A blockchain-based advertising data classification method and system

Abstract

The present invention provides a method and system for classifying advertising data. The method comprises the following steps: collecting advertising data from multiple sources and preprocessing the collected advertising data; parsing the preprocessed advertising data using a generative artificial intelligence (AI) model and converting the parsed text data into high-dimensional vectors; constructing an asymmetric autoencoder neural network with unshared connection weight matrices for the encoder and decoder, with the encoder using sparse connections and the decoder being a fully connected network; and constructing a loss function to train the asymmetric autoencoder neural network; using the trained asymmetric autoencoder neural network to convert the high-dimensional vectors into low-dimensional feature representations, which are then input into a classifier for classification. Compared to existing technologies, the method of the present invention can achieve more efficient advertising classification.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an advertising data classification method and system based on blockchain and a large model. Background Art

[0002] In the digital advertising industry, targeted advertising delivery relies on the effective classification of advertising data. However, existing ad classification methods still face numerous challenges. First, the sources of advertising data are complex, involving multiple channels such as social media, search engines, and website advertising systems. This makes data authenticity difficult to guarantee and makes it susceptible to data tampering or false advertising. Second, advertising content is diverse, encompassing multiple formats such as text, images, and videos. Traditional machine learning methods have limitations in feature extraction and semantic understanding, making it difficult to accurately capture the implicit information in advertising copy. Especially when processing high-dimensional sparse data, traditional dimensionality reduction methods such as principal component analysis (PCA) and symmetric autoencoder neural networks struggle to strike a balance between feature compression and information preservation, resulting in limited classification effectiveness. Furthermore, current ad classification models generally employ fixed learning rates or traditional gradient descent optimization strategies. Faced with the complex distribution of high-dimensional advertising data, model training is prone to falling into local optima and exhibits slow convergence.

[0003] For example, a Chinese invention patent with publication number CN111191445A proposes a method for classifying advertising text, including: obtaining a text to be classified; calculating the word vector of the text to be classified using a trained word vector model; calculating the similarity of similar words corresponding to a given category based on the word vector of the text to be classified to obtain a similarity score between the text to be classified and each of the given categories; and configuring the given category with the highest similarity score to the text to be classified as the classification result of the text to be classified. This method can automatically classify the text to be classified into the corresponding sub-category with the highest similarity score, thereby achieving a high degree of match between the text classification and the given sub-category of the business, and further achieving accurate classification of the advertising text.

[0004] For another example, the Chinese invention patent with publication number CN113220966A proposes a method for classified display of advertising creatives, including the following steps: obtaining advertising creatives of preset media channels and regions through crawler technology to construct an original database of channel creatives; standardizing the creatives of each channel in the original database of channel creatives to construct a standard database of creatives; parsing the advertising creative data in the standard database of creatives, performing advertising creative back-labeling processing, and obtaining advertising creative data after back-labeling processing; sorting the advertising creative data after back-labeling processing based on preset evaluation criteria to obtain sorted advertising creative data; and indexing the sorted advertising creative data based on the received search instructions and returning the retrieved advertising creative data for classified display of advertising creatives. This method can improve the efficiency of acquiring advertising creatives and enable the recommendation of high-quality creatives.

[0005] However, the prior art including the above method still has the following deficiencies:

[0006] 1) In the advertising data classification task, traditional autoencoder neural networks have limitations. The lack of symmetric structure limits their adaptability to high-dimensional sparse data.

[0007] 2) In the advertising data classification task, the traditional model has insufficient sparsity control capabilities, and traditional L1 / L2 regularization cannot effectively control feature sparsity.

[0008] 3) In the advertising data classification task, the traditional fixed learning rate method is inefficient, cannot adapt to different data distributions, and is prone to falling into local optimality or gradient explosion. Summary of the Invention

[0009] Based on the above background, the present invention proposes an advertisement data classification method and system to overcome at least one of the shortcomings of the above-mentioned prior art. Specifically, the following technical solutions are adopted:

[0010] A first aspect of the present invention provides an advertisement data classification method, comprising the following steps:

[0011] Collect advertising data from various sources and pre-process the collected advertising data;

[0012] Use a generative AI model to parse pre-processed advertising data and convert the parsed text data into high-dimensional vectors;

[0013] Constructing an asymmetric autoencoding neural network in which the connection weight matrices of the encoder and decoder are not shared, the encoder adopts sparse connections, and the decoder is a fully connected asymmetric autoencoding neural network, and constructing a loss function to train the asymmetric autoencoding neural network;

[0014] The high-dimensional vector is converted into a low-dimensional feature representation using a trained asymmetric autoencoder neural network, and then input into a classifier for classification.

[0015] Furthermore, the pre-processing of the collected advertising data includes:

[0016] Data cleaning, including removing invalid data, deduplicating data, and / or filling in missing values;

[0017] Data transformation, including normalization and standardization of data and encoding of categorical variables.

[0018] Furthermore, the use of a generative artificial intelligence model to parse the pre-processed advertising data and convert the parsed text data into a high-dimensional vector includes:

[0019] Use generative AI models to parse ad text and extract the ad's theme, sentiment, target audience, and / or market positioning information;

[0020] Understand the semantics of advertising texts based on natural language processing technology to identify the core intent of advertising content;

[0021] Convert the parsed text data into high-dimensional vectors.

[0022] Furthermore, the asymmetric autoencoder neural network uses learnable latent variables to enhance the expression of complex patterns. The specific calculation method is expressed as follows:

[0023]

[0024] Where, is the feature representation after dimensionality reduction; is the weight matrix, which represents the connection strength between neurons in the asymmetric autoencoder neural network; Input data represents the vectorized advertising copy data; is the weighted coefficient of the interaction term, which represents the strength of the influence of the latent variable; is a latent variable matrix that represents the automatic learning of complex patterns or associated features; is the bias term of the asymmetric autoencoder neural network; is a nonlinear activation function, which represents the nonlinear transformation of the mapping result. Its calculation method is expressed as:

[0025]

[0026] Where, is an exponential function, representing a transformation that exponentially amplifies the input value; is the input signal of the current neuron.

[0027] Furthermore, a loss function is constructed by combining the reconstruction error and the sparsity regularization term to train the asymmetric autoencoder neural network. The constructed loss function expression is as follows:

[0028]

[0029] Where, is the total loss function of the asymmetric autoencoder neural network; is the total number of samples, representing the size of the copywriting samples that can be used for training; For the Input data; Characterization The L2 distance between the input and output of a data sample; is the regularization strength parameter; For the dimensionality reduction features The absolute value of the power, is a positive integer; is the regularization coefficient, which represents the influence of the additional entropy term in the sparsity constraint; Characterizing entropy-based sparsity measures; is the variance of the current batch of data, which represents the degree of dispersion of the distribution of the batch of advertising copy vectors; is the skewness adjustment factor, which represents the correction strength of the distribution skewness; is the sparsity regularization term, which is calculated as follows:

[0030]

[0031] Where, is the number of features after dimensionality reduction, representing the dimension of the final low-dimensional space.

[0032] Furthermore, training the asymmetric autoencoder neural network includes:

[0033] Setting the number of network layers and the number of neurons in each layer of the asymmetric autoencoder neural network, and performing parameter initialization operations;

[0034] During the training process, the loss function is calculated step by step and the asymmetric autoencoder neural network is updated according to its gradient with respect to the weight matrix, bias term and latent variable matrix, and the learning rate is adaptively decayed with reference to the simulated annealing process;

[0035] After each training cycle, a dynamic correction mechanism is used to adjust the parameter update direction;

[0036] Repeat the above steps until the preset stopping conditions are met and the training is completed.

[0037] Furthermore, the parameter initialization operation is performed in the following manner:

[0038]

[0039] Where, Symbols indicating compliance with a specific distribution; The mean is 0 and the variance is The normal distribution of represents the random generation of initial weights; is the identity matrix, representing the basic linear transformation that does not change the direction of the vector.

[0040] Furthermore, during the training process, the weight matrix of the asymmetric autoencoder neural network is updated based on the momentum mechanism and combined with gradient accumulation. The calculation method is expressed as:

[0041]

[0042] Where, is the modified weight update amount, representing the additional correction to the current weight; For the The weight matrix of the iteration; For the The weight matrix of the iteration; is the gradient of the loss function with respect to the weight matrix, representing the update direction of the asymmetric autoencoder neural network due to the advertising copy reconstruction loss and sparse regularization; is the weighted coefficient of the momentum term, which represents the proportion of the impact of momentum on this update; For the The momentum term of the iteration is updated as follows:

[0043]

[0044] Where, is the momentum decay coefficient, which represents the retention ratio of historical gradient; For the The momentum term for the iteration.

[0045] Furthermore, the calculation method for iteratively updating the bias term of the asymmetric autoencoder neural network during training is:

[0046]

[0047] Where, For the The bias vector for the iteration; For the The bias vector for the iteration; is the learning rate of the asymmetric autoencoder neural network; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the bias, which represents the optimization direction of the bias term in the dimensionality reduction objective.

[0048] Furthermore, the calculation method for iteratively updating the latent variable matrix of the asymmetric autoencoder neural network during training is expressed as:

[0049]

[0050] Where, For the The latent variable matrix of the iteration; For the The latent variable matrix of the iteration; is the learning rate of the asymmetric autoencoder neural network, which represents the update step size of the asymmetric autoencoder neural network parameters in each iteration; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the latent variable matrix, representing the optimization direction of the vectorized advertising copy data in the latent space.

[0051] Furthermore, during the training process, the adaptive decay of the learning rate is calculated by referring to the simulated annealing process:

[0052]

[0053] Where, For the The learning rate at the iteration; is the diagonal approximation of the Hessian matrix; is the learning rate adjustment coefficient, which represents the decay rate of the learning rate with the number of iterations.

[0054] Furthermore, the calculation method for adjusting the parameter update direction using the dynamic correction mechanism is:

[0055]

[0056] Where, is a symbolic function, Representation symbolizes the weight direction to enhance nonlinear regulation; is a decay function based on the time step, representing the gradual reduction of the correction strength during the iteration process; is the correction term adjustment coefficient.

[0057] Furthermore, the classifier is any one of logistic regression, support vector machine, random forest or decision tree.

[0058] A second aspect of the present invention provides an advertisement data classification system for implementing the advertisement data classification method described in the first aspect, comprising:

[0059] A data collection module is used to collect advertising data from various sources and pre-process the collected advertising data;

[0060] A data processing module is used to parse the pre-processed advertising data using a generative artificial intelligence model and convert the parsed text data into high-dimensional vectors; and to convert the high-dimensional vectors into low-dimensional feature representations using a trained asymmetric autoencoder neural network, which are then input into a classifier for classification;

[0061] The blockchain evidence storage and verification module is used to store the collected data on the blockchain, ensure the credibility of the model training data, and store the model inference results in a verifiable manner;

[0062] And the system interaction and visualization module is used to provide a user interface, display classification results, and support users to query and adjust model parameters.

[0063] The beneficial technical effects of the present invention are as follows:

[0064] 1) The asymmetric autoencoding neural network adopts an asymmetric coding structure, which enhances the adaptability to high-dimensional advertising data.

[0065] 2) The asymmetric autoencoder neural network combined with the latent variable matrix improves the ability to recognize complex patterns in advertising data.

[0066] 3) The asymmetric autoencoder neural network combined with sparsity-induced regularization and a composite regularization strategy are used to improve the retention rate of key features of advertising data.

[0067] 4) The asymmetric autoencoder neural network adopts adaptive learning rate optimization, combined with local curvature perception and dynamic adjustment strategy, to improve the convergence speed and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Schematic diagram of the flow of an embodiment of the advertising data classification method of the present invention.

[0069] Figure 2 This is a schematic diagram of the results of verification experiment 1 in an embodiment of the present invention.

[0070] Figure 3 This is a schematic diagram of the results of verification experiment 2 in an embodiment of the present invention.

[0071] Figure 4 This is a schematic diagram of the results of verification experiment three in an embodiment of the present invention.

[0072] Figure 5 Schematic diagram of the results of verification experiment 4 in the embodiment of the present invention.

[0073] Figure 6 This is a schematic diagram of the results of verification experiment 5 in the embodiment of the present invention. DETAILED DESCRIPTION

[0074] Embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0075] See also Figure 1 , an embodiment of the present invention provides an advertisement data classification method, comprising the following steps:

[0076] S1. Collect advertising data from multiple sources and pre-process the collected advertising data.

[0077] Specifically, crawler technology and API interfaces can be used to collect advertising data in real time from social media platforms, search engine advertising, website advertising systems and other channels, while blockchain technology can be used to store the data to ensure the traceability and non-tamperability of the data source.

[0078] As a preferred implementation scheme, in this embodiment, pre-processing the collected advertising data includes:

[0079] Data cleaning, including removing invalid data, deduplicating data, and / or filling in missing values;

[0080] Data transformation, including normalization and standardization of data and encoding of categorical variables.

[0081] Furthermore, feature engineering can be used to perform keyword extraction, sentiment analysis, and user behavior feature construction to enhance the availability and effectiveness of data.

[0082] S2. Use a generative AI model to parse the preprocessed advertising data and convert the parsed text data into a high-dimensional vector.

[0083] As a preferred implementation scheme, in this embodiment, the specific process includes the following:

[0084] Use generative AI models to parse ad text and extract the ad's theme, sentiment, target audience, and / or market positioning information;

[0085] Understand the semantics of advertising texts based on natural language processing technology to identify the core intent of advertising content;

[0086] Convert the parsed text data into high-dimensional vectors.

[0087] Since the use of generative artificial intelligence large models to process and convert text data is an existing technical means in this field, it will not be elaborated here.

[0088] S3. Construct a connection weight matrix between the encoder and decoder that is not shared, and the encoder uses sparse connections (for example, there is no connection between some neurons), the decoder is a fully connected asymmetric autoencoder neural network, and a loss function is constructed to train the asymmetric autoencoder neural network.

[0089] In this embodiment, the asymmetric autoencoder neural network is different from the traditional autoencoder neural network (composed of a symmetric encoder and decoder). It is designed to meet the complex distribution modeling requirements of high-dimensional sparse data such as advertising copy vectors. For example, advertising copy data often has high-dimensional sparse characteristics (such as the bag-of-words model). The asymmetric structure can be used to design the encoder in a targeted manner, capturing the potential complex patterns in the data (such as the implicit associations of keyword combinations) through a deeper network, while simplifying the decoder to avoid overfitting. The asymmetric autoencoder neural network breaks the symmetry of the structure, and its asymmetry is reflected in two aspects:

[0090] 1) Asymmetric structure: The encoder and decoder have different numbers of layers, neurons, or connections. In a specific example, the encoder may be deeper (e.g., 5 layers) and the decoder shallower (e.g., 4 layers) to accommodate high-dimensional to low-dimensional compression.

[0091] 2) Parameter asymmetry: The weight matrices of the encoder and decoder are not shared, and the encoder uses additional latent variables (latent variable matrix ), and enhance the ability to express nonlinear relationships.

[0092] As a preferred embodiment, in this embodiment, the sparse connection method is realized by a sparse matrix, and a matrix with the same dimension as the weight matrix of the asymmetric autoencoder neural network is generated in the initialization stage, that is, the sparse matrix , whose elements follow the Bernoulli distribution, and the sparsity matrix is obtained by element-by-element product Some weights are forced to be 0. At the same time, under the constraint of the sparsity regularization term, key features are screened out, and the model can adaptively adjust the sparsity pattern during training.

[0093] As a preferred implementation scheme, in this embodiment, the asymmetric autoencoder neural network uses learnable latent variables to enhance the expression of complex patterns. The specific calculation method is expressed as follows:

[0094]

[0095] Where, is the feature representation after dimensionality reduction; is the weight matrix, which represents the connection strength between neurons in the asymmetric autoencoder neural network; Input data represents the vectorized advertising copy data; is the interaction term weight coefficient, which characterizes the strength of the latent variable influence. Preferably, Set to 0.1; is a latent variable matrix that represents the automatic learning of complex patterns or associated features; is the bias term of the asymmetric autoencoder neural network; is a nonlinear activation function, which represents the nonlinear transformation of the mapping result. Its calculation method is expressed as:

[0096]

[0097] Where, is an exponential function, representing a transformation that exponentially amplifies the input value; is the input signal of the current neuron.

[0098] In this embodiment, the latent variable matrix is inspired by the latent variable concept in the probabilistic graphical model and is used to model the potential associations in the data. As a learnable low-rank matrix, it captures the implicit relationship between advertising copy (such as the correlation between "price" and "promotion") through iterative updates, and directly integrates latent variables into the forward calculation of the autoencoder as optimizable parameters (rather than random variables), simplifying the training process.

[0099] In this embodiment, in order to fully model the nonlinear relationship in the advertising copy vector, a nonlinear activation function based on the principle of material expansion in physics is adopted, so that the activation function can capture the irregular characteristic distribution of the advertising copy data while maintaining the smoothness of the output, thereby enhancing the modeling ability of the long-tail distribution (dominated by a few keywords) in the advertising copy.

[0100] Furthermore, a loss function is constructed by combining the reconstruction error and the sparsity regularization term to train the asymmetric autoencoder neural network. The constructed loss function expression is as follows:

[0101]

[0102] Where, is the total loss function of the asymmetric autoencoder neural network; is the total number of samples, representing the size of the copywriting samples that can be used for training; For the Input data; Characterization The L2 distance between the input and output of a data sample; is the regularization strength parameter, preferably, Set to 0.3; For the dimensionality reduction features The absolute value of the power, preferably, Set to 2, is a positive integer; is the regularization coefficient, which characterizes the influence of the additional entropy term in the sparsity constraint. Preferably, Set to 0.5; Characterizing entropy-based sparsity measures; is the variance of the current batch of data, which represents the degree of dispersion of the distribution of the batch of advertising copy vectors; is the skewness adjustment factor, which characterizes the correction strength of the distribution skewness. Preferably, Set to 2, pass To dynamically adjust the loss weights under different data distributions; is a sparsity regularization term, which is used to filter out the most discriminative features for advertising copy classification or key descriptions, thereby suppressing redundant and noisy features and improving the generalization performance and interpretability of the dimensionality reduction model. The calculation method is expressed as:

[0103]

[0104] Where, is the number of features after dimensionality reduction, representing the dimension of the final low-dimensional space.

[0105] As a preferred embodiment, in this embodiment, training the asymmetric autoencoder neural network includes:

[0106] Setting the number of network layers and the number of neurons in each layer of the asymmetric autoencoder neural network, and performing parameter initialization operations;

[0107] During the training process, the loss function is calculated step by step and the asymmetric autoencoder neural network is updated according to its gradient with respect to the weight matrix, bias term and latent variable matrix, and the learning rate is adaptively decayed with reference to the simulated annealing process;

[0108] After each training cycle, a dynamic correction mechanism is used to adjust the parameter update direction;

[0109] Repeat the above steps until the preset stopping conditions are met and the training is completed.

[0110] Furthermore, the parameter initialization operation is performed in the following manner:

[0111]

[0112] Where, Symbols indicating compliance with a specific distribution; The mean is 0 and the variance is The normal distribution of represents the random generation of initial weights; is the identity matrix, representing the basic linear transformation that does not change the direction of the vector.

[0113] Furthermore, during the training process, the weight matrix of the asymmetric autoencoder neural network is updated based on the momentum mechanism and combined with gradient accumulation. The calculation method is expressed as:

[0114]

[0115] Where, is the modified weight update amount, representing the additional correction to the current weight; For the The weight matrix of the iteration; For the The weight matrix of the iteration; is the gradient of the loss function with respect to the weight matrix, representing the update direction of the asymmetric autoencoder neural network due to the advertising copy reconstruction loss and sparse regularization; is the weighted coefficient of the momentum term, which represents the impact ratio of momentum on this update. Preferably, Set to 0.4; For the The momentum term of the iteration is updated as follows:

[0116]

[0117] Where, is the momentum attenuation coefficient, which characterizes the retention ratio of the historical gradient. Preferably, Set to 0.3; For the The momentum term for the iteration.

[0118] Furthermore, the calculation method for iteratively updating the bias term of the asymmetric autoencoder neural network during training is:

[0119]

[0120] Where, For the The bias vector for the iteration; For the The bias vector for the iteration; is the learning rate of the asymmetric autoencoder neural network; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the bias, which represents the optimization direction of the bias term in the dimensionality reduction objective.

[0121] Furthermore, the calculation method for iteratively updating the latent variable matrix of the asymmetric autoencoder neural network during training is expressed as:

[0122]

[0123] Where, For the The latent variable matrix of the iteration; For the The latent variable matrix of the iteration; is the learning rate of the asymmetric autoencoder neural network, which represents the update step size of the asymmetric autoencoder neural network parameters for each iteration. Preferably, Set to 0.01; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the latent variable matrix, representing the optimization direction of the vectorized advertising copy data in the latent space.

[0124] Furthermore, during the training process, the adaptive decay of the learning rate is calculated by referring to the simulated annealing process:

[0125]

[0126] Where, For the The learning rate at the iteration; is the diagonal approximation of the Hessian matrix; is the learning rate adjustment coefficient, which represents the decay rate of the learning rate with the number of iterations.

[0127] Furthermore, after each cycle of asymmetric autoencoder neural network training, a dynamic correction mechanism is used to adjust the parameter update direction. By combining the current gradient and historical error information, the update path of the high-dimensional copy feature space is precisely controlled to prevent falling into local optimality and reduce oscillation. The calculation method is expressed as:

[0128]

[0129] Where, is a symbolic function, Representation symbolizes the weight direction to enhance nonlinear regulation; is a decay function based on the time step, representing the gradual reduction of the correction strength during the iteration process; is the correction term adjustment coefficient.

[0130] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0131] S4. Utilize the trained asymmetric autoencoder neural network to convert the high-dimensional vector into a low-dimensional feature representation, and then input it into a classifier for classification.

[0132] Optionally, in this embodiment, the classifier is any one of logistic regression, support vector machine, random forest or decision tree. Since using a classifier to classify low-dimensional features is a conventional technical means in this field, it will not be elaborated in detail here.

[0133] The following further illustrates the technical advantages of the technical solution disclosed in the present invention (referred to as the present technology in the figure) through several experimental data.

[0134] Experiment 1

[0135] This experiment aims to verify the effect of asymmetric structure design on the ability to select advertising copy features, and compares the performance of conventional dimensionality reduction methods such as principal component analysis, traditional symmetric autoencoders, and manifold learning algorithms with the solution of this invention in terms of feature importance discrimination. Figure 2 Experimental results show that the method of the present invention is significantly better than other methods in distinguishing deep features such as semantic coherence, especially in capturing the implicit associations of keyword combinations in advertising copy, showing stronger pattern recognition capabilities. The asymmetric structure of the present invention effectively enhances the adaptability to high-dimensional sparse data distribution through the combination of encoder deep modeling and decoder simplified design, so that the feature space after dimensionality reduction can more accurately reflect the core elements of advertising semantics. Traditional methods have difficulty balancing feature compression and information retention due to the limitations of symmetrical structures.

[0136] Experiment 2

[0137] This experiment focuses on comparing the effects of sparsity-induced regularization and traditional sparsity constraint methods on feature selection, and comparing the retention of key features under different regularization strengths. Figure 3 Experimental results show that the composite regularization strategy proposed by the method of this invention significantly improves the retention rate of high-value information such as core keywords and sentiment intensity while suppressing noise features. The single sparsity constraint adopted by traditional methods easily leads to excessive feature clipping. However, this invention achieves the coordinated control of feature sparsity and distribution irregularity through entropy constraints and dynamic weighting mechanisms. This allows rare keywords with long-tail distribution in advertising copy to be effectively expressed during the dimensionality reduction process, enhancing the practicality of the model in real business scenarios.

[0138] Experiment 3

[0139] This experiment compares the convergence process of the adaptive learning rate strategy of the present invention with that of the traditional fixed learning rate training to verify the role of local curvature perception and dynamic adjustment mechanism in improving training efficiency. Figure 4 Experimental data shows that the method of the present invention outperforms traditional methods in terms of loss reduction speed and stability. In particular, it uses curvature perception to quickly avoid high-gradient oscillation areas in the early stages of training, and combines learning rate decay to achieve smooth convergence in the middle and late stages. Traditional methods lack adaptability to the distribution characteristics of high-dimensional data and are prone to falling into local optimality or gradient explosion. However, the present method uses a simulated annealing parameter update strategy, which enables the model to autonomously adjust the optimization path based on the geometric characteristics of the feature space, significantly shortening the training cycle required to achieve stable convergence.

[0140] Experiment 4

[0141] This experiment uses heatmap visualization to compare the distribution differences of reconstruction errors of different methods in high-dimensional space to verify the effect of the dynamic correction mechanism on enhancing the generalization ability of the model. Figure 5 Experimental data show that the reconstruction error of the proposed method is more uniform in spatial distribution and lower overall, indicating that it can better maintain the topological structure of the original data. Traditional methods lack a correction mechanism for the parameter update direction, resulting in abnormal peaks in the reconstruction error in specific dimensions. However, the proposed method effectively suppresses directional deviations during the parameter update process through the dynamic fusion of gradient sign correction and historical error feedback, enabling the model to steadily optimize along the optimal path of the loss surface, thereby achieving more accurate low-dimensional mapping on complex high-dimensional advertising data.

[0142] Experiment 5

[0143] This experiment analyzes the effect of dynamic correction mechanism on the efficiency of high-dimensional parameter space exploration by visually comparing the spatial characteristics of parameter optimization paths. Figure 6 Compared with the disordered oscillations and path wandering exhibited by traditional gradient descent methods in parameter space, the optimization trajectory of the method of the present invention shows obvious direction correction and path focusing characteristics, especially in areas where the curvature changes drastically, it can autonomously adjust the update step size and direction. The experimental results show that the parameter update trajectory of the present invention stably approaches the optimal area in the form of spiral convergence in three-dimensional space, while the traditional method has a large number of redundant exploration paths in its trajectory due to the lack of perception of the geometric characteristics of the parameter space. The experimental results show that the dynamic correction mechanism, by fusing historical gradient information and curvature perception, gives the model the ability to autonomously plan the optimal path in a complex parameter space, effectively avoiding the convergence delay problem caused by direction offset in traditional methods.

[0144] A second embodiment of the present invention further provides an advertisement data classification system for implementing the advertisement data classification method described in the first embodiment, including:

[0145] A data collection module is used to collect advertising data from various sources and pre-process the collected advertising data;

[0146] A data processing module is used to parse the pre-processed advertising data using a generative artificial intelligence model and convert the parsed text data into high-dimensional vectors; and to convert the high-dimensional vectors into low-dimensional feature representations using a trained asymmetric autoencoder neural network, which are then input into a classifier for classification;

[0147] The blockchain evidence storage and verification module is used to store the collected data on the blockchain, ensure the credibility of the model training data, and store the model inference results in a verifiable manner;

[0148] And the system interaction and visualization module is used to provide a user interface, display classification results, and support users to query and adjust model parameters.

[0149] Specifically, the functions of the blockchain evidence storage and verification module include data traceability, operation record storage and result verification. Among them, data traceability makes data traceable by storing data fingerprints on the chain; operation record storage is used to record all key steps such as data collection, preprocessing, large model analysis, machine learning modeling, etc. to ensure the reliability of system operation; result verification ensures the integrity and consistency of reasoning results through the blockchain consensus mechanism.

[0150] The functions of the system interaction and visualization module include data query, result display and model tuning interface. Among them, data query supports users to retrieve historical data based on blockchain; result display presents advertising classification results and credibility assessment in a visual manner; model tuning interface allows users to adjust hyperparameters, update training data, and observe changes in model performance.

[0151] It should be noted that the method of the embodiment of the present invention can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiment of the present invention, and the multiple devices will interact with each other to complete the method.

[0152] The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the scope of protection of the present invention.

Claims

1. A blockchain-based advertising data classification method, characterized in that: The steps include: Collect advertising data from various sources and pre-process the collected advertising data; Use a generative AI model to parse pre-processed advertising data and convert the parsed text data into high-dimensional vectors; The connection weight matrices of the encoder and decoder are not shared, the encoder adopts sparse connection, and the decoder is a fully connected asymmetric autoencoder neural network. A loss function is constructed to train the asymmetric autoencoder neural network. The asymmetric autoencoder neural network adopts learnable latent variables to enhance the expression of complex patterns. The specific calculation method is expressed as: Where, is the feature representation after dimensionality reduction; is the weight matrix, which represents the connection strength between neurons in the asymmetric autoencoder neural network; Input data represents the vectorized advertising copy data; is the weighted coefficient of the interaction term, which represents the strength of the influence of the latent variable; is a latent variable matrix that represents the automatic learning of complex patterns or associated features; is the bias term of the asymmetric autoencoder neural network; is a nonlinear activation function, which represents the nonlinear transformation of the mapping result. Its calculation method is expressed as: Where, is an exponential function, representing a transformation that exponentially amplifies the input value; is the input signal of the current neuron; The high-dimensional vector is converted into a low-dimensional feature representation using a trained asymmetric autoencoder neural network, and then input into a classifier for classification.

2. The advertising data classification method according to claim 1, wherein: The pre-processing of the collected advertising data includes: Data cleaning, including removing invalid data, deduplicating data, and / or filling in missing values; Data transformation, including normalization and standardization of data and encoding of categorical variables.

3. The advertising data classification method according to claim 1, wherein: The method of using a generative artificial intelligence model to parse the pre-processed advertising data and converting the parsed text data into a high-dimensional vector includes: Use generative AI models to parse ad text and extract the ad's theme, sentiment, target audience, and / or market positioning information; Understand the semantics of advertising texts based on natural language processing technology to identify the core intent of advertising content; Convert the parsed text data into high-dimensional vectors.

4. The advertising data classification method according to claim 1, wherein: The loss function is constructed by combining the reconstruction error and the sparsity regularization term to train the asymmetric autoencoder neural network. The constructed loss function expression is as follows: Where, is the total loss function of the asymmetric autoencoder neural network; is the total number of samples, representing the size of the copywriting samples that can be used for training; For the Input data; Characterization The L2 distance between the input and output of a data sample; is the regularization strength parameter; For the dimensionality reduction features The absolute value of the power, is a positive integer; is the regularization coefficient, which represents the influence of the additional entropy term in the sparsity constraint; Characterizing entropy-based sparsity measures; is the variance of the current batch of data, which represents the degree of dispersion of the distribution of the batch of advertising copy vectors; is the skewness adjustment factor, which represents the correction strength of the distribution skewness; is the sparsity regularization term, which is calculated as follows: Where, is the number of features after dimensionality reduction, representing the dimension of the final low-dimensional space.

5. The advertisement data classification method according to claim 4, wherein: Training the asymmetric autoencoder neural network includes: Setting the number of network layers and the number of neurons in each layer of the asymmetric autoencoder neural network, and performing parameter initialization operations; During the training process, the loss function is calculated step by step and the asymmetric autoencoder neural network is updated according to its gradient with respect to the weight matrix, bias term and latent variable matrix, and the learning rate is adaptively decayed with reference to the simulated annealing process; After each training cycle, a dynamic correction mechanism is used to adjust the parameter update direction; Repeat the above steps until the preset stopping conditions are met and the training is completed.

6. The advertisement data classification method according to claim 5, wherein: The parameter initialization operation is performed in the following manner: Where, Symbols indicating compliance with a specific distribution; The mean is 0 and the variance is The normal distribution of represents the random generation of initial weights; is the identity matrix, representing the basic linear transformation that does not change the direction of the vector.

7. The advertisement data classification method according to claim 5, wherein: During the training process, the weight matrix of the asymmetric autoencoder neural network is updated based on the momentum mechanism and combined with gradient accumulation. The calculation method is expressed as: Where, is the modified weight update amount, representing the additional correction to the current weight; For the The weight matrix of the iteration; For the The weight matrix of the iteration; is the gradient of the loss function with respect to the weight matrix, representing the update direction of the asymmetric autoencoder neural network due to the advertising copy reconstruction loss and sparse regularization; is the weighted coefficient of the momentum term, which represents the proportion of the impact of momentum on this update; For the The momentum term of the iteration is updated as follows: Where, is the momentum decay coefficient, which represents the retention ratio of the historical gradient; For the The momentum term for the iteration.

8. The advertisement data classification method according to claim 5, wherein: The calculation method for iteratively updating the bias term of the asymmetric autoencoder neural network during training is: Where, For the The bias vector for the iteration; For the The bias vector for the iteration; is the learning rate of the asymmetric autoencoder neural network; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the bias, which represents the optimization direction of the bias term in the dimensionality reduction objective.

9. The advertisement data classification method according to claim 5, wherein: The calculation method for iteratively updating the latent variable matrix of the asymmetric autoencoder neural network during training is expressed as: Where, For the The latent variable matrix of the iteration; For the The latent variable matrix of the iteration; is the learning rate of the asymmetric autoencoder neural network, which represents the update step size of the asymmetric autoencoder neural network parameters in each iteration; It is the gradient of the loss function of the asymmetric autoencoder neural network with respect to the latent variable matrix, representing the optimization direction of the vectorized advertising copy data in the latent space.

10. The advertisement data classification method according to claim 5, wherein: The calculation method for adaptively decaying the learning rate during training is based on the simulated annealing process: Where, For the The learning rate at the iteration; is the diagonal approximation of the Hessian matrix; is the learning rate adjustment coefficient, which represents the decay rate of the learning rate with the number of iterations.

11. The advertisement data classification method according to claim 7, wherein: The calculation method for adjusting the parameter update direction using the dynamic correction mechanism is: Where, is a symbolic function, Representation symbolizes the weight direction to enhance nonlinear regulation; is a decay function based on the time step, representing the gradual reduction of the correction strength during the iteration process; is the correction term adjustment coefficient.

12. The advertisement data classification method according to claim 1, wherein: The classifier is any one of logistic regression, support vector machine, random forest or decision tree.

13. A blockchain-based advertising data classification system, used to implement the advertising data classification method according to any one of claims 1 to 12, characterized in that: include: A data collection module is used to collect advertising data from various sources and pre-process the collected advertising data; A data processing module is used to parse the pre-processed advertising data using a generative artificial intelligence model and convert the parsed text data into high-dimensional vectors; and to convert the high-dimensional vectors into low-dimensional feature representations using a trained asymmetric autoencoder neural network, which are then input into a classifier for classification; The blockchain evidence storage and verification module is used to store the collected data on the blockchain, ensure the credibility of the model training data, and store the model inference results in a verifiable manner; And the system interaction and visualization module is used to provide a user interface, display classification results, and support users to query and adjust model parameters.

Images