of the i-th sample, with a value of 0 or 1, indicating that it belongs to category 0 or category 1. This represents the probability that the convolutional neural network model predicts the i-th sample, which is the probability that it is predicted to be class 1.

[0093] When y i When = 1, the first term in the loss function It takes effect; at this point, the second item... The cross-entropy is 0; it represents the probability that the convolutional neural network model hopes to predict. The closer to 1, the better, because along with It increases as it approaches 1.

[0094] When y i When = 0, the second term in the loss function It takes effect; the first item at this time... Cross-entropy is 0; it represents the probability that the model hopes to predict. The closer to 0, the better, because along with It increases as it approaches 0.

[0095] The average loss L is obtained by summing these two losses for all samples and taking the average value. This average loss L is used to measure the model's performance in the binary classification task. The smaller the loss, the better the model's performance.

[0096] Specifically, the gradient of the loss function with respect to the parameters of the convolutional neural network model is calculated using the gradient descent method, i.e., backpropagation. The gradient of the loss function with respect to the network parameters (convolutional kernel weights, fully connected layer weights, etc.) is calculated using the gradient descent method. The model parameters are then continuously adjusted using optimization algorithms (such as gradient descent) to minimize this loss function.

[0097] The functions of backpropagation include calculating the output error, i.e., calculating the error of the output layer based on the loss function; calculating the gradient, i.e. calculating the gradient with respect to the weights W and biases b; and updating the weights, i.e. using an optimization algorithm (such as stochastic gradient descent) to update the weights W and biases b based on the calculated gradients.

[0098] In practical applications, fully connected layers are often used in combination with convolutional layers, pooling layers, etc., to form deep neural networks and work together to complete complex learning tasks.

[0099] Specifically, based on the gradients of the convolutional neural network (CNN) model parameters, optimization algorithms are used to update the CNN model parameters. This involves using optimization algorithms (such as Stochastic Gradient Descent (SGD), Adaptive Gradient Descent (Adagrad), and Adaptive Increment (Adadelta)) to update the network parameters according to the calculated gradients. For example, in SGD, the parameter update formula is expressed as:

[0100]

[0101] In the formula, θ t θ represents the parameters at the t-th iteration (e.g., weights and biases in a neural network); t+1 η represents the updated parameters, i.e., the parameters at the (t+1)th iteration; η represents the learning rate, which controls the step size of parameter updates. If the learning rate is too large, the algorithm may fail to converge; if it is too small, the convergence speed may be too slow. The loss function J represents the loss function with respect to the parameter θ. t The gradient.

[0102] Repeat the training and iterative updates until the model's performance on the validation set reaches a satisfactory level, thus obtaining the active component element recognition model.

[0103] Specifically, the performance of the active component element recognition model is evaluated by accuracy, recall, and precision. That is, the accuracy, recall, F1 score, and other metrics of the active component element recognition model are evaluated on the test set to determine the model's performance.

[0104] The steps for using Region of Interest (RoI) pooling to evaluate metrics such as accuracy, recall, and F1 score of a model are as follows:

[0105] First, prediction results are generated. For a given test dataset, the model generates prediction results for each sample, including the predicted region of interest (RoI) and the corresponding class label.

[0106] Second, ground truth labeling, which means that there are already ground truth labels on the test dataset, including real RoIs and class labels.

[0107] Third, match the predictions with the true values, that is, match the predicted RoIs with the true RoIs to determine which are true positives (TP), false positives (FP), false negatives (FN), and true negatives (TN).

[0108] Accuracy is calculated as follows: Accuracy = (Number of correctly predicted samples) / (Total number of samples), or Accuracy = (TP + TN) / (TP + TN + FP + FN).

[0109] Calculate the recall rate: Recall = (number of samples correctly predicted as positive) / (actual number of positive samples), or recall rate = TP / (TP + FN).

[0110] Calculate precision: Precision = (number of samples correctly predicted as positive) / (number of samples predicted as positive), or precision = TP / (TP + FP).

[0111] Calculate the F1 score: F1 = 2 * (Precision * Recall) / (Precision + Recall), which means F1 score = 2 * (Precision * Recall) / (Precision + Recall).

[0112] In actual calculations, the matching and classification of RoIs need to be appropriately defined and processed according to the specific task and data characteristics in order to accurately calculate these metrics.

[0113] S22. Input the marketing campaign design diagram into the campaign component element recognition model to identify the marketing elements in the marketing campaign design diagram and obtain the characteristics of the campaign component elements.

[0114] Specifically, the marketing campaign design is input into the campaign component element recognition model to identify the marketing elements in the campaign design and obtain the characteristics of the campaign component elements, including the following steps:

[0115] S221. Input the marketing campaign design diagram into the campaign component element recognition model, and recognize the marketing elements through convolution operation to obtain the local features of the campaign component elements.

[0116] S222. Reduce the feature dimension of local features of active component elements through pooling operations to obtain the main features of active component elements;

[0117] S223. Based on the local features and main features of the active component elements, feature synthesis is performed using a fully connected operation to obtain the features of the active component elements.

[0118] Specifically, the convolution operation involves sliding a convolution kernel across the input marketing campaign design to extract local features of the campaign component elements.

[0119] Given the marketing campaign design diagram X and the convolution kernel K, the convolution operation can be represented as:

[0120] N = X * K;

[0121] In the formula, * represents the convolution operation; N represents the result of the convolution operation.

[0122] Specifically, N is the feature map obtained after convolving the input marketing campaign design map X with the convolution kernel K.

[0123] Each element in the feature map is obtained by performing convolution calculations on the corresponding position of the input data using a convolution kernel; these feature maps contain the feature information extracted and filtered from the input data by the convolution kernel, providing a foundation for subsequent processing and classification tasks.

[0124] For example, if the input X is a marketing campaign design diagram, then N is the image feature representation obtained after processing with a specific convolutional kernel.

[0125] Specifically, pooling operations reduce the dimensionality of the feature map, thereby reducing computation, while retaining the main features of the active component elements; commonly used methods include max pooling and average pooling.

[0126] Assume the input feature map is H×W (height H, width W), the pooling window is kH×kW (pooling window height kH, width kW), and the stride is sH and sW (stride in the height and width directions).

[0127] Max pooling can be expressed as:

[0128]

[0129] In the formula, Y represents the output after the pooling operation. It is the result obtained after performing max pooling or average pooling on the input feature map, and represents the feature value after pooling processing; Y ijThis represents the value of the element in the output feature map at row i and column j; i represents the index of the row in the output feature map; j represents the index of the column in the output feature map; m and n represent the pooling window regions R used to traverse the corresponding input feature map. ij The index of the inner element; max indicates the operation of taking the maximum value; X m，n This represents the value of the element located at the m-th row and n-th column in the input feature map.

[0130] Pooling reduces the dimensionality of the feature map. The new feature map composed of Y retains the main features while reducing the amount of data and computation, which helps with subsequent processing and analysis.

[0131] The meaning of the max pooling formula is to output the element value Y at position (i,j) in the feature map. ij It is the pooling window region R in the input feature map. ij The maximum value among all elements.

[0132] Average pooling can be expressed as:

[0133]

[0134] In the formula, Q ij This represents the value of the element located at the i-th row and j-th column in the output feature map.

[0135] The meaning of the average pooling formula is to first pool the corresponding pooling window region R in the input feature map. ij All element values ​​X m，n Summing is performed, and then the summation result is divided by the total number of elements in the pooling window region (i.e., kH×kW). The result is the element value Y at position (i,j) in the output feature map. ij .

[0136] Specifically, the steps of a fully connected layer operation include: data preparation, which involves receiving the input feature vector x from the previous layer (usually a convolution or pooling operation); weight initialization, where the fully connected layer has a weight matrix W and a bias vector b, and the parameters are randomly initialized before training begins; linear transformation, which involves calculating the linear combination z = Wx + b, where Wx is matrix multiplication, i.e., multiplying the input feature vector by the weight matrix and then adding the bias vector; activation function application, i.e., inputting the result of the linear transformation z into the activation function to obtain the output a; common activation functions include ReLU, Sigmoid, Tanh, etc.; output propagation, i.e., passing the activated output a to the next layer or as the final output (if this is the last layer of the network); during training, the gradient is calculated based on the loss function, and the weights W and biases b of the fully connected layer are updated through the backpropagation algorithm to optimize the model's performance.

[0137] S23. Using the object detection algorithm of the convolutional neural network model and combining it with the features of the active component elements, identify the bounding box position of the active component elements to obtain the position information of the extracted active component.

[0138] Specifically, the object detection algorithm using a convolutional neural network model, combined with the features of active component elements, identifies the bounding box positions of active component elements, and extracts the position information of active components, including the following steps:

[0139] S231. Based on the features of the active component elements, a region proposal network is used to generate candidate regions to obtain a candidate region dataset.

[0140] S232. Convert the candidate region dataset into a fixed-size feature map set, and extract fixed-size features through pooling operations;

[0141] S233. Classify fixed-size features using object classification, and adjust the bounding box positions of fixed-size features using bounding box regression to obtain the position information of the extracted active components.

[0142] Specifically, the object detection algorithm using a convolutional neural network model, combined with the features of active component elements, identifies the bounding box positions of active component elements and obtains the position information of the active component. That is, the position information of the element is inferred or approximated by using the output of the CNN.

[0143] Object detection algorithms, specifically CNN-based object detection algorithms such as Faster Region-Based Convolutional Neural Networks (Faster R-CNN) and YOLO, can not only identify the object category in an image but also provide the approximate bounding box location of the object, thereby indirectly obtaining the object's location range in the image.

[0144] Faster R-CNN includes a Region Proposal Network (RPN), which generates candidate regions that may contain objects. Specifically, it uses a convolutional neural network to slide a window on the feature map, predicting the probability of the presence and position offset of multiple candidate boxes (anchors) at each window position, thus obtaining a candidate region dataset.

[0145] For each anchor, predict two values: the probability of containing an object (foreground / background) and the position offset (dx, dy, dw, dh), which are used to correct the anchor's position and size.

[0146] Region of Interest (RoI) pooling is a process that converts candidate regions of different sizes into a fixed-size feature map set. Specifically, it involves extracting the corresponding region from the feature map based on the position and size of the candidate region, and then obtaining a fixed-size feature map through pooling.

[0147] The steps to extract the corresponding region from the feature map based on the position and size of the candidate region are as follows:

[0148] First, determine the coordinates of the candidate region on the feature map. That is, assume that the coordinates of the upper left corner of the candidate region are (x1, y1) and the coordinates of the lower right corner are (x2, y2), and the size of the feature map is H×W (height is H and width is W).

[0149] Second, extract the corresponding region, that is, cut off the part from (y1,x1) to (y2,x2) from the feature map to obtain the feature map block corresponding to the candidate region.

[0150] For calculating features of a fixed size (assuming n×n) obtained through pooling operations, taking max pooling as an example, the formula can be expressed as:

[0151]

[0152] In the formula, F ij represents the value in the i-th row and j-th column of the fixed-size feature; w represents the width of the extracted candidate region; h represents the height of the extracted candidate region; f(x,y) is the value at coordinates (x,y) in the candidate region feature map.

[0153] The pooling operation formula means that for each position (i,j) in a fixed-size feature, the maximum value in the corresponding sub-region (determined by i and j) of the candidate region is taken as the feature value of that position.

[0154] Taking average pooling as an example, the formula can be expressed as:

[0155]

[0156] The meanings are similar, except that here it calculates the average value of the feature values ​​within a sub-region.

[0157] Third, classification and regression, namely, classifying and refining the location of candidate regions; using fully connected layers to classify RoI features and obtain object category probabilities; at the same time, predicting the location offset (dx,dy,dw,dh) again to refine the position and size of candidate regions.

[0158] The formula for calculating position offset is usually based on the following form:

[0159]

[0160] Among them, (x a ,y a ) represents the center coordinates of the anchor, w a and h a These are the width and height of the anchor, and (dx,dy,dw,dh) are the predicted offsets.

[0161] Depending on the actual needs, optional measures such as data augmentation, regularization, and hyperparameter tuning can be taken to improve the model's performance and generalization ability.

[0162] S24. Extract key element information of the marketing campaign based on the characteristics and location information of the campaign components, and convert the key element information into descriptive data through data transformation.

[0163] Specifically, the key element information is converted into descriptive data, that is, the data format for generating descriptive key element summary data is as follows:

[0164]

[0165] Specifically, key element information includes the name, type, width, height, position, background, and text of the activity component.

[0166] Specifically, key element information refers to the key marketing information contained in buttons, text, product images, tables, etc.

[0167] S3. Based on the pre-acquired component library, combine descriptive data and marketing campaign scenarios to obtain campaign component templates, and then render the page according to the obtained campaign component templates to generate the marketing campaign page.

[0168] Specifically, page rendering based on the obtained activity component template is performed by a rendering engine, and the activity page is fully displayed on the terminal screen.

[0169] Specifically, based on a pre-acquired component library, descriptive data, and marketing campaign scenarios, activity component templates are obtained. The process of rendering the page based on these templates to generate the marketing campaign page includes the following steps:

[0170] S31. Perform data parsing on the descriptive data to obtain the attributes of each activity component, and extract the component type based on the attributes of each activity component;

[0171] S32. Based on the component type and marketing campaign scenario, retrieve the campaign component template from the pre-acquired component library, and convert the campaign component template into campaign message data by combining it with descriptive data.

[0172] Specifically, a component template refers to a general model or example created in software development or page design for a certain type of components with similar functions and structures. It usually includes the basic structure, style, attribute settings, and default behavior logic of the component. Component templates provide developers or designers with a standardized starting point, so that when creating specific component instances, they can be customized and modified based on the template, thereby improving development efficiency, ensuring consistency, and reducing the possibility of errors.

[0173] For example, when building a webpage, the "button" component has a template that specifies the default size, color, font style, mouse hover effect, etc. Developers can create specific buttons based on this template where they need to use buttons, and adjust certain attributes such as text content and click actions according to specific needs.

[0174] Specifically, activity message data has a unique activity identifier (activityId) and the data is persistently stored.

[0175] S33. Parse the activity message data and render the page based on the parsing results to generate the marketing activity page.

[0176] Specifically, data parsing is performed on descriptive data, which means reading and parsing the descriptive data to extract the attributes of each activity component, such as name, type, width, height, position, background, and text.

[0177] Specifically, the activity component template is converted into activity message data by combining descriptive data. First, a unified basic data structure is defined for the activity component. All activity components and activity component templates follow the unified data structure.

[0178] Specifically, the activity component template is retrieved from a pre-acquired component library; that is, the corresponding activity component template is retrieved from the corresponding component library. This activity component template has already encapsulated the functional implementation of the associated marketing activity scenario; it only needs to render the component to the corresponding position based on the position and width and height data described in the component type; the width, height, and position information in the design draft, etc., from the summary information are filled into the component template. The unified data structure of the activity components is as follows:

[0179]

[0180]

[0181] Here, attributes represent custom properties in the component, applied to the component to achieve specific styles or functions; action represents the data in the array as the IDs of the operations supported by the current component, and the specific operation type and parameters can be queried in the actionList data by ID; actionList represents the data in the array as detailed information on the operations supported by the current component; with this configuration information, after rendering and parsing, the corresponding operation can be completed.

[0182] The data structure stored in actionList is as follows:

[0183]

[0184]

[0185] The above configuration means that a button component has two event handlers, id1 and id2, in its action. The event type of id1 is "init", which means that the component is triggered when it is loaded and initialized. When the rendering engine processes the id1 event, it determines what action to perform based on the type. For example, if type = 7, it will prompt the user "Data loading". The rendering engine then uses the content as the prompt message and displays the specific prompt message.

[0186] The converted activity message data format is as follows:

[0187]

[0188] Specifically, page rendering based on the parsing results includes registering and finding activity components, laying out activity components on the page, setting the attributes and styles of activity components, and setting the interaction logic of activity components.

[0189] Specifically, the activity message data is parsed, meaning the page rendering sends a request to the server, which includes the activityId of the activity to be retrieved. After receiving the request, the server sends the data. The page then receives the JavaScript object representation (JSON) data sent by the server and converts the data into a data object with the following structure:

[0190] [{Component1},{Component2},...,{ComponentN}].

[0191] Specifically, the parsing result, i.e., the data after traversal and parsing, includes:

[0192] First, for each type of component (such as button, tab, text, input, etc.), the corresponding component template is obtained from the corresponding component library.

[0193] Second, set the component's style and position on the page based on the extracted attributes (width, height, position, etc.).

[0194] Third, if the component has text content (such as the Txt attribute), populate the component with that text content.

[0195] Fourth, if the component has a background color or image (such as Background or Sr attributes), set it accordingly.

[0196] Specifically, the registration and search of activity components involves searching for the corresponding component templates based on the business scenario ID and componentType field values ​​of various component libraries registered during the initialization of the rendering engine.

[0197] For example, if "componentType" is "COMPONT_BUTTON", which is a button component, then the relevant information and functions of the button component registered in the corresponding component library will be retrieved.

[0198] Specifically, the layout of active components on the page involves extracting properties such as top, left, width, and height from the rectangle object to determine the component's position and size on the page; setting the component's rotation angle based on the rotate property; processing the zIndex property to determine the component's stacking order; and determining whether the component is fixed to a certain position on the page based on the fixed property.

[0199] Specifically, you can set the properties and styles of the active component, including optional animation effects; if the `animate` object exists and has relevant properties, retrieve the animation name from the `name` field; use `duration` to set the animation duration, `delay` to set the delay time, and `repeat` to set the number of repetitions; set the border by retrieving properties such as `style`, `color`, and `width` from the `border` object, and setting the component's border style, color, and width; set the rounded corners of the border based on the radius properties of the four corners; and apply custom properties from the `attributes` object to the component to achieve specific styles or functions.

[0200] Specifically, set the interaction logic for the activity component. If the action array or actionList object has definitions, set the corresponding actions and interaction logic for the component, such as click events, mouse hover effects, etc.

[0201] The process of executing actions and interactive functions involves loading the configuration data from the component's action and actionList into the currently active button component. For example, if the component template is configured with "attributes", which includes a custom attribute named "buttonLabel" with the value "click button to execute", then when the button is displayed, the label content of the button component is set to "click button to execute". When parsing the "action" array, which lists two event handler IDs, "licai_001" and "daikuan_002", the configuration data corresponding to the ID in "actionList" is retrieved based on the ID.

[0202] The "actionList" object details the specific information for these two events: First, the "licai_001" event is triggered by "click" and has a business scenario type of 1. Assuming type:1, this means the event is triggered when the user clicks the button, sending a network request to obtain configuration parameters such as "apiId" and "param". When the user clicks, the button component's network request interface is called, passing in the parsed "apiId" and "param" information. The apiId uniquely corresponds to the interface of the current business scenario in the service system. After receiving the data, the service system processes it and returns the processing result. The button component completes the data processing returned by the server and displays it on the page. Second, the "licai_002" event has an actionType of "doubleClick", which means that double-clicking the button executes the relevant event. Similarly, the "ApiId" and "params" configurations are parsed, and the component method is called to complete the data processing.

[0203] Example of activity message data during execution:

[0204]

[0205]

[0206] The description explains that attributes are data associated with the current component and can store any data related to the component; action stores the operation IDs corresponding to the marketing scenarios supported by the current component; actionList contains the specific operations corresponding to each action; actionList contains the actions of the current component corresponding to the marketing scenario, such as sending an interface request; based on the above configuration information, the relevant scenario operations are executed when the component is rendered.

[0207] The complete example message is as follows:

[0208]

[0209]

[0210]

[0211]

[0212]

[0213]

[0214]

[0215]

[0216] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatically identifying design draft element generation activities using deep learning, characterized in that, The method includes the following steps: S1. Based on the pre-acquired marketing campaign design and user information, and through customer group identification and marketing scenario identification, obtain the marketing campaign scenario; S2. Using a convolutional neural network model, identify marketing elements in the marketing campaign design diagram, obtain the features of campaign component elements, and extract the location information of campaign components based on the features of campaign component elements to obtain descriptive data. S2 includes: S21. Annotate the marketing campaign design diagram to obtain marketing campaign data. Train the convolutional neural network model using the marketing campaign data to obtain the campaign component element recognition model. S22. Input the marketing campaign design diagram into the campaign component element recognition model to identify the marketing elements in the marketing campaign design diagram and obtain the characteristics of the campaign component elements. S23. Using the object detection algorithm of the convolutional neural network model and combining it with the features of active component elements, identify the bounding box position of the active component elements and obtain the position information of the active component. S24. Extract key element information of the marketing campaign based on the characteristics and location information of the campaign components, and convert the key element information into descriptive data through data transformation; S23 includes: S231. Based on the features of the active component elements, a region proposal network is used to generate candidate regions to obtain a candidate region dataset. S232. Convert the candidate region dataset into a fixed-size feature map set, and extract fixed-size features through pooling operations; S233. Classify fixed-size features using object classification, and adjust the bounding box positions of fixed-size features using bounding box regression to obtain the position information of the extracted active components; S3. Based on the pre-acquired component library, combine descriptive data and marketing campaign scenarios to obtain campaign component templates, and then render the page according to the obtained campaign component templates to generate the marketing campaign page.

2. The method for automatically identifying design element generation activities using deep learning according to claim 1, characterized in that, The process of obtaining marketing campaign scenarios based on pre-acquired marketing campaign design diagrams and user information, and through customer group identification and marketing scenario identification, includes the following steps: S11. Based on the pre-acquired user information, customer groups are segmented using customer group identification methods to obtain customer group tag data; S12. Based on the pre-acquired marketing campaign design diagram, marketing scenarios are bound through marketing scenario recognition to obtain the marketing scenario; S13. Link customer group tag data with marketing scenarios to obtain marketing campaign scenarios.

3. The method for automatically identifying design draft element generation activities using deep learning according to claim 2, characterized in that, The process of segmenting customers based on pre-acquired user information and obtaining customer tag data through customer group identification includes the following steps: S111. Based on the pre-acquired user information, create user profiles and analyze user behavior according to user images to extract user features; S112. Based on the extracted user characteristics, the users are divided into customer groups to obtain the customer group segmentation results; S113. The customer segmentation results are managed by labeling to obtain customer segment label data.

4. The method for automatically identifying design element generation activities using deep learning according to claim 1, characterized in that, The process of annotating the marketing campaign design to obtain marketing campaign data, and then training a convolutional neural network model using that data to obtain a campaign component element recognition model, includes the following steps: S211. Annotate the marketing campaign design diagram, obtain marketing campaign data, and initialize the parameters of the convolutional neural network model; S212. Input the marketing campaign data into the initialized convolutional neural network model, optimize the convolutional neural network model through the loss function, and use the gradient descent method to calculate the gradient of the loss function with respect to the parameters of the convolutional neural network model. S213. Based on the gradient of the convolutional neural network model parameters, update the parameters of the convolutional neural network model using an optimization algorithm to obtain the active component element recognition model, and evaluate the performance of the active component element recognition model through accuracy, recall, and precision.

5. The method for automatically identifying design element generation activities using deep learning according to claim 1, characterized in that, The process of inputting the marketing campaign design diagram into the campaign component element recognition model to identify the marketing elements in the marketing campaign design diagram and obtain the features of the campaign component elements includes the following steps: S221. Input the marketing campaign design diagram into the campaign component element recognition model, and recognize the marketing elements through convolution operation to obtain the local features of the campaign component elements. S222. Reduce the feature dimension of local features of active component elements through pooling operations to obtain the main features of active component elements; S223. Based on the local features and main features of the active component elements, feature synthesis is performed using a fully connected operation to obtain the features of the active component elements.

6. The method for automatically identifying design element generation activities using deep learning according to claim 1, characterized in that, The key element information includes the name, type, width, height, position, background, and text of the active component.

7. The method for automatically identifying design element generation activities using deep learning according to claim 1, characterized in that, The process of obtaining activity component templates based on a pre-acquired component library, combined with descriptive data and marketing campaign scenarios, and then rendering the page according to the obtained activity component templates to generate the marketing campaign page includes the following steps: S31. Perform data parsing on the descriptive data to obtain the attributes of each activity component, and extract the component type based on the attributes of each activity component; S32. Based on the component type and marketing campaign scenario, retrieve the campaign component template from the pre-acquired component library, and convert the campaign component template into campaign message data by combining it with descriptive data; S33. Parse the activity message data and render the page based on the parsing results to generate the marketing activity page.

8. The method for automatically identifying design element generation activities through deep learning according to claim 7, characterized in that, The page rendering based on the parsing results includes registering and finding activity components, laying out activity components on the page, setting the attributes and styles of activity components, and setting the interaction logic of activity components.

Images