Advertisement data statistical method based on deep learning

Abstract

The invention relates to the technical field of advertisement data, and discloses an advertisement data statistical method based on deep learning, comprising: step 1, constructing a user-advertisement interaction matrix of advertisement data, rows of the user-advertisement interaction matrix representing different users, columns of the user-advertisement interaction matrix representing different advertisements, and elements of the matrix representing interaction degrees between the users and the advertisements; and 2, encoding the obtained features of the user and the advertisement, converting the user features and the advertisement features into low-dimensional dense vectors, and mapping the discrete user features and advertisement features through an embedding technology to enable the user features and the advertisement features to be expressed in a low-dimensional space. By adopting the adaptive matrix decomposition method, the matrix rank is dynamically adjusted, the limitation of using a fixed rank in the traditional method is avoided, the potential user and advertisement relation characteristics are effectively extracted from the high-dimensional sparse advertisement data, and higher calculation efficiency and more accurate advertisement effect prediction are obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising data, and specifically provides an advertising data statistics method based on deep learning. Background Art

[0002] In modern advertising placement, the statistics and analysis of advertising data are key links in optimizing advertising effects. With the diversification of advertising placement platforms, advertisers are faced with challenges from various types of advertising data. The data usually has high-dimensional sparsity, and traditional data analysis methods cannot effectively extract potential user-advertising relationship features. Therefore, how to efficiently process sparse data and accurately predict advertising effects has become the core issue in improving advertising placement efficiency and return on investment.

[0003] Current advertising data statistics methods mainly rely on traditional machine learning models and statistical methods. These methods process advertising data through simple feature engineering and matrix factorization techniques. However, traditional methods usually use a fixed rank for matrix factorization, resulting in poor processing effects for sparse data. In addition, the prediction models of advertising effects usually ignore the differences between different advertising data sets and scenarios, lacking adaptability, and thus cannot obtain ideal prediction results in complex advertising data environments.

[0004] With the development of deep learning, especially in the ability to process high-dimensional sparse data, deep neural networks demonstrate powerful feature learning capabilities and can automatically extract meaningful patterns and relationships in data. However, traditional deep learning methods still face challenges in how to effectively process high-dimensional sparse data and how to dynamically adjust the model structure to adapt to different advertising scenarios.

[0005] Therefore, those skilled in the art provide an advertising data statistics method based on deep learning to solve the above-mentioned problems. Summary of the Invention

[0006] Aiming at the deficiencies of the prior art, the present invention provides an advertising data statistics method based on deep learning to solve the problems raised in the above background art.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: An advertising data statistics method based on deep learning, comprising:

[0008] Step 1: Construct a user-advertising interaction matrix for advertising data. The rows of the user-advertising interaction matrix represent different users, the columns represent different advertisements, and each element of the matrix represents the degree of interaction between the user and the advertisement;

[0009] Step 2: Encode the obtained user and advertisement features, and transform the user features and advertisement features into low-dimensional dense vectors. Map the discrete user features and advertisement features through embedding technology so that the user features and advertisement features can be expressed in a low-dimensional space;

[0010] Step 3: Based on the obtained user features and advertisement features, perform low-rank matrix factorization. Represent the user-advertisement interaction matrix as the product of matrices, reduce the dimension of the matrix through factorization, and extract the potential relationship between users and advertisements;

[0011] Step 4: On the basis of low-rank matrix factorization, use regularization technology to constrain the obtained user and advertisement matrices. By adding L1 regularization during the matrix factorization process, make the irrelevant feature terms in the matrix tend to zero and optimize the learning effect of the model;

[0012] Step 5: Process the low-rank matrix using a neural network structure. Perform a non-linear mapping on the user and advertisement embedding vectors extracted from the low-rank matrix through the neural network, capture the non-linear relationship, and output the final predicted value of the advertisement effect;

[0013] Step 6: Introduce the mutual information maximization strategy. By maximizing the mutual information between the feature vector and the advertisement effect, enable the model to learn effective feature representations;

[0014] Step 7: Use an optimization algorithm to update all parameters, including the embedding vectors of users and advertisements and the weight matrix of the neural network. During the training process, optimize through the backpropagation algorithm and the gradient descent method, and minimize the training error using the objective function;

[0015] Step 8: After the model training is completed, use the test data to evaluate the model. Evaluate the accuracy of the model by comparing the prediction results with the real data, and adjust the model parameters when necessary;

[0016] Step 9: Apply the trained model to the real-time optimization of advertisement placement. According to the prediction results of the model, adjust the advertisement display strategy, optimize the target audience, display frequency, and advertisement creativity of the advertisement to maximize the effect of advertisement placement.

[0017] Preferably, in the low-rank matrix factorization in step 3, an adaptive matrix factorization method is used to dynamically adjust the matrix rank r, and the matrix factorization is performed through the following optimization objective function:

[0018]

[0019] where L adapt is the loss function of the adaptive matrix factorization,

[0020] A is the user-advertisement interaction matrix,

[0021] U and V are low-rank embedding matrices for users and advertisements,

[0022] r is the matrix rank, and α is a hyperparameter controlling the matrix complexity,

[0023] λ1 and λ2 are regularization parameters,

[0024] ∥U∥1 is the L1 norm of the user feature matrix U,

[0025] ∥V∥1 is the L1 norm of the advertisement feature matrix V,

[0026] represents the sum of the squares of the elements of the matrix.

[0027] Preferably, in the regularization process of step 4, L1 regularization is processed through the following formula:

[0028]

[0029] where L total represents the comprehensive loss function, λ1 and λ2 are regularization parameters,

[0030] ∥U∥1 is the L1 norm of the user feature matrix U,

[0031] ∥V∥1 is the L1 norm of the advertisement feature matrix V,

[0032] represents the sum of the squares of the elements of the matrix,

[0033] A is the user-advertisement interaction matrix,

[0034] U and V are low-rank embedding matrices for users and advertisements.

[0035] Preferably, in the neural network structure of step 5, a deep neural network is used to process the embedding vectors obtained by low-rank matrix factorization. The network structure includes multiple sets of fully connected layers, and the output of each layer is non-linearly mapped through an activation function to finally output the advertisement effect prediction value y:

[0036] y = f(U, V) = σ(W3·σ(W2·σ(W1·(U·V T ))),

[0037] where W1, W2, and W3 are the weight matrices of the fully connected layers,

[0038] y is the advertisement effect prediction value, U is the user feature matrix,

[0039] V is the advertisement feature matrix, and σ is the activation function,

[0040] (U·V T) is the product of the user feature matrix U and the advertisement feature matrix V obtained by low-rank matrix factorization. Preferably, the mutual information maximization strategy in step 6 is maximized through the following objective function:

[0041] I(X, Y) = H(X) + H(Y) - H(X, Y),

[0042] where I(X, Y) is the mutual information, H(X) is the entropy of the user and advertisement feature matrix X,

[0043] H(Y) is the entropy of the advertisement effect Y, and H(X, Y) is the joint entropy of the user features and the advertisement effect.

[0044] Preferably, the optimization algorithm in step 7 is an optimization algorithm based on gradient descent. During the training process, the parameters are updated. The parameters include the user embedding vector U, the advertisement embedding vector V, and the neural network weight matrices W1, W2, and W3. The gradient calculation formula is:

[0045]

[0046] where A is the user-advertisement interaction matrix, U is the user feature matrix, and V is the advertisement feature matrix,

[0047] represents the direction and step size for optimizing the user feature matrix,

[0048] λ1 and λ2 are regularization parameters, and sign(U) represents the sign function of matrix U,

[0049] sign(V) represents the sign function of matrix V.

[0050] Preferably, the optimization algorithm adopted is the Adam optimizer. The parameters are updated by calculating the gradient and the adaptive learning rate to accelerate the convergence of the model. The specific optimization process is carried out through the following update rules:

[0051]

[0052] where θ t+1 represents the parameter value at the (t + 1)-th iteration, θ t represents the parameter value at the t-th iteration, η is the learning rate, m t is the moving average of the gradient, v t is the moving average of the square of the gradient, and ∈ is a constant.

[0053] Preferably, L2 regularization is adopted during the training process. The L2 regularization term is controlled by the following formula:

[0054]

[0055] Among them, L reg represents the regularization loss of the neural network weight matrix, and λ3 is the regularization parameter. is an element in the neural network weight matrices W1, W2, and W3.

[0056] Preferably, the early stopping method is adopted during the training process to avoid overfitting. The early stopping method monitors the validation set loss. If the validation set loss fails to decrease in consecutive multiple training rounds, the training is stopped in advance to prevent the overfitting of the model.

[0057] Preferably, after the training, the obtained advertisement effect prediction model is applied to the real-time optimization of the advertisement placement strategy. According to the predicted advertisement effect, the strategies of the advertisement display frequency, advertisement creativity, and audience targeting are adjusted to improve the overall effect of the advertisement placement.

[0058] The present invention provides a method for advertising data statistics based on deep learning, which has the following beneficial effects:

[0059] 1. By adopting the adaptive matrix decomposition method, the present invention dynamically adjusts the matrix rank, avoids the limitation of using a fixed rank in the traditional method, effectively extracts the potential user-advertisement relationship features in the high-dimensional sparse advertising data, and obtains higher computational efficiency and more accurate advertisement effect prediction.

[0060] 2. By optimizing the objective function for matrix rank adjustment, the present invention enables the matrix decomposition process to dynamically adjust the rank according to the actual changes of the data, enhances the adaptability of the model to different advertisement data sets and scenarios, and obtains a more flexible and efficient advertising data processing solution.

[0061] 3. By introducing the mutual information maximization strategy, the present invention strengthens the model's ability to capture the deep relationship between the input features and the advertisement effect, realizes more accurate and robust feature learning, and obtains more accurate advertisement effect prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] To enable those skilled in the art to understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0064] The present invention will be described in detail below with reference to the accompanying drawings:

[0065] Example:

[0066] Please refer to the attached Figure 1 , this embodiment of the present invention provides a method for statistically analyzing advertising data based on deep learning, including:

[0067] Step 1: Construct a user-advertisement interaction matrix for advertising data. The rows of the user-advertisement interaction matrix represent different users, the columns represent different advertisements, and each element of the matrix represents the degree of interaction between the user and the advertisement;

[0068] Step 2: Encode the characteristics of the obtained users and advertisements, and transform the user characteristics and advertisement characteristics into low-dimensional dense vectors. Map the discrete user characteristics and advertisement characteristics through embedding technology so that the user characteristics and advertisement characteristics can be expressed in a low-dimensional space;

[0069] Step 3: Based on the obtained user characteristics and advertisement characteristics, perform low-rank matrix factorization, represent the user-advertisement interaction matrix as the product of matrices, reduce the dimension of the matrix through factorization, and extract the potential relationship between users and advertisements;

[0070] Step 4: On the basis of low-rank matrix factorization, use regularization technology to constrain the obtained user and advertisement matrices. By adding L1 regularization during the matrix factorization process, make the irrelevant feature terms in the matrix tend to zero and optimize the learning effect of the model;

[0071] Step 5: Use a neural network structure to process the low-rank matrix. Perform a non-linear mapping on the user and advertisement embedding vectors extracted from the low-rank matrix through the neural network, capture the non-linear relationship, and output the final predicted value of the advertising effect;

[0072] Step 6: Introduce a mutual information maximization strategy. By maximizing the mutual information between the feature vector and the advertising effect, enable the model to learn effective feature representations;

[0073] Step 7: Use an optimization algorithm to update all parameters, including the embedding vectors of users and advertisements and the weight matrix of the neural network. During the training process, optimize through the backpropagation algorithm and the gradient descent method, and minimize the training error using the objective function;

[0074] Step 8: After the model training is completed, use test data to evaluate the model. Evaluate the accuracy of the model by comparing the prediction results with the real data, and adjust the model parameters when necessary;

[0075] Step 9: Apply the trained model to the real-time optimization of advertising placement. According to the prediction results of the model, adjust the advertising display strategy, optimize the target audience, display frequency, and advertising creativity of the advertisement to maximize the advertising placement effect.

[0076] The advantage of Step 1 is that it can provide clear basic data structures for subsequent advertisement effect prediction, laying a foundation for the model to learn the potential relationship between user behavior and advertisements.

[0077] The advantage of Step 2 lies in reducing the complexity of the feature space, avoiding the computational burden brought by high-dimensional data processing, and improving the learning efficiency of the model.

[0078] The advantage of Step 3 is that it can reveal the deep relationship between advertisements and users, while improving the efficiency of data processing and avoiding the computational bottleneck in traditional sparse matrix methods.

[0079] The advantage of Step 4 is to reduce unnecessary features, enhance the generalization ability of the model, make feature selection more accurate, and improve the stability and accuracy of the model.

[0080] The advantage of Step 5 is to automatically discover potential complex patterns in the data through a deep learning model, improve the prediction accuracy of advertisement effects, and enable the model to handle more complex data relationships.

[0081] The advantage of Step 6 is that the model can learn linear relationships and capture deep non-linear information, thereby improving the prediction accuracy of advertisement effects.

[0082] The advantage of Step 7 is to continuously optimize the model according to feedback, ensuring that the training process is efficient and accurate.

[0083] The advantage of Step 8 is to ensure the generalization ability of the model, play a better role in actual applications, and avoid overfitting and underfitting phenomena.

[0084] The advantage of Step 9 is to adjust the strategy in real time according to the advertisement effect, thereby improving the return on investment of advertisement placement and ensuring that advertisers obtain the best results.

[0085] In the low-rank matrix factorization of Step 3, an adaptive matrix factorization method is used to dynamically adjust the matrix rank r, and matrix factorization is performed through the following optimization objective function:

[0086]

[0087] where L adapt is the loss function of the adaptive matrix factorization,

[0088] A is the user-advertisement interaction matrix,

[0089] U and V are the low-rank embedding matrices of users and advertisements,

[0090] r is the matrix rank, α is the hyperparameter controlling the matrix complexity,

[0091] λ1 and λ2 are regularization parameters,

[0092] ∥U∥1 is the L1 norm of the user feature matrix U,

[0093] ∥V∥1 is the L1 norm of the advertisement feature matrix V,

[0094] which represents the sum of the squares of the elements of the matrix.

[0095] Traditional low-rank matrix factorization methods use a fixed rank to factorize the user-advertisement interaction matrix, while the present invention can automatically select the most suitable rank according to different advertisement data sets and scenarios by dynamically adjusting the matrix rank r. This method ensures that the model can adaptively adjust in a variety of data environments, and the complexity and scale of different advertisement data sets can be effectively processed, thereby improving the flexibility and generalization ability of the model.

[0096] By optimizing the objective function, it is ensured that during the matrix factorization process, the optimal rank is dynamically selected according to the actual data situation, avoiding overfitting and over-simplification. A lower matrix rank can reduce the computational complexity, enabling the model to improve the computational efficiency while maintaining high accuracy. In addition, appropriate selection of the matrix rank helps to reduce redundant features and avoid model overfitting, thereby enhancing the stability and accuracy of the model.

[0097] During the matrix factorization process, an L1 regularization term is added to make the irrelevant feature terms in the matrix tend to zero, thereby effectively performing feature selection. The introduction of the regularization term helps to reduce feature redundancy and enhance the generalization ability of the model, making the model more robust when processing new data.

[0098] Low-rank matrix factorization can effectively extract the potential relationship features between users and advertisements. By adaptively adjusting the matrix rank, the model can more accurately capture the complex relationship between users and advertisements, ensuring that more valuable feature information can be extracted from sparse data. This is particularly important for advertising effect prediction because the relationship between advertisements and users is usually very complex, and low-rank factorization can help capture this potential deep relationship.

[0099] During the regularization process of step 4, the L1 regularization is processed through the following formula:

[0100]

[0101] where L total represents the comprehensive loss function, and λ1 and λ2 are regularization parameters,

[0102] ∥U∥1 is the L1 norm of the user feature matrix U,

[0103] ∥V∥1 is the L1 norm of the advertisement feature matrix V,

[0104] Denotes the sum of the squares of the elements of the matrix,

[0105] A is the user-advertisement interaction matrix,

[0106] U and V are the low-rank embedding matrices of users and advertisements.

[0107] The core advantage of L1 regularization is that it can make the irrelevant feature terms in the matrix tend to zero. By applying L1 regularization to the user and advertisement matrices U and V, the model can automatically select the most representative features and remove the irrelevant features, achieving the effect of feature selection.

[0108] The regularization term helps reduce the model complexity and effectively avoid overfitting. Especially when dealing with high-dimensional sparse data, L1 regularization reduces the degrees of freedom of the model by compressing unimportant features, ensuring good performance of the model on both training data and unseen data. This makes the model more generalizable and better able to adapt to changing data in practical applications.

[0109] Since L1 regularization promotes matrix sparsity, it can reduce the redundant part in the calculation and improve the computational efficiency of the model. In a high-dimensional sparse data environment, L1 regularization makes the calculation process more efficient, especially when dealing with large-scale advertisement data, which can significantly reduce the computational complexity.

[0110] In the neural network structure of step 5, a deep neural network is used to process the embedding vectors obtained from low-rank matrix factorization. The network structure includes multiple sets of fully connected layers, and the output of each layer is non-linearly mapped through an activation function, and finally the predicted value y of the advertisement effect is output:

[0111] y = f(U, V) = σ(W3·σ(W2·σ(W1·(U·V T )))),

[0112] where W1, W2, and W3 are the weight matrices of the fully connected layers,

[0113] y is the predicted value of the advertisement effect, U is the user feature matrix,

[0114] V is the advertisement feature matrix, σ is the activation function,

[0115] (U·V T ) is the product of the user feature matrix U and the advertisement feature matrix V obtained from low-rank matrix factorization.

[0116] Through the combination of multiple fully connected layers and activation functions, the neural network can automatically capture the non-linear relationships in the advertising data. In advertising effect prediction, the relationship between users and advertisements is complex and non-linear. The deep neural network can map the features obtained from low-rank matrix factorization to a higher-dimensional space through layer-by-layer non-linear mappings, improving the model's ability to model complex relationships.

[0117] Through the deep neural network, the embedding vectors of the user feature matrix U and the advertisement feature matrix V obtained from low-rank matrix factorization are used to generate richer and more discriminative feature representations. After the features are mapped through non-linear activation functions, the learning of potential advertising effects can be enhanced, improving the accuracy of advertising effect prediction.

[0118] The hierarchical structure of the deep neural network enables the model to flexibly adapt to different datasets and advertising scenarios. The outputs of each layer are processed through activation functions, providing the model with stronger feature extraction capabilities. The non-linear combinations between different layers can help the model automatically discover the most meaningful patterns from the data, adapting to the complexity in various advertising placement scenarios.

[0119] The mutual information maximization strategy in step 6 is maximized through the following objective function:

[0120] I(X, Y) = H(X) + H(Y) - H(X, Y),

[0121] where I(X, Y) is the mutual information, H(X) is the entropy of the user and advertisement feature matrix X,

[0122] H(Y) is the entropy of the advertising effect Y, and H(X, Y) is the joint entropy of the user features and the advertising effect.

[0123] Mutual information maximization helps the model learn more representative and higher-quality feature representations. By maximizing the mutual information between user features and advertising effects, the model can automatically identify and strengthen the features closely related to advertising effects, enabling better extraction of potential information in advertising data, improving the expressive ability of features, and enhancing the prediction accuracy of advertising effects.

[0124] Traditional models usually learn linear relationships. Through the mutual information maximization strategy, the model can capture deeper non-linear associations between user features and advertising effects. This process enables the model to understand the complex interaction relationships between user behaviors and advertisement contents, improving the accuracy of predicting advertising effects.

[0125] Mutual information maximization enhances the robustness of the model by strengthening the dependence relationship between features and advertising effects. When dealing with incomplete and noisy data, the model can optimize feature learning by maximizing mutual information, reducing the interference of irrelevant information, and improving the stability and reliability of advertising effect prediction.

[0126] The optimization algorithm for step 7 is an optimization algorithm based on gradient descent. During the training process, the parameters are updated. The parameters include the user embedding vector U, the advertisement embedding vector V, and the neural network weight matrices W1, W2, and W3. The gradient calculation formula is as follows:

[0127]

[0128] where A is the user-advertisement interaction matrix, U is the user feature matrix, and V is the advertisement feature matrix.

[0129] represents the direction and step size for optimizing the user feature matrix.

[0130] λ1 and λ2 are regularization parameters, and sign(U) represents the sign function of matrix U.

[0131] sign(V) represents the sign function of matrix V.

[0132] The optimization algorithm adopted is the Adam optimizer. The parameters are updated by calculating the gradient and the adaptive learning rate to accelerate the convergence of the model. The specific optimization process is carried out through the following update rules:

[0133]

[0134] where θ t+1 represents the parameter value at the (t + 1)-th iteration, θ t represents the parameter value at the t-th iteration, η is the learning rate, m t is the moving average of the gradient, v t is the moving average of the square of the gradient, and ∈ is a constant.

[0135] The optimization algorithm based on gradient descent calculates the gradient and adjusts the parameters, continuously reducing the loss function, and thus optimizing the model. The gradient calculation formula can effectively guide the update of the model parameters in the direction of minimizing the loss function, improving the accuracy of the model. The Adam optimizer accelerates the training process by dynamically adjusting the learning rate and combining the moving average of the gradient, avoiding getting stuck in local minima in the parameter space, and ensuring the rapid convergence of the model.

[0136] The Adam optimizer has the advantage of an adaptive learning rate and can adjust the update step size of each parameter according to the gradient. Specifically, the algorithm automatically adjusts the learning rate of each parameter according to the historical information of the gradient. When facing different parameter update requirements, it can perform more precise optimization. The adaptive learning rate makes the parameter update more flexible and efficient, especially when facing complex and changing advertisement data, maintaining stable convergence.

[0137] During the training process, the regularization term helps control the complexity of the model and reduce overfitting to the training data. The regularization parameter controls the sparsity of the user feature matrix U and the advertisement feature matrix V, prompting the model to select more meaningful features and improving the generalization ability and stability of the model.

[0138] L2 regularization is adopted during the training process, and the L2 regularization term is controlled by the following formula:

[0139]

[0140] where L reg represents the regularization loss of the neural network weight matrix, λ3 is the regularization parameter, and \(w_{ij}\) are the elements in the neural network weight matrices \(W_1\), \(W_2\), and \(W_3\).

[0141] The early stopping method is adopted during the training process to avoid overfitting. The early stopping method monitors the validation set loss. If the validation set loss fails to decrease in consecutive multiple training epochs, the training is stopped in advance to prevent the model from overfitting.

[0142] After training, the obtained advertisement effect prediction model is applied to the real-time optimization of the advertisement placement strategy. According to the predicted advertisement effect, the strategies of advertisement display frequency, advertisement creativity, and audience targeting are adjusted to improve the overall effect of advertisement placement.

[0143] Through L2 regularization, early stopping, and real-time optimization, the present invention can effectively improve the performance of the advertisement data statistics method. L2 regularization controls the model complexity, prevents overfitting, and improves the generalization ability of the model; the early stopping method monitors the validation set loss, avoids overfitting during the training process, and enhances the stability of the model; real-time optimization enables the model to dynamically adjust the advertisement strategy according to the advertisement effect prediction result, improving the accuracy and effect of advertisement placement. The combination of technical means makes the advertisement effect prediction model perform more efficiently, stably, and accurately during the training, evaluation, and application processes.

[0144] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for advertising data statistics based on deep learning, characterized in that, Including: Step 1: Construct a user-advertisement interaction matrix for advertisement data. The rows of the user-advertisement interaction matrix represent different users, the columns represent different advertisements, and each element of the matrix represents the degree of interaction between a user and an advertisement; Step 2: Encode the obtained features of users and advertisements, and transform the user features and advertisement features into low-dimensional dense vectors. Map the discrete user features and advertisement features through an embedding technique so that the user features and advertisement features can be expressed in a low-dimensional space; Step 3: Based on the obtained user features and advertisement features, perform low-rank matrix factorization. Represent the user-advertisement interaction matrix as a product of matrices, reduce the dimension of the matrix through factorization, and extract the potential relationship between users and advertisements; Step 4: On the basis of low-rank matrix factorization, use a regularization technique to constrain the obtained user and advertisement matrices. By adding L1 regularization during the matrix factorization process, make the irrelevant feature terms in the matrix tend to zero and optimize the learning effect of the model; Step 5: Use a neural network structure to process the low-rank matrix. Perform a non-linear mapping on the user and advertisement embedding vectors extracted from the low-rank matrix through the neural network, capture the non-linear relationship, and output the final predicted value of the advertisement effect; Step 6: Introduce a mutual information maximization strategy. By maximizing the mutual information between the feature vector and the advertisement effect, enable the model to learn effective feature representations; Step 7: Use an optimization algorithm to update all parameters, including the embedding vectors of users and advertisements and the weight matrix of the neural network. During the training process, optimize through the backpropagation algorithm and the gradient descent method, and minimize the training error using the objective function; Step 8: After the model training is completed, use test data to evaluate the model. Evaluate the accuracy of the model by comparing the prediction results with the real data, and adjust the model parameters when necessary; Step 9: Apply the trained model to the real-time optimization of advertisement placement. According to the prediction results of the model, adjust the advertisement display strategy, optimize the target audience, display frequency and advertisement creativity of the advertisement to maximize the effect of advertisement placement.

2. The advertising data statistics method based on deep learning according to claim 1, characterized in that In the low-rank matrix factorization of Step 3, an adaptive matrix factorization method is used to dynamically adjust the matrix rank r, and the matrix factorization is performed through the following optimization objective function: Among them, L adapt is the loss function of adaptive matrix factorization, A is the user-advertisement interaction matrix, U and V are the low-rank embedding matrices of users and advertisements, r is the matrix rank, α is a hyperparameter controlling the matrix complexity, λ1 and λ2 are regularization parameters, ∥U∥1 is the L1 norm of the user feature matrix U, ∥V∥1 is the L1 norm of the advertisement feature matrix V, Represents the sum of the squares of the elements of the matrix.

3. The method for statistically analyzing advertisement data based on deep learning according to claim 2, wherein In the regularization process of Step 4, the L1 regularization is processed through the following formula: Among them, L total represents the comprehensive loss function, and λ1 and λ2 are regularization parameters. ∥U∥1 is the L1 norm of the user feature matrix U, ∥V∥1 is the L1 norm of the advertisement feature matrix V, represents the sum of the squares of the elements of the matrix, A is the user-advertisement interaction matrix, U and V are the low-rank embedding matrices of users and advertisements.

4. A method for statistically analyzing advertising data based on deep learning according to claim 1, characterized in that, In the neural network structure of Step 5, a deep neural network is used to process the embedding vectors obtained from the low-rank matrix factorization. The network structure includes multiple sets of fully connected layers, and the output of each layer is non-linearly mapped through an activation function, and finally the predicted value y of the advertisement effect is output: y = f(U, V) = σ(W3 · σ(W2 · σ(W1 · (U · V T )))), Among them, W1, W2, and W3 are the weight matrices of the fully connected layers, y is the predicted value of the advertising effect, U is the user feature matrix, V is the advertising feature matrix, σ is the activation function, (U·V T ) is the product of the user feature matrix U and the advertisement feature matrix V obtained by low-rank matrix factorization.

5. A method for advertising data statistics based on deep learning according to claim 1, characterized in that, The mutual information maximization strategy in step 6 is maximized through the following objective function: I(X, Y) = H(X) + H(Y) - H(X, Y), where I(X, Y) is the mutual information, H(X) is the entropy of the user and advertising feature matrix X, H(Y) is the entropy of the advertising effect Y, and H(X, Y) is the joint entropy of the user features and the advertising effect.

6. A method for advertising data statistics based on deep learning according to claim 1, characterized in that, The optimization algorithm in step 7 is an optimization algorithm based on gradient descent. During the training process, the parameters are updated. The parameters include the user embedding vector U, the advertising embedding vector V, and the neural network weight matrices W1, W2, and W3. The gradient calculation formula is: where A is the user-advertising interaction matrix, U is the user feature matrix, and V is the advertising feature matrix, Indicates the direction and step size for optimizing the user feature matrix, λ1 and λ2 are regularization parameters, sign(U) represents the sign function of matrix U, sign(V) represents the sign function of matrix V.

7. A method for advertising data statistics based on deep learning according to claim 6, characterized in that, The optimization algorithm adopted is the Adam optimizer. The parameters are updated by calculating the gradient and the adaptive learning rate to accelerate the convergence of the model. The specific optimization process is carried out through the following update rules: Among them, θ t+1 represents the parameter value at the (t + 1)-th iteration, θ t represents the parameter value at the t-th iteration, η is the learning rate, m t is the moving average of the gradient, v t is the moving average of the square of the gradient, and ∈ is a constant.

8. A method for statistically analyzing advertising data based on deep learning according to claim 6, characterized in that L2 regularization is adopted during the training process. The L2 regularization term is controlled by the following formula: Among them, L reg represents the regularization loss of the neural network weight matrix, and λ3 is the regularization parameter. is an element in the neural network weight matrices W1, W2, and W3.

9. A method for advertising data statistics based on deep learning according to claim 8, characterized in that Early stopping is adopted during the training process to avoid overfitting. Early stopping monitors the validation set loss. If the validation set loss fails to decrease in consecutive multiple training rounds, the training is stopped in advance to prevent the model from overfitting.

10. A method for advertising data statistics based on deep learning according to claim 9, characterized in that, After the training, the obtained advertising effect prediction model is applied to the real-time optimization of the advertising placement strategy. According to the predicted advertising effect, the strategies of the display frequency, advertising creativity, and audience targeting of the advertisement are adjusted to improve the overall effect of the advertising placement.

Images