Artificial intelligence advertisement publishing method

Abstract

The invention relates to the technical field of information, in particular to an artificial intelligence advertisement publishing method, which comprises the following steps: collecting and preprocessing advertisement data, so that the data is more standardized and consistent, and the quality of the data is effectively improved; by establishing the sentiment analysis model, the sentiment polarity of the input vector can be accurately estimated, the individuation degree of the advertisement is improved, and meanwhile, a sentiment basis is provided for individualized advertisement creativity; by training the sentiment analysis model and updating model parameters by using a gradient descent method, the performance of the model can be continuously optimized, and the accuracy of sentiment classification is improved; by obtaining the user feedback data and calculating the comprehensive user feedback index, the effect of the personalized advertisement creativity is evaluated, targeted adjustment and optimization are facilitated, and the overall effect of the advertisement and the user experience are improved. The invention is used for solving the technical problem of poor advertisement release effect in the existing scheme.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an artificial intelligence advertising publishing method. Background Art

[0002] Information technology is a general term for a class of technologies that can expand the functions of human information organs, including software, hardware, network and communication technologies, and is an indispensable infrastructure in all fields of modern society. Advertising belongs to the field of information technology, and the publication and management of advertising cannot be separated from the Internet and big data technology.

[0003] There are still some problems in the current advertising publishing and management, including the following aspects: data quality and consistency issues. In the current advertising publishing and management, due to the diversity of data sources, there are problems with diverse formats and uneven quality; insufficient personalization of advertisements. Traditional advertising publishing methods often lack targeting and find it difficult to accurately match user interests and needs; lack of specialized sentiment analysis. Current technologies often lack in-depth analysis of the emotional tendencies of advertising content, resulting in a lack of personalization in advertising creativity and difficulty in generating emotional resonance with the target audience; it is difficult to evaluate the effectiveness of advertising creativity. Traditional advertising publishing methods lack a scientific evaluation mechanism and are unable to accurately measure the effectiveness of advertising creativity. In response to these problems, corresponding solutions and technical means need to be adopted to improve the efficiency and effectiveness of advertising publishing. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the background technology and to propose an artificial intelligence advertising publishing method.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An artificial intelligence advertising publishing method, comprising:

[0007] Step 1: Collect advertisement-related data, including text, image, and video data, and pre-process the collected data;

[0008] Step 2: Obtain preprocessed data and establish a sentiment analysis model based on the preprocessed data;

[0009] Step 3: Use the loss function to train the established sentiment analysis model and use the gradient descent method to update the model parameters;

[0010] Step 4: Obtain user feedback data and evaluate the effectiveness of personalized advertising creativity in order to publish the next round of advertising.

[0011] It should be noted that the artificial intelligence advertising publishing method proposed in the present invention can be widely applied to various advertising delivery scenarios. The method can be used to improve the efficiency and effectiveness of advertising delivery. Specifically, it can comprehensively analyze and accurately evaluate advertising data through artificial intelligence technology, and ensure accurate advertising delivery and promote the optimization of advertising effects by collecting advertising-related data, establishing sentiment analysis models, training models, and obtaining user feedback. The application of the present invention can significantly improve the pertinence and effectiveness of advertising delivery, reduce advertising costs, and improve user experience and satisfaction. Through intelligent means, the full-chain optimization of advertising delivery from data collection, model building to creative generation and effect evaluation is achieved.

[0012] Furthermore, the process of collecting advertisement-related data and preprocessing the collected data includes:

[0013] Collect text, image and video data to form the original data set D = {T, I, V}; where T, I, V represent text, image and video data respectively. The original data set is text, image and video data from different channels, including social media platforms, online forums, blogs, etc.

[0014] Preprocess the original data set: For text data T, define the jth word in text data T as t j ; The word t' after natural language processing j Represented as t' j =f T (t j ), where f T Indicates the word t j The function for processing; then the processed text data T'={t' j t j ∈T};

[0015] It can be understood that natural language processing operations include lexical analysis, syntactic analysis, semantic analysis, etc.;

[0016] For image data I, the pixel coordinates in image data I are defined as (a, b), and its original pixel value is P(a, b); the pixel value P'(a, b) after size normalization and color correction is expressed as P'(a, b) = g I (P(a,b)), where g I represents the function of processing image pixels; then the processed image data I'={P'(a,b)|(a,b)∈I};

[0017] It can be understood that size normalization refers to adjusting the size of an image to a uniform standard size. Size normalization is achieved by scaling the image, such as nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. Color correction refers to adjusting the color balance and tone of an image to make it more accurate and consistent. In image processing, color correction involves adjusting the three color channels of RGB (red, green, and blue);

[0018] For video data V, define the kth frame in video data V as v k ; Frame v' after frame rate adjustment and resolution adjustment k =c V (v k ), where c V represents the function of processing the video frame; then the processed video data V'={v' k |v k ∈V};

[0019] It can be understood that frame rate adjustment refers to changing the frame rate of the video to adapt to different application scenarios or requirements. The specific method involves reading the video file, adjusting the frame rate parameters, and then re-encoding and saving the video; resolution adjustment refers to changing the resolution of the image to adapt to different display devices or requirements by using a machine learning framework to specify the resolution parameters in the model input layer or data loading function;

[0020] Integrate the preprocessed data set D'={T',I',V'}.

[0021] Furthermore, the process of obtaining the preprocessed data and establishing a sentiment analysis model based on the preprocessed data includes:

[0022] Get the preprocessed data set and mark it as D'; each data has been converted into a vector representation of fixed length, that is, x∈D', and the input vector x contains information of multiple modalities such as text, image or video;

[0023] A sentiment analysis model EAM(x; Θ) is established to estimate the sentiment polarity of the input vector x. The model establishment includes the setting of feature embedding layer, multi-layer perceptron and output layer; where Θ represents all parameters of the model.

[0024] Furthermore, the process of setting the feature embedding layer, multi-layer perceptron and output layer of the sentiment analysis model includes:

[0025] For the feature embedding layer of the sentiment analysis model, in order to capture more complex semantics and representation capabilities, the input vector x is mapped to a high-dimensional space through the embedding layer. Suppose the embedding layer consists of the weight matrix W emb ∈ R m×nParameterization, where m represents the size of the hidden layer, n represents the size of the input vector, and R does not directly represent a specific meaning or variable, but is used to represent the number of linearly independent rows or columns in the matrix;

[0026] Through the feature embedding layer formula z = W emb x+ ε emb Realize the transformation from low-level representation to high-level abstract feature space. In the formula, z represents the embedded representation of the input vector x after linear transformation, that is, the embedded vector. The embedded vector z transforms the original input vector x (which may be the original word vector, image pixel or audio spectrum, etc.) into a high-dimensional space. This new representation contains more semantic information and intrinsic associations, which is more conducive to the subsequent model to capture and understand the deep features of the data; the embedding layer weight matrix W emb The role of is to reshape and project the input vector x, which is equivalent to a linear combination of the data, reflecting the correlation between the input features and each dimension in the embedding space; ε emb Represents the bias vector, which belongs to the model parameter Θ and is used to ensure that z has a certain non-zero basis even when the input vector x is a zero vector, thus enhancing the flexibility of the model;

[0027] It is understandable that the embedding vector z obtained by the feature embedding layer formula is more suitable for downstream tasks such as sentiment analysis, because it integrates the semantic information of the input vector x to a certain extent, so that it can be more effectively understood in the subsequent nonlinear layer.

[0028] The multi-layer perceptron part of the sentiment analysis model includes several hidden layers, each of which increases nonlinearity by applying an activation function σ; the output of each layer is passed to the next layer as input until the last layer produces the final ad sentiment prediction;

[0029] Let L be the number of layers, z l is the output of the lth layer, W l , ε l are the weight matrix and bias vector of the layer respectively, then there is the formula z l =σ(W l z l-1 +ε l );

[0030] For the output layer of the sentiment analysis model, the output layer is responsible for mapping the output of the hidden layer to the sentiment category;

[0031] Set the output to three categories, including positive, neutral and negative emotions. The output layer has a softmax layer to generate the probability of each category.

[0032] Define Wout , ε out are the weight and bias parameters of the output layer, which are also part of Θ, the parameter of the sentiment analysis model;

[0033] Using the formula y=softmax(W out z L +ε out ) Output the predicted advertising sentiment category probability distribution y, where z L Represents the last layer of the multi-layer perceptron, that is, the output of the Lth layer;

[0034] Understandably, L It has the function of summarizing all the features extracted by the previous layers. Its essence is the composite result of multiple layers of nonlinear transformations, which can capture the high-level abstract representation of the input data. For sentiment analysis tasks, z L Contains the key information that the model considers most relevant to sentiment classification. This information is extracted and condensed step by step from the initial input vector x, and has undergone multiple feature learning and pattern recognition processes.

[0035] Among them, the softmax function is defined as:

[0036] In the formula, softmax(z) i Represents the probability of sentiment category i; is an exponential function, representing a specific category, used to convert the output z of the multilayer perceptron l Mapping into the probability domain ensures that all categories have a non-negative probability value; Indicates all categories The sum of; i represents the sentiment category index, when i = 1, then z1 Indicates positive emotion. When i=2, z2 Indicates neutral sentiment. When i=3, then z3 Represents negative sentiment; J is the sum index variable, traversing all possible output categories;

[0037] It is understandable that the output layer will then convert z L Mapped to a specific emotion category, a probability distribution y is generated by applying the softmax function, and each element y in y i It represents the predicted probability of the corresponding emotional category. In other words, the vector form before the output layer carries sufficient emotional motivation information, and the softmax function converts it into an easy-to-interpret category probability form, which facilitates the model to make the final emotional classification decision.

[0038] Furthermore, the process of training the established sentiment analysis model using the loss function and updating the model parameters using the gradient descent method includes:

[0039] Set the loss function to train the model, where the loss function is:

[0040] Lo(x,τ;Θ)=-Σ i τ i log(y i ), where τ represents the real sentiment label of the advertisement, τi Represents the true label corresponding to the emotion category i, corresponding to the output layer z i The value of y i represents the predicted probability of advertising sentiment category i;

[0041] Use the gradient descent method to update the model parameters Θ, that is,

[0042] In the formula, h represents the number of iterations, η represents the learning rate, represents the gradient with respect to the parameter Θ.

[0043] Furthermore, the process of obtaining user feedback data and evaluating the effectiveness of personalized advertising creativity includes:

[0044] Get the preset advertising creative template M;

[0045] It can be understood that an advertising creative template is a predefined advertising creative structure used to guide the advertising creative generation process, covering specific text format, image layout, color matching, etc.;

[0046] Introduce advertising creative generation function G;

[0047] The advertising creative generation function G accepts the advertising creative template M, the predicted emotion category probability distribution y and the corresponding advertising real emotion label τ as input, and outputs personalized advertising creative s = G(M, y, τ,);

[0048] It is understandable that this is a multi-level mapping process. Function mapping refers to mapping elements in the input space to elements in the output space through certain rules or relationships. The input space consists of creative templates, predicted probability distribution of emotional categories and corresponding real emotional labels, while the output space is personalized advertising creativity. The creative generation function needs to find or generate content that matches these emotions in the creative template based on the real emotional labels, which involves parsing emotional labels, replacing or adjusting variable elements in the template, etc.

[0049] Get user feedback data U = {(u 1 ,r 1 ),(u 2 ,r 2 ),...,(u H ,r H)}, where u represents the user's operation on the personalized advertising creative s (such as clicks, viewing time, etc.), r represents the corresponding evaluation score (such as 0-10 points), and H represents the number of types of operations performed by the user on the personalized advertising creative (such as clicks, viewing time, sharing, comments, etc.);

[0050] Combine the user's various operations on personalized advertising creatives and their corresponding evaluation scores to construct a comprehensive user feedback index: In the formula, F(q) represents the comprehensive user feedback index value, q represents the index of the type of operation performed by the user on the personalized advertising creativity, and w q represents the weight of the qth operation, which is used to reflect the importance of the operation to the overall feedback;

[0051] Use the creative evaluation function to calculate the effectiveness value of personalized advertising creative s:

[0052] In the formula, E(s) represents the effectiveness value of personalized advertising creative s, N s An index collection representing user feedback data related to personalized advertising creative s;

[0053] According to the effectiveness value of personalized advertising creative s, the creative generation function G and creative template M are adjusted to ensure smooth progress to the next round of advertising release: according to user feedback and the evaluation results of personalized advertising creative effectiveness value, the layout of the creative template is optimized, for example, the layout of the advertising content is optimized, the matching of pictures and texts is adjusted, etc., to improve the attractiveness and readability of the advertisement; the design of the creative template is improved, such as changing the color scheme, adjusting the font style, etc., to better match the aesthetic preferences of the target audience; at the same time, according to the evaluation results of personalized advertising creative effectiveness value, the elements with poor performance in the creative template are replaced, such as replacing pictures, modifying copy, etc., to improve the overall effect of the advertisement; after adjusting the creative generation function and creative template, it is necessary to use user feedback data again to evaluate the effectiveness value of personalized advertising creative to verify the effect of the adjustment, which can be achieved by comparing the effectiveness values ​​of the advertising creative before and after the adjustment. If the effectiveness value after the adjustment is improved, it means that the adjustment is effective; if the effectiveness value of the personalized advertising creative does not change significantly or even decreases, it is necessary to further analyze the reasons and make adjustments.

[0054] Compared with the existing technology, the advantages of the artificial intelligence advertising publishing method provided by the present invention are:

[0055] 1. The present invention can significantly improve the quality and consistency of data through preprocessing, namely natural language processing of text, size normalization and color correction of images, frame rate adjustment and resolution adjustment of videos, and provide a reliable basis for subsequent analysis. By integrating data from different channels, advertising analysis can cover a wider range of audience groups and points of interest, which helps to understand market demand more comprehensively.

[0056] 2. The present invention can accurately capture the emotional tendency in the advertising content by establishing a sentiment analysis model, provide an emotional basis for personalized advertising creativity, and enhance the attractiveness and resonance of the advertisement. The model can process information including text, images, videos and other modalities, realize cross-modal sentiment analysis, and improve the accuracy and comprehensiveness of the analysis; through the training of loss function and gradient descent method, the model can continuously optimize its parameters and improve the generalization ability of sentiment analysis;

[0057] 3. By acquiring user feedback data and calculating comprehensive user feedback indicators, the present invention can objectively evaluate the effectiveness of personalized advertising creativity and provide data support for advertising optimization. At the same time, based on the quantitative evaluation results, the creative generation function and creative template can be adjusted and optimized to form a continuously improved advertising publishing process and improve the overall advertising effect and user experience.

[0058] In summary, the present invention is interconnected and mutually supportive through the above steps, together forming an efficient and personalized artificial intelligence advertising publishing method, which can not only comprehensively and accurately collect and process advertising data, but also deeply understand the emotions conveyed by the advertising content by establishing a sentiment analysis model, and then generate personalized advertising ideas that meet the emotional needs of users. In addition, by obtaining user feedback data and evaluating the effectiveness of personalized advertising ideas, the advertising content and form can be continuously optimized to ensure that the effect of advertising is maximized. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of an artificial intelligence advertising publishing method proposed by the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Reference Figure 1 , an artificial intelligence advertising publishing method, comprising:

[0062] Step 1: Collect advertisement-related data, including text, image, and video data, and pre-process the collected data;

[0063] Step 2: Obtain preprocessed data and establish a sentiment analysis model based on the preprocessed data;

[0064] Step 3: Use the loss function to train the established sentiment analysis model and use the gradient descent method to update the model parameters;

[0065] Step 4: Obtain user feedback data and evaluate the effectiveness of personalized advertising creativity in order to publish the next round of advertising.

[0066] It should be noted that the artificial intelligence advertising publishing method proposed in the present invention can be widely applied to various advertising delivery scenarios. The method can be used to improve the efficiency and effectiveness of advertising delivery. Specifically, it can comprehensively analyze and accurately evaluate advertising data through artificial intelligence technology, and ensure accurate advertising delivery and promote the optimization of advertising effects by collecting advertising-related data, establishing sentiment analysis models, training models, and obtaining user feedback. The application of the present invention can significantly improve the pertinence and effectiveness of advertising delivery, reduce advertising costs, and improve user experience and satisfaction. Through intelligent means, the full-chain optimization of advertising delivery from data collection, model building to creative generation and effect evaluation is achieved.

[0067] See also Figure 1 The present invention provides an artificial intelligence advertising publishing method. In step 1, the step of collecting advertisement-related data and preprocessing the collected data includes:

[0068] Step 101: collect text, image and video data to form an original data set D = {T, I, V}; wherein T, I, V represent text, image and video data respectively, and the original data set is text, image and video data from different channels, including social media platforms, online forums, blogs, etc.;

[0069] Step 102: preprocess the original data set. The specific steps are as follows:

[0070] Step A1: For text data T, define the jth word in the text data T as t j ; The word t' after natural language processing j Represented as t' j =f T (t j ), where f T Indicates the word t j The function for processing; then the processed text data T'={t' j tj ∈T};

[0071] In step A1, natural language processing operations include lexical analysis, syntactic analysis, semantic analysis, etc.;

[0072] Step A2: for image data I, define the pixel coordinates in image data I as (a, b), and its original pixel value as P(a, b); the pixel value P'(a, b) after size normalization and color correction is expressed as P'(a, b) = g I (P(a,b)), where g I represents the function of processing image pixels; then the processed image data I'={P'(a,b)(a,b)∈I};

[0073] In step A2, size normalization refers to adjusting the size of the image to a uniform standard size. Size normalization is achieved by scaling the image, such as nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. Color correction refers to adjusting the color balance and tone of the image to make it more accurate and consistent. In image processing, color correction involves adjusting the three color channels of RGB (red, green, and blue).

[0074] Step A3: for the video data V, define the kth frame in the video data V as v k ; Frame v' after frame rate adjustment and resolution adjustment k =c V (v k ), where c V represents the function of processing the video frame; then the processed video data V'={v' k |v k ∈V};

[0075] In step A3, frame rate adjustment refers to changing the frame rate of the video to adapt to different application scenarios or requirements. The specific method involves reading the video file, adjusting the frame rate parameters, and then re-encoding and saving the video; resolution adjustment refers to changing the resolution of the image to adapt to different display devices or requirements by specifying the resolution parameters in the model input layer or data loading function using a machine learning framework;

[0076] Step 103: Integrate the preprocessed data set D'={T', I', V'}.

[0077] See also Figure 1 The present invention provides an artificial intelligence advertising publishing method. In the step 2, the step of obtaining pre-processed data and establishing a sentiment analysis model based on the pre-processed data includes:

[0078] Step 201: Obtain a preprocessed data set and mark it as D'; each data has been converted into a vector representation of a fixed length, that is, x∈D', and the input vector x contains information of multiple modalities such as text, image or video;

[0079] Step 202: Establish a sentiment analysis model EAM (x; Θ) to estimate the sentiment polarity of the input vector x, where Θ represents all parameters of the model; wherein the model establishment includes the setting of a feature embedding layer, a multi-layer perceptron, and an output layer, and the setting steps are as follows:

[0080] Step B1: For the feature embedding layer of the sentiment analysis model, in order to capture more complex semantics and representation capabilities, the input vector x is mapped to a high-dimensional space through the embedding layer. Suppose the embedding layer consists of a weight matrix W emb ∈ R m×n Parameterization, where m represents the size of the hidden layer, n represents the size of the input vector, and R does not directly represent a specific meaning or variable, but is used to represent the number of linearly independent rows or columns in the matrix;

[0081] Through the feature embedding layer formula z = W emb x+ ε emb Realize the transformation from low-level representation to high-level abstract feature space. In the formula, z represents the embedded representation of the input vector x after linear transformation, that is, the embedded vector. The embedded vector z transforms the original input vector x (which may be the original word vector, image pixel or audio spectrum, etc.) into a high-dimensional space. This new representation contains more semantic information and intrinsic associations, which is more conducive to the subsequent model to capture and understand the deep features of the data; the embedding layer weight matrix W emb The role of is to reshape and project the input vector x, which is equivalent to a linear combination of the data, reflecting the correlation between the input features and each dimension in the embedding space; ε emb Represents the bias vector, which belongs to the model parameter Θ and is used to ensure that z has a certain non-zero basis even when the input vector x is a zero vector, thus enhancing the flexibility of the model;

[0082] In step B1, the embedding vector z obtained by the feature embedding layer formula is more suitable for downstream tasks such as sentiment analysis, because it integrates the semantic information of the input vector x to a certain extent, so that it can be more effectively understood in the subsequent nonlinear layer.

[0083] Step B2: The multi-layer perceptron part of the sentiment analysis model includes several hidden layers, each of which increases nonlinearity by applying an activation function σ; wherein the output of each layer is passed to the next layer as input until the last layer produces the final advertisement sentiment prediction;

[0084] Let L be the number of layers, z l is the output of the lth layer, W l , ε l are the weight matrix and bias vector of the layer respectively, then there is the formula z l =σ(W l z l-1 +ε l );

[0085] Step B3: For the output layer of the sentiment analysis model, the output layer is responsible for mapping the output of the hidden layer to the sentiment category;

[0086] Set the output to three categories, including positive, neutral and negative emotions. The output layer has a softmax layer to generate the probability of each category.

[0087] Define W out , ε out are the weight and bias parameters of the output layer, which are also part of Θ, the parameter of the sentiment analysis model;

[0088] Using the formula y=softmax(W out z L +ε out ) Output the predicted advertising sentiment category probability distribution y, where z L Represents the last layer of the multi-layer perceptron, that is, the output of the Lth layer;

[0089] Understandably, L It has the function of summarizing all the features extracted by the previous layers. Its essence is the composite result of multiple layers of nonlinear transformations, which can capture the high-level abstract representation of the input data. For sentiment analysis tasks, z L Contains the key information that the model considers most relevant to sentiment classification. This information is extracted and condensed step by step from the initial input vector x, and has undergone multiple feature learning and pattern recognition processes.

[0090] Among them, the softmax function is defined as:

[0091] In the formula, softmax(z) i Represents the probability of sentiment category i; is an exponential function, representing a specific category, used to convert the output z of the multilayer perceptron l Mapping into the probability domain ensures that all categories have a non-negative probability value; Indicates all categories The sum of; i represents the sentiment category index, when i = 1, then z 1 Indicates positive emotion. When i=2, z 2 Indicates neutral sentiment. When i=3, z 3 Represents negative sentiment; J is the sum index variable, traversing all possible output categories;

[0092] In step B3, the output layer then converts z L Mapped to a specific emotion category, a probability distribution y is generated by applying the softmax function, and each element y in y i It represents the predicted probability of the corresponding emotional category. In other words, the vector form before the output layer carries sufficient emotional motivation information, and the softmax function converts it into an easy-to-interpret category probability form, which facilitates the model to make the final emotional classification decision.

[0093] See also Figure 1 The present invention provides an artificial intelligence advertising publishing method. In step three, the step of using the loss function to train the established sentiment analysis model and using the gradient descent method to update the model parameters includes:

[0094] Step 301: Set a loss function training model, where the loss function is:

[0095] Lo(x,τ;Θ)=-Σ i τ i log(y i ), where τ represents the real sentiment label of the advertisement, τi Represents the true label corresponding to the emotion category i, corresponding to the output layer z i The value of y i represents the predicted probability of advertising sentiment category i;

[0096] Step 302: Use the gradient descent method to update the model parameters θ, that is,

[0097] In the formula, h represents the number of iterations, η represents the learning rate, represents the gradient with respect to the parameter Θ;

[0098] See also Figure 1 The present invention provides an artificial intelligence advertising publishing method. In step 4, the steps of obtaining user feedback data and evaluating the personalized advertising creative effect so as to publish the next round of advertising include:

[0099] Step 401: Obtain a preset advertising creative template M;

[0100] In step 401, the advertisement creative template is a predefined advertisement creative structure used to guide the advertisement creative generation process, covering specific text format, image layout, color matching, etc.;

[0101] Step 402: introducing an advertisement creative generation function G;

[0102] Step 403, the advertising creative generation function G accepts the advertising creative template M, the predicted emotion category probability distribution y and the corresponding advertising real emotion label τ as input, and outputs the personalized advertising creative s=G(M,y,τ,);

[0103] In step 403, this is a multi-level mapping process. Function mapping refers to mapping elements in the input space to elements in the output space through certain rules or relationships. The input space consists of creative templates, predicted probability distribution of emotion categories and corresponding real emotion labels, while the output space is personalized advertising creativity. The creative generation function needs to find or generate content matching these emotions in the creative template according to the real emotion labels, which involves parsing the emotion labels, replacing or adjusting the variable elements in the template, etc.

[0104] Step 404: Obtain user feedback data U = {(u 1 ,r 1 ),(u 2 ,r 2 ),...,(u H ,r H )}, where u represents the user's operation on the personalized advertising creative s (such as clicks, viewing time, etc.), r represents the corresponding evaluation score (such as 0-10 points), and H represents the number of types of operations performed by the user on the personalized advertising creative (such as clicks, viewing time, sharing, comments, etc.);

[0105] Step 405: Combine the user's various operations on the personalized advertising creative and the corresponding evaluation scores to construct a comprehensive user feedback index: In the formula, F(q) represents the comprehensive user feedback index value, q represents the index of the type of operation performed by the user on the personalized advertising creativity, and w q represents the weight of the qth operation, which is used to reflect the importance of the operation to the overall feedback;

[0106] Step 406: Calculate the effectiveness value of the personalized advertising creative s using the creative evaluation function:

[0107] In the formula, E(s) represents the effectiveness value of personalized advertising creative s, N s An index collection representing user feedback data related to personalized advertising creative s;

[0108] Step 407: According to the effectiveness value of the personalized advertising creative s, the creative generation function G and the creative template M are adjusted to ensure smooth progress to the next round of advertising release: according to user feedback and the evaluation results of the effectiveness value of the personalized advertising creative, the layout of the creative template is optimized, for example, the layout of the advertising content is optimized, the matching of pictures and texts is adjusted, etc., to improve the attractiveness and readability of the advertisement; the design of the creative template is improved, such as changing the color scheme, adjusting the font style, etc., to better match the aesthetic preferences of the target audience; at the same time, according to the evaluation results of the effectiveness value of the personalized advertising creative, the elements with poor performance in the creative template are replaced, such as replacing pictures, modifying the copy, etc., to improve the overall effect of the advertisement; after adjusting the creative generation function and the creative template, it is necessary to use the user feedback data again to evaluate the effectiveness value of the personalized advertising creative to verify the effect of the adjustment, which can be achieved by comparing the effectiveness values ​​of the advertising creative before and after the adjustment. If the effectiveness value after the adjustment is improved, it means that the adjustment is effective; if the effectiveness value of the personalized advertising creative does not change significantly or even decreases, it is necessary to further analyze the reasons and make adjustments.

[0109] In the embodiment of the present invention, text, image and video data are collected from different channels such as social media platforms, online forums, blogs, etc. to form an original data set. The preprocessing process further improves the data quality, including natural language processing of text, size normalization and color correction of images, frame rate adjustment and resolution adjustment of videos, so that the data is more standardized and consistent, providing a solid foundation for subsequent sentiment analysis and advertising creative generation. By establishing a sentiment analysis model, the sentiment information in the advertising data can be deeply mined. Through the setting of the feature embedding layer, the multi-layer perceptron and the output layer, the model can accurately estimate the sentiment polarity of the input vector, helping advertisers to better understand the target audience. emotional tendencies, so as to formulate more accurate and effective advertising strategies, improve the personalization of advertisements, and make the advertising content closer to the emotional needs of users. By training the sentiment analysis model and using the gradient descent method to update the model parameters, the performance of the model can be continuously optimized, and the accuracy of sentiment classification can be improved. By obtaining user feedback data, it is possible to directly understand the user's acceptance and satisfaction with personalized advertising creativity, help advertisers to adjust advertising strategies in a timely manner, optimize advertising creativity, and improve the attractiveness and conversion rate of advertisements. By constructing a comprehensive user feedback index and calculating the effectiveness value of personalized advertising creativity, advertisers can quantitatively evaluate the advertising effect, so as to formulate advertising budgets and delivery plans more scientifically. In summary, the examples of the present invention involve data processing, comprehensive analysis and intelligent optimization decisions, and solve the technical problem of poor advertising publishing effects in existing solutions. In actual situations, more data and contextual information may be needed to make specific decisions and optimization plans.

[0110] In addition, the formulas involved in the above are all calculated by removing dimensions and taking their numerical values. They are a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The proportional coefficient in the formula and the various preset thresholds in the analysis process are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data; the size of the proportional coefficient is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the proportional coefficient depends on the amount of sample data and the preliminary setting of the corresponding processing coefficient for each group of sample data by technical personnel in this field; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0111] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically based on the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0112] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0113] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0115] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0117] Secondly: In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other;

[0118] Finally: The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An artificial intelligence advertising publishing method, characterized by: Step 1: Collect advertisement-related data, including text, image, and video data, and pre-process the collected data; Step 2: Obtain preprocessed data and establish a sentiment analysis model based on the preprocessed data; Step 3: Use the loss function to train the established sentiment analysis model and use the gradient descent method to update the model parameters; Step 4: Obtain user feedback data and evaluate the effectiveness of personalized advertising creativity in order to publish the next round of advertising.

2. The artificial intelligence advertising publishing method according to claim 1, characterized in that: In step 1, the process of collecting advertisement-related data and preprocessing the collected data includes: Collect text, image and video data to form the original data set D = {T, I, V}; where T, I, V represent text, image and video data respectively; Preprocess the original data set: For text data T, define the jth word in text data T as t j ; The word t' after natural language processing j Denoted as t' j =f T (t j ), where f T Represents the word t j The function for processing; then the processed text data T'={t' j |t j ∈T}; For image data I, the pixel coordinates in image data I are defined as (a, b), and its original pixel value is P(a, b); the pixel value P'(a, b) after size normalization and color correction is expressed as P'(a, b) = g I (P(a,b)), where g I represents the function of processing image pixels; then the processed image data I'={P'(a,b)|(a,b)∈I}; For video data V, define the kth frame in video data V as v k ; Frame v' after frame rate adjustment and resolution adjustment k =c V (v k ), where c V represents the function of processing the video frame; then the processed video data V'={v' k |v k ∈V}; Integrate the preprocessed data set D'={T',I',V'}.

3. The artificial intelligence advertising publishing method according to claim 1, characterized in that: In the step 2, the process of obtaining the preprocessed data and establishing a sentiment analysis model based on the preprocessed data includes: Get the preprocessed data set and mark it as D'; A sentiment analysis model EAM(x; Θ) is established to estimate the sentiment polarity of the input vector x. The model establishment includes the setting of feature embedding layer, multi-layer perceptron and output layer; where Θ represents all parameters of the model.

4. The artificial intelligence advertising publishing method according to claim 3, characterized in that: In step 2, the process of setting the feature embedding layer, multi-layer perceptron and output layer of the sentiment analysis model includes: For the feature embedding layer of the sentiment analysis model, the input vector x is mapped to a high-dimensional space through the embedding layer. Suppose the embedding layer is composed of a weight matrix W emb ∈R m×n Parameterization, where m represents the hidden layer size and n represents the size of the input vector; Through the feature embedding layer formula z = W emb x+ε emb Realize the transition from low-level representation to high-level abstract feature space. In the formula, z represents the embedded representation of the input vector x after linear transformation, that is, the embedded vector; the embedding layer weight matrix W emb The role of is to perform reshaping and projection operations on the input vector x; ε emb represents the bias vector, which also belongs to the model parameter Θ; The multi-layer perceptron part of the sentiment analysis model includes several hidden layers, each of which increases nonlinearity by applying an activation function σ; the output of each layer is passed to the next layer as input until the last layer produces the final ad sentiment prediction; Let L be the number of layers, z l is the output of the lth layer, W l , ε l are the weight matrix and bias vector of the layer respectively, then there is the formula z l =σ(W l z l-1 +ε l ); For the output layer of the sentiment analysis model, the output layer is responsible for mapping the output of the hidden layer to the sentiment category; Set the output to three categories, including positive, neutral and negative emotions. The output layer has a softmax layer to generate the probability of each category. Define W out , ε out are the weight and bias parameters of the output layer respectively; Use the formula y=soft max(W out z L +ε out ) Output the predicted advertising sentiment category probability distribution y, where z L represents the output of the last layer of the multi-layer perceptron, i.e., the Lth layer; the softmax function is defined as: In the formula, softmax(z) i Represents the probability of sentiment category i; is an exponential function; Indicates all categories The sum of; i represents the sentiment category index, when i = 1, z1 represents positive sentiment, when i = 2, z2 represents neutral sentiment, when i = 3, z3 represents negative sentiment; J is the sum index variable, traversing all possible output category numbers.

5. The artificial intelligence advertising publishing method according to claim 4, characterized in that: In step 3, the process of training the established sentiment analysis model using the loss function and updating the model parameters using the gradient descent method includes: Set the loss function to train the model, where the loss function is: Lo(x,τ;Θ)=-Σ i τ i log(y i ), where τ represents the real sentiment label of the advertisement, τ i Represents the true label corresponding to the emotion category i, corresponding to the output layer z i The value of y i represents the predicted probability of advertising sentiment category i; Use the gradient descent method to update the model parameters Θ, that is, In the formula, h represents the number of iterations, η represents the learning rate, represents the gradient with respect to the parameter Θ.

6. The artificial intelligence advertising publishing method according to claim 1, characterized in that: In step 4, the process of obtaining user feedback data and evaluating the creative effect of personalized advertising includes: Get the preset advertising creative template M; Introduce advertising creative generation function G; The advertising creative generation function G accepts the advertising creative template M, the predicted emotion category probability distribution y and the corresponding advertising real emotion label τ as input, and outputs personalized advertising creative s = G(M, y, τ,); Get user feedback data U = {(u1,r1),(u2,r2),...,(u H ,r H )}, where u represents the user's operation on the personalized advertising creative s, r represents the corresponding evaluation score, and H represents the number of types of operations performed by the user on the personalized advertising creative; Combine the user's various operations on personalized advertising creatives and their corresponding evaluation scores to construct a comprehensive user feedback index: In the formula, F(q) represents the comprehensive user feedback index value, q represents the index of the type of operation performed by the user on the personalized advertising creative, and w q represents the weight of the qth operation, which is used to reflect the importance of the operation to the overall feedback; Use the creative evaluation function to calculate the effectiveness value of personalized advertising creative s: Where E(s) represents the effectiveness value of personalized advertising creative s, N s An index collection representing user feedback data related to personalized advertising creative s; According to the effectiveness value of personalized advertising creative s, the creative generation function G and creative template M are adjusted to ensure smooth progress to the next round of advertising release.

Images