A real-time advertising effect tracking method based on multi-objective optimization algorithm

Abstract

The present invention discloses a method for tracking the real-time effect of advertising based on a multi-objective optimization algorithm, including S1, capturing user behavior data related to advertising, collecting market dynamics and competitor advertising performance data from a third-party data provider; S2, formatting, cleaning and preliminarily analyzing the collected user behavior data, market dynamics and competitor advertising performance data; S3, extracting key features from advertising content using natural language processing and image recognition technology; S4, designing an advertising performance optimization model using a multi-objective genetic algorithm; S5, automatically adjusting the advertising delivery strategy according to the output results of the multi-objective genetic algorithm; S6, using a machine learning model to predict the market reaction after the adjusted advertising delivery strategy is delivered; S7, and performing performance data after the advertising is delivered. The present invention uses a multi-objective genetic algorithm to design an advertising performance optimization model to ensure the scientific nature and optimization effect of the advertising delivery strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement delivery, and in particular to a method for tracking the real-time effect of advertisement delivery based on a multi-objective optimization algorithm. Background Art

[0002] In the field of digital marketing, real-time effect tracking and optimization of advertising has always been an important research direction. Traditional advertising strategies usually rely on static data analysis and manual adjustments, which often fail to achieve ideal results in a rapidly changing market environment. In existing technologies, advertising mainly faces the following problems:

[0003] Advertising involves multiple performance indicators, such as cost efficiency, audience coverage, and user engagement, which are often conflicting. Existing advertising strategies often focus on optimizing a single goal while ignoring other equally important indicators, resulting in an overall decline in advertising effectiveness. For example, in some cases, improving user click-through rates may increase advertising costs and reduce the economic benefits of advertising.

[0004] In existing technologies, tracking of advertising effectiveness usually relies on batch data processing, which cannot achieve real-time data capture and analysis. This delayed data processing method makes it difficult for advertisers to understand the immediate effects of advertising in a timely manner, and they cannot quickly adjust advertising strategies to cope with market changes, thus missing out on many potential optimization opportunities.

[0005] The data involved in the advertising process comes from a wide range of sources, including user behavior data, market dynamics data, and competitor advertising performance data. Existing technologies often lack effective methods for integrating and analyzing this data, resulting in low data utilization and difficulty in fully evaluating advertising effectiveness. For example, some advertising platforms may only be able to obtain user click-through rate and viewing time data, but cannot integrate market dynamics and competitor performance data from third-party data providers, thus affecting the scientific nature of advertising decisions.

[0006] Traditional advertising strategy adjustments mainly rely on manual experience and subjective judgment, lacking automated and adaptive optimization mechanisms. In existing technologies, although some systems have introduced machine learning algorithms to predict advertising effects, they are usually limited to static models and cannot be dynamically adjusted based on real-time data, resulting in the adjustment of advertising strategies lagging behind market changes.

[0007] Although multi-objective optimization algorithms can theoretically optimize multiple objectives simultaneously, in practical applications, due to the complexity of the algorithm and the limitation of computing resources, the application of multi-objective genetic algorithms in existing technologies still has many limitations. For example, how to effectively integrate user behavior data, market dynamics and competitor performance data in multi-objective genetic algorithms, and find the best advertising delivery strategy through Pareto frontier analysis, these problems have not been effectively solved in existing technologies.

[0008] Therefore, how to provide a real-time effect tracking method for advertisement delivery based on a multi-objective optimization algorithm is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0009] One purpose of the present invention is to propose a real-time effect tracking method for advertising based on a multi-objective optimization algorithm. The present invention uses a multi-objective genetic algorithm to design an advertising performance optimization model, determines the trade-off relationship between different advertising performance indicators through Pareto frontier analysis, and finds the optimal trade-off point between the indicators through iterative update solutions to ensure the scientific nature and optimization effect of the advertising strategy.

[0010] A method for tracking the real-time effect of advertisement delivery based on a multi-objective optimization algorithm according to an embodiment of the present invention comprises the following steps:

[0011] S1. Through data collection modules deployed on multiple digital platforms, real-time capture of user behavior data related to advertising, including user click-through rate, viewing time, audience response and interaction data, as well as collection of market dynamics and competitor advertising performance data from third-party data providers;

[0012] S2. Format, clean and preliminarily analyze the collected user behavior data, market dynamics and competitor advertising performance data;

[0013] S3. Use natural language processing and image recognition technology to extract key features including sentiment, visual elements and topic classification from advertising content, and use statistical methods to screen out key data that has a decisive impact on advertising effectiveness evaluation;

[0014] S4. Use a multi-objective genetic algorithm to design an advertising performance optimization model. The advertising performance optimization model combines pre-processed user behavior data, market dynamics, and competitor advertising performance data, uses Pareto frontier analysis to determine the trade-off relationship between different advertising performance indicators, and finds the optimal trade-off point between various indicators through iterative update solutions;

[0015] S5. Automatically adjust the advertising delivery strategy based on the output results of the multi-objective genetic algorithm. The advertising delivery strategy includes modification of advertising content, optimization of delivery time, and precise positioning of target audiences.

[0016] S6. Use machine learning models to predict the market response after the adjusted advertising strategy is launched, and adjust the advertising strategy parameters in real time;

[0017] S7. Re-enter the performance data after the advertisement is released, including the audience's purchasing behavior, user feedback and revisit rate, into the system, and use the optimized multi-objective genetic algorithm to continue performance evaluation and strategy adjustment, forming a closed-loop advertising effect automatic optimization system with dynamic learning and adaptive adjustment.

[0018] Optionally, the S1 includes the following steps:

[0019] S11. Deploy a data collection module on multiple digital platforms, including a social media platform, a search engine, and a video platform, wherein the data collection module is capable of capturing user behavior data in real time;

[0020] S12, captured user behavior data including user click-through rate CTR, viewing time WT, audience response AR and interaction data ID;

[0021] Click-through rate is calculated by weighting the number of user clicks to the number of impressions, taking into account the relative ratio of impressions to clicks and the sinusoidal relationship between impressions and clicks:

[0022] ;

[0023] in, represents the number of clicks on the i-th display, represents the weight value of the i-th display, represents the number of impressions of the ith impression, Represents the total number of all impressions. Represents the total number of clicks for all impressions;

[0024] The viewing time is calculated by weighting the cosine function relationship between the user's viewing time and the maximum viewing time, and combining the square root of the viewing time with the tangent function relationship:

[0025] ;

[0026] in, represents the viewing time of the i-th user, represents the maximum viewing time, k represents the adjustment coefficient, represents the viewing time of the i-th user at the j-th viewing time, represents the total viewing time of the i-th user at the j-th viewing time;

[0027] Audience reactions are calculated by taking the weighted average of the emotion score and emotion intensity, and are corrected by combining the emotion change rate, emotion fluctuation, and its cosine function relationship:

[0028] ;

[0029] in, represents the weight of the sentiment score, represents the weight of the emotion intensity, represents the sentiment score of the i-th user, represents the emotional intensity of the i-th user, represents the emotion change rate of the i-th user, represents the emotional fluctuation of the i-th user, represents the sentiment index of the i-th user;

[0030] The interaction data is calculated by the weighted sum of the number of interactions and the interaction depth, and is corrected by combining the interaction quality score with the square root relationship of the interaction frequency:

[0031] ;

[0032] in, represents the weight of the number of interactions, represents the weight of the interaction quality score, represents the number of interactions of the i-th user, represents the interaction depth of the i-th user, represents the interaction quality score of the i-th user, represents the interaction frequency of the i-th user, represents the interaction index of the i-th user;

[0033] S13. Collect market dynamics data and competitor advertising performance data from third-party data providers, including market trends and competitor advertising effectiveness;

[0034] The market trend is calculated by weighting the sine function relationship between the market change rate and the search index, and is determined by combining the hyperbolic cosine function of the volatility factor and the industry growth rate, the square root of the economic index and the market share, and the tangent function relationship:

[0035] ;

[0036] in, represents the weight of the market change rate, represents the weight of the industry growth rate, represents the rate of change of the ith market, represents the search index of the ith market, represents the volatility factor of the ith market, represents the growth rate of the ith industry, represents the economic index of the ith market, represents the market share of the ith market;

[0037] Competitor advertising effectiveness is weighted by the cosine function relationship between competitor clicks and impressions, and confirmed by the sine function relationship between advertising influence and advertising costs:

[0038] ;

[0039] in, represents the number of clicks on the jth competitor's ad, represents the number of impressions of the jth competitor ad, represents the influence of the jth competitor’s advertisement, represents the cost of the jth competitor’s advertisement, Denotes the number of times the jth competitor ad is displayed.

[0040] Optionally, S3 includes the following steps:

[0041] S31, extracting emotional tendency features from the advertisement content using natural language processing technology, wherein the emotional tendency features include emotional score, emotional intensity and emotional change rate:

[0042] ;

[0043] in, represents the sentiment score of the ith advertisement, represents the weight of sentiment words, represents the score of the jth sentiment word in the i-th advertisement;

[0044] ;

[0045] in, represents the emotional intensity of the ith advertisement, represents the score of the kth emotional intensity word in the ith advertisement, and n represents the total number of emotional intensity words;

[0046] ;

[0047] in, represents the sentiment change rate of the i-th advertisement, represents the score of the jth sentiment word in the i-th advertisement, represents the score of the previous sentiment word in the i-th advertisement;

[0048] S32. Extract visual element features from the advertisement content using image recognition technology, wherein the visual element features include color distribution, image texture and shape features:

[0049] ;

[0050] in, represents the color distribution of the i-th advertisement, represents the color value of the lth pixel in the ith advertisement, and p represents the total number of pixels;

[0051] ;

[0052] in, represents the image texture of the i-th advertisement, represents the value of the mth texture feature in the i-th advertisement, and q represents the total number of texture features;

[0053] ;

[0054] in, represents the shape feature of the i-th advertisement, represents the area of ​​the oth shape in the ith advertisement, represents the total area of ​​all shapes in the ith advertisement, represents the perimeter of the oth shape in the ith ad;

[0055] S33, extracting topic classification features from the advertisement content, classifying the advertisement text through a text classification algorithm, and extracting the main topic categories:

[0056] ;

[0057] in, represents the main topic score of the i-th advertisement, represents the score of the tth topic word in the ith advertisement, and s represents the total number of topic words;

[0058] S34. Through correlation analysis, emotional tendency characteristics, visual element characteristics and topic classification characteristics are screened to select key data that have a decisive influence on the evaluation of advertising effectiveness.

[0059] Optionally, the S34 includes the following steps:

[0060] S341. Use nonlinear correlation analysis method to perform correlation analysis on sentiment tendency features, where sentiment tendency features include sentiment scores , emotional intensity and emotional change rate ,The correlation between the sentiment tendency characteristics and the advertising effect evaluation results is calculated by the following formula:

[0061] ;

[0062] in, , , are the weights of sentiment score, sentiment intensity and sentiment change rate, respectively. and are the means of the sentiment tendency feature set and the effect evaluation results, respectively. The nonlinear term in the formula Used to capture the interrelationships of sentiment tendency features;

[0063] S342. Use high-dimensional correlation analysis method to perform correlation analysis on visual element features, including color distribution , Image Texture and shape features ,The correlation between the visual element features and the advertising effect evaluation results is calculated by the following formula:

[0064] ;

[0065] in, , , are the weights of color distribution, image texture and shape features, respectively. and are the mean of the visual element feature set and the effect evaluation result, respectively. The high-dimensional term in the formula Used to capture the interrelationships of visual element features;

[0066] S343. Use multidimensional correlation analysis method to conduct correlation analysis on topic classification features. Topic classification features include main topic scores ,The correlation between the topic classification features and the advertising effect evaluation results is calculated by the following formula:

[0067] ;

[0068] in, and are the mean of the topic classification feature set and the effect evaluation result, respectively. The multidimensional term in the formula Used to capture the interrelationships of topic classification features;

[0069] S344. Conduct correlation analysis on comprehensive sentiment tendency features, visual element features and topic classification features, and calculate the correlation between the comprehensive features and the advertising effect evaluation results.

[0070] Optionally, S4 includes the following steps:

[0071] S41. Design an advertising performance optimization model using a multi-objective genetic algorithm, which combines pre-processed user behavior data, market dynamics, and competitor advertising performance data;

[0072] S42, initialize the population, generate an initial population containing multiple individuals, each individual represents an advertising strategy, and the individual is composed of a feature vector of user behavior data, market dynamics and competitor advertising performance data constitute;

[0073] S43. Define a fitness function, taking the cost efficiency, audience coverage and user engagement of advertising performance indicators as targets, and calculate the fitness function by the following formula:

[0074] ;

[0075] in, represents the feature vector of the i-th individual, , and are the weights of advertising performance indicators, They are the weights of click-through rate, viewing time, audience response and interactive data, MT represents market dynamics, SI represents search index, BF represents volatility factor, IG represents industry growth rate, EI represents economic index, MS represents market share, CC represents competitor clicks, CD represents competitor impressions, AI represents advertising influence, and AF represents advertising cost;

[0076] S44. Use Pareto frontier analysis to determine the trade-offs between different advertising performance indicators, construct a Pareto frontier set, and find the optimal balance point between multiple advertising performance indicators:

[0077] ;

[0078] Among them, P is the population, f is the fitness function;

[0079] S45, selection operation, selects excellent individuals in the population based on the fitness function value. The selection probability is proportional to the fitness value of the individual. The selection probability is calculated by the following formula:

[0080] ;

[0081] in, represents the i-th individual, N represents the total number of the population, To select the pressure parameter;

[0082] S46, crossover operation, generating new individuals by performing a crossover operation on the selected individuals, and the crossover operation is performed by the following formula:

[0083] ;

[0084] in, represents the jth gene of the offspring individual, and They represent the jth gene of the parent individual, is the cross coefficient, is the fusion coefficient of market and competitor data;

[0085] S47, mutation operation, mutation operation is performed on the newly generated individuals to increase the diversity of the population. The mutation operation is performed using the following formula:

[0086] ;

[0087] in, represents the jth gene of the mutated offspring individual, is the coefficient of variation, and user behavior data includes click-through rate, viewing time, audience response and interaction data;

[0088] S48, iterative update, through multiple iterative selection, crossover and mutation operations, gradually optimize the individuals in the population until the preset termination conditions are met. The termination conditions include the maximum number of iterations or the population fitness value reaches stability. The iterative update is performed by the following formula:

[0089] ;

[0090] Where k represents the number of iterations, represents the i-th individual in the k-th iteration, is the fusion coefficient;

[0091] S49. Output the optimal solution and select the individual with the highest fitness value from the final population as the optimal advertising delivery strategy.

[0092] Optionally, S5 includes the following steps:

[0093] S51. According to the output result of the multi-objective genetic algorithm, determine the advertisement content that needs to be modified, including advertisement text, image and video elements. The modification of the advertisement content is performed by the following formula:

[0094] ;

[0095] in, represents the modified content of the jth ad element, Represents the original content of the jth ad element, represents the optimization suggestion for the jth ad element. is the content modification coefficient;

[0096] S52. According to the output results of the multi-objective genetic algorithm, the advertisement delivery time is optimized, and the optimal delivery time window is calculated by the following formula:

[0097] ;

[0098] Where t represents time, , and They are the weights of click-through rate, viewing time and interaction data respectively;

[0099] S53. According to the output results of the multi-objective genetic algorithm, the target audience is accurately positioned, and the audience positioning parameters are calculated by the following formula:

[0100] ;

[0101] in, represents the characteristics of the i-th audience, represents the reaction of the i-th audience to the advertisement, represents the interaction data of the i-th audience, and N is the total number of audiences;

[0102] S54. Apply the modified advertising content, optimized delivery time and precisely positioned target audience to the actual advertising delivery strategy, form a dynamic adjustment mechanism, and continuously optimize the advertising delivery strategy based on real-time feedback.

[0103] Optionally, the S6 includes the following steps:

[0104] S61. Construct a machine learning model. The machine learning model uses pre-processed user behavior data, market dynamics, and competitor advertising performance data as input to predict the market response to the adjusted advertising strategy. The machine learning model makes predictions using the following formula:

[0105] ;

[0106] Where y represents the predicted market reaction, , and are the weights of user behavior data, market dynamics, and competitor advertising performance, respectively, and f represents the prediction function of the machine learning model;

[0107] S62. Train the model, use historical advertising data and its actual market response as a training set, optimize the model parameters through the gradient descent algorithm, and calculate the loss function of the model through the following formula:

[0108] ;

[0109] Among them, L represents the loss function, N is the number of training set samples, represents the actual market reaction of the ith sample, represents the predicted market reaction of the i-th sample, and R(w) is the regularization term of the model parameters;

[0110] S63. Use the trained machine learning model to predict the market reaction to the adjusted advertising strategy, input the latest user behavior data, market dynamics and competitor advertising performance data, and the model outputs the predicted market reaction value;

[0111] S64. Monitor the actual market reaction in real time, compare the actual market reaction data with the market reaction value predicted by the model, and calculate the error using the following formula:

[0112] ;

[0113] Among them, e represents the error, M is the number of monitoring data samples, represents the actual market reaction of the jth sample, represents the predicted market reaction of the jth sample;

[0114] S65. Adjust the advertisement delivery strategy parameters in real time based on the error value, and optimize the parameters using the following formula:

[0115] ;

[0116] in, is the learning rate, is the partial derivative of the error with respect to the parameter, so that the advertising delivery strategy can maintain the best effect in the dynamic changes of the market.

[0117] The beneficial effects of the present invention are:

[0118] (1) The present invention deploys data collection modules on multiple digital platforms to capture user behavior data including user click-through rate, viewing time, audience response and interaction data in real time, and integrates market dynamics and competitor advertising performance data collected from third-party data providers, formats, cleans and preliminarily analyzes these data to ensure the timeliness and accuracy of the data.

[0119] (2) The present invention uses natural language processing and image recognition technology to extract key features including emotional tendencies, visual elements and topic classification from advertising content, and uses statistical methods to screen out key data that have a decisive influence on the evaluation of advertising effectiveness, thereby ensuring the comprehensiveness and effectiveness of feature extraction.

[0120] (3) The present invention uses a multi-objective genetic algorithm to design an advertising performance optimization model, determines the trade-off relationship between different advertising performance indicators through Pareto frontier analysis, and finds the optimal trade-off point between the indicators through iterative update solutions, thereby ensuring the scientific nature and optimization effect of the advertising delivery strategy.

[0121] (4) The present invention automatically adjusts the advertising delivery strategy based on the output results of the multi-objective genetic algorithm, including the modification of advertising content, optimization of delivery time and precise positioning of target audiences. Through continuous real-time monitoring and dynamic adjustment, the optimal execution of the advertising delivery strategy is ensured.

[0122] (5) The present invention uses a machine learning model to predict the market response to the adjusted advertising delivery strategy, monitors the actual market response data in real time, and adjusts the advertising delivery strategy parameters in real time through error analysis to ensure that the advertising delivery strategy maintains the best effect in the dynamic changes of the market. BRIEF DESCRIPTION OF THE DRAWINGS

[0123] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0124] Figure 1 The present invention provides a flowchart of a method for tracking the real-time effect of advertisement delivery based on a multi-objective optimization algorithm. DETAILED DESCRIPTION

[0125] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0126] refer to Figure 1 A method for tracking the real-time effect of advertising based on a multi-objective optimization algorithm comprises the following steps:

[0127] S1. Through data collection modules deployed on multiple digital platforms, real-time capture of user behavior data related to advertising, including user click-through rate, viewing time, audience response and interaction data, as well as collection of market dynamics and competitor advertising performance data from third-party data providers;

[0128] In this implementation, S1 includes the following steps:

[0129] S11. Deploy data collection modules on multiple digital platforms, including social media platforms, search engines, and video platforms, and the data collection modules can capture user behavior data in real time;

[0130] S12, captured user behavior data including user click-through rate CTR, viewing time WT, audience response AR and interaction data ID;

[0131] Click-through rate is calculated by weighting the number of user clicks to the number of impressions, taking into account the relative ratio of impressions to clicks and the sinusoidal relationship between impressions and clicks:

[0132] ;

[0133] in, represents the number of clicks on the i-th display, represents the weight value of the i-th display, represents the number of impressions of the ith impression, Represents the total number of all impressions. Represents the total number of clicks for all impressions;

[0134] The viewing time is calculated by weighting the cosine function relationship between the user's viewing time and the maximum viewing time, and combining the square root of the viewing time with the tangent function relationship:

[0135] ;

[0136] in, represents the viewing time of the i-th user, represents the maximum viewing time, k represents the adjustment coefficient, represents the viewing time of the i-th user at the j-th viewing time, represents the total viewing time of the i-th user at the j-th viewing time;

[0137] Audience reactions are calculated by taking the weighted average of the emotion score and emotion intensity, and are corrected by combining the emotion change rate, emotion fluctuation, and its cosine function relationship:

[0138] ;

[0139] in, represents the weight of the sentiment score, represents the weight of the emotion intensity, represents the sentiment score of the i-th user, represents the emotional intensity of the i-th user, represents the emotion change rate of the i-th user, represents the emotional fluctuation of the i-th user, represents the sentiment index of the i-th user;

[0140] The interaction data is calculated by the weighted sum of the number of interactions and the interaction depth, and is corrected by combining the interaction quality score with the square root relationship of the interaction frequency:

[0141] ;

[0142] in, represents the weight of the number of interactions, represents the weight of the interaction quality score, represents the number of interactions of the i-th user, represents the interaction depth of the i-th user, represents the interaction quality score of the i-th user, represents the interaction frequency of the i-th user, represents the interaction index of the i-th user;

[0143] S13. Collect market dynamics data and competitor advertising performance data from third-party data providers, including market trends and competitor advertising effectiveness;

[0144] The market trend is calculated by weighting the sine function relationship between the market change rate and the search index, and is determined by combining the hyperbolic cosine function of the volatility factor and the industry growth rate, the square root of the economic index and the market share, and the tangent function relationship:

[0145] ;

[0146] in, represents the weight of the market change rate, represents the weight of the industry growth rate, represents the rate of change of the ith market, represents the search index of the ith market, represents the volatility factor of the ith market, represents the growth rate of the ith industry, represents the economic index of the ith market, represents the market share of the ith market;

[0147] Competitor advertising effectiveness is weighted by the cosine function relationship between competitor clicks and impressions, and confirmed by the sine function relationship between advertising influence and advertising costs:

[0148] ;

[0149] in, represents the number of clicks on the jth competitor's ad, represents the number of impressions of the jth competitor ad, represents the influence of the jth competitor’s advertisement, represents the cost of the jth competitor’s advertisement, Denotes the number of times the jth competitor ad is displayed.

[0150] This implementation method realizes the comprehensive optimization of multiple performance indicators (such as cost efficiency, audience coverage and user engagement) during the advertising delivery process through a real-time advertising delivery effect tracking method based on a multi-objective optimization algorithm, thereby ensuring the overall effect of advertising delivery. At the same time, by capturing user behavior data and market dynamic data related to advertising delivery in real time, the timeliness and accuracy of the data are improved, supporting real-time decision-making and optimization.

[0151] S2. Format, clean and preliminarily analyze the collected user behavior data, market dynamics and competitor advertising performance data;

[0152] S3. Use natural language processing and image recognition technology to extract key features including sentiment, visual elements and topic classification from advertising content, and use statistical methods to screen out key data that has a decisive impact on advertising effectiveness evaluation;

[0153] In this implementation, S3 includes the following steps:

[0154] S31. Use natural language processing technology to extract emotional tendency features from advertising content. Emotional tendency features include emotional score, emotional intensity and emotional change rate:

[0155] ;

[0156] in, represents the sentiment score of the ith advertisement, represents the weight of sentiment words, represents the score of the jth sentiment word in the i-th advertisement;

[0157] ;

[0158] in, represents the emotional intensity of the ith advertisement, represents the score of the kth emotional intensity word in the ith advertisement, and n represents the total number of emotional intensity words;

[0159] ;

[0160] in, represents the sentiment change rate of the i-th advertisement, represents the score of the jth sentiment word in the i-th advertisement, represents the score of the previous sentiment word in the i-th advertisement;

[0161] S32. Use image recognition technology to extract visual element features from the advertising content. The visual element features include color distribution, image texture and shape features:

[0162] ;

[0163] in, represents the color distribution of the i-th advertisement, represents the color value of the lth pixel in the ith advertisement, and p represents the total number of pixels;

[0164] ;

[0165] in, represents the image texture of the i-th advertisement, represents the value of the mth texture feature in the i-th advertisement, and q represents the total number of texture features;

[0166] ;

[0167] in, represents the shape feature of the i-th advertisement, represents the area of ​​the oth shape in the ith advertisement, represents the total area of ​​all shapes in the ith advertisement, represents the perimeter of the oth shape in the ith ad;

[0168] S33, extracting topic classification features from the advertisement content, classifying the advertisement text through a text classification algorithm, and extracting the main topic categories:

[0169] ;

[0170] in, represents the main topic score of the i-th advertisement, represents the score of the tth topic word in the ith advertisement, and s represents the total number of topic words;

[0171] S34. Through correlation analysis, emotional tendency characteristics, visual element characteristics and topic classification characteristics are screened to select key data that have a decisive influence on the evaluation of advertising effectiveness.

[0172] This implementation method deploys data collection modules on multiple digital platforms to capture user behavior data including user click-through rate, viewing time, audience reaction and interaction data in real time, and integrates market dynamics and competitor advertising performance data collected from third-party data providers, thereby achieving comprehensive data integration and real-time updating, ensuring the scientific nature and accuracy of advertising delivery strategies.

[0173] In this implementation, S34 includes the following steps:

[0174] S341. Use nonlinear correlation analysis method to perform correlation analysis on sentiment tendency features, where sentiment tendency features include sentiment scores , emotional intensity and emotional change rate ,The correlation between the sentiment tendency characteristics and the advertising effect evaluation results is calculated by the following formula:

[0175] ;

[0176] in, , , are the weights of sentiment score, sentiment intensity and sentiment change rate, respectively. and are the means of the sentiment tendency feature set and the effect evaluation results, respectively. The nonlinear term in the formula Used to capture the interrelationships of sentiment tendency features;

[0177] S342. Use high-dimensional correlation analysis method to perform correlation analysis on visual element features, including color distribution , Image Texture and shape features ,The correlation between the visual element features and the advertising effect evaluation results is calculated by the following formula:

[0178] ;

[0179] in, , , are the weights of color distribution, image texture and shape features, respectively. and are the mean of the visual element feature set and the effect evaluation result, respectively. The high-dimensional term in the formula Used to capture the interrelationships of visual element features;

[0180] S343. Use multidimensional correlation analysis method to conduct correlation analysis on topic classification features. Topic classification features include main topic scores ,The correlation between the topic classification features and the advertising effect evaluation results is calculated by the following formula:

[0181] ;

[0182] in, and are the mean of the topic classification feature set and the effect evaluation result, respectively. The multidimensional term in the formula Used to capture the interrelationships of topic classification features;

[0183] S344. Conduct correlation analysis on comprehensive sentiment tendency features, visual element features and topic classification features, and calculate the correlation between the comprehensive features and the advertising effect evaluation results.

[0184] This implementation method uses natural language processing and image recognition technology to extract key features such as emotional tendencies, visual elements and topic classifications from advertising content, and uses statistical methods to screen out key data that have a decisive influence on advertising effectiveness evaluation, thereby achieving comprehensive extraction and effective analysis of advertising content features and improving the accuracy of advertising effectiveness evaluation.

[0185] S4. Use a multi-objective genetic algorithm to design an advertising performance optimization model. The advertising performance optimization model combines pre-processed user behavior data, market dynamics, and competitor advertising performance data, uses Pareto frontier analysis to determine the trade-off relationship between different advertising performance indicators, and finds the optimal trade-off point between various indicators through iterative update solutions;

[0186] In this implementation, S4 includes the following steps:

[0187] S41. Design an advertising performance optimization model using a multi-objective genetic algorithm, which combines pre-processed user behavior data, market dynamics, and competitor advertising performance data;

[0188] S42, initialize the population, generate an initial population containing multiple individuals, each individual represents an advertising strategy, and the individual is composed of a feature vector of user behavior data, market dynamics and competitor advertising performance data constitute;

[0189] S43. Define a fitness function, taking the cost efficiency, audience coverage and user engagement of advertising performance indicators as targets, and calculate the fitness function by the following formula:

[0190] ;

[0191] in, represents the feature vector of the i-th individual, , and are the weights of advertising performance indicators, They are the weights of click-through rate, viewing time, audience response and interactive data, MT represents market dynamics, SI represents search index, BF represents volatility factor, IG represents industry growth rate, EI represents economic index, MS represents market share, CC represents competitor clicks, CD represents competitor impressions, AI represents advertising influence, and AF represents advertising cost;

[0192] S44. Use Pareto frontier analysis to determine the trade-offs between different advertising performance indicators, construct a Pareto frontier set, and find the optimal balance point between multiple advertising performance indicators:

[0193] ;

[0194] Among them, P is the population, f is the fitness function;

[0195] S45, selection operation, selects excellent individuals in the population based on the fitness function value. The selection probability is proportional to the fitness value of the individual. The selection probability is calculated by the following formula:

[0196] ;

[0197] in, represents the i-th individual, N represents the total number of the population, To select the pressure parameter;

[0198] S46, crossover operation, generating new individuals by performing a crossover operation on the selected individuals, and the crossover operation is performed by the following formula:

[0199] ;

[0200] in, represents the jth gene of the offspring individual, and They represent the jth gene of the parent individual, is the cross coefficient, is the fusion coefficient of market and competitor data;

[0201] S47, mutation operation, mutation operation is performed on the newly generated individuals to increase the diversity of the population. The mutation operation is performed using the following formula:

[0202] ;

[0203] in, represents the jth gene of the mutated offspring individual, is the coefficient of variation, and user behavior data includes click-through rate, viewing time, audience response and interaction data;

[0204] S48, iterative update, through multiple iterative selection, crossover and mutation operations, gradually optimize the individuals in the population until the preset termination conditions are met. The termination conditions include the maximum number of iterations or the population fitness value reaches stability. The iterative update is performed by the following formula:

[0205] ;

[0206] Where k represents the number of iterations, represents the i-th individual in the k-th iteration, is the fusion coefficient;

[0207] S49. Output the optimal solution and select the individual with the highest fitness value from the final population as the optimal advertising delivery strategy.

[0208] This implementation method designs an advertising performance optimization model using a multi-objective genetic algorithm, combines user behavior data, market dynamics and competitor advertising performance data, uses Pareto frontier analysis to determine the trade-off relationship between different advertising performance indicators, and finds the optimal trade-off point between each indicator through iterative update solutions, thereby achieving scientific optimization of advertising delivery strategies and improving the overall effectiveness of advertising delivery.

[0209] S5. Automatically adjust the advertising delivery strategy based on the output results of the multi-objective genetic algorithm. The advertising delivery strategy includes modification of advertising content, optimization of delivery time, and precise positioning of target audiences.

[0210] In this implementation, S5 includes the following steps:

[0211] S51. According to the output result of the multi-objective genetic algorithm, determine the advertisement content that needs to be modified, including advertisement text, image and video elements. The modification of the advertisement content is performed by the following formula:

[0212] ;

[0213] in, represents the modified content of the jth ad element, Represents the original content of the jth ad element, represents the optimization suggestion for the jth ad element. is the content modification coefficient;

[0214] S52. According to the output results of the multi-objective genetic algorithm, the advertisement delivery time is optimized, and the optimal delivery time window is calculated by the following formula:

[0215] ;

[0216] Where t represents time, , and They are the weights of click-through rate, viewing time and interaction data respectively;

[0217] S53. According to the output results of the multi-objective genetic algorithm, the target audience is accurately positioned, and the audience positioning parameters are calculated by the following formula:

[0218] ;

[0219] in, represents the characteristics of the i-th audience, represents the reaction of the i-th audience to the advertisement, represents the interaction data of the i-th audience, and N is the total number of audiences;

[0220] S54. Apply the modified advertising content, optimized delivery time and precisely positioned target audience to the actual advertising delivery strategy, form a dynamic adjustment mechanism, and continuously optimize the advertising delivery strategy based on real-time feedback.

[0221] This implementation method automatically adjusts the advertising delivery strategy according to the output results of the multi-objective genetic algorithm, including the modification of advertising content, optimization of delivery time and precise positioning of target audiences, thereby achieving dynamic adjustment and optimization of the advertising delivery strategy and ensuring the best execution effect of the advertising delivery strategy in the dynamic changes of the market.

[0222] S6. Use machine learning models to predict the market response after the adjusted advertising strategy is launched, and adjust the advertising strategy parameters in real time;

[0223] In this implementation, S6 includes the following steps:

[0224] S61. Construct a machine learning model. The machine learning model uses pre-processed user behavior data, market dynamics, and competitor advertising performance data as input to predict the market response to the adjusted advertising strategy. The machine learning model makes predictions using the following formula:

[0225] ;

[0226] Where y represents the predicted market reaction, , and are the weights of user behavior data, market dynamics, and competitor advertising performance, respectively, and f represents the prediction function of the machine learning model;

[0227] S62. Train the model, use historical advertising data and its actual market response as a training set, optimize the model parameters through the gradient descent algorithm, and calculate the loss function of the model through the following formula:

[0228] ;

[0229] Among them, L represents the loss function, N is the number of training set samples, represents the actual market reaction of the ith sample, represents the predicted market reaction of the i-th sample, and R(w) is the regularization term of the model parameters;

[0230] S63. Use the trained machine learning model to predict the market reaction to the adjusted advertising strategy, input the latest user behavior data, market dynamics and competitor advertising performance data, and the model outputs the predicted market reaction value;

[0231] S64. Monitor the actual market reaction in real time, compare the actual market reaction data with the market reaction value predicted by the model, and calculate the error using the following formula:

[0232] ;

[0233] Among them, e represents the error, M is the number of monitoring data samples, represents the actual market reaction of the jth sample, represents the predicted market reaction of the jth sample;

[0234] S65. Adjust the advertisement delivery strategy parameters in real time based on the error value, and optimize the parameters using the following formula:

[0235] ;

[0236] in, is the learning rate, is the partial derivative of the error with respect to the parameter, so that the advertising delivery strategy can maintain the best effect in the dynamic changes of the market.

[0237] This implementation method constructs a machine learning model, takes pre-processed user behavior data, market dynamics and competitor advertising performance data as input, predicts the market response of the adjusted advertising strategy, and adjusts the advertising strategy parameters in real time based on the error value, thereby achieving real-time optimization and dynamic adjustment of the advertising strategy, ensuring continuous optimization of the advertising strategy and optimal results.

[0238] S7. Re-enter the performance data after the advertisement is released, including the audience's purchasing behavior, user feedback and revisit rate, into the system, and use the optimized multi-objective genetic algorithm to continue performance evaluation and strategy adjustment, forming a closed-loop advertising effect automatic optimization system with dynamic learning and adaptive adjustment.

[0239] In summary, the present invention provides an efficient and intelligent method for tracking the real-time effects of advertising by combining multi-objective optimization algorithms, natural language processing, image recognition technology and machine learning models, which overcomes many shortcomings in the prior art, realizes comprehensive optimization and continuous improvement of advertising delivery strategies, and has significant practical value and market prospects.

[0240] 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. A method for tracking the real-time effect of advertising based on a multi-objective optimization algorithm, characterized in that: The following steps are involved: S1. Through data collection modules deployed on multiple digital platforms, real-time capture of user behavior data related to advertising, including user click-through rate, viewing time, audience response and interaction data, as well as collection of market dynamics and competitor advertising performance data from third-party data providers; S2. Format, clean and preliminarily analyze the collected user behavior data, market dynamics and competitor advertising performance data; S3. Use natural language processing and image recognition technology to extract key features including sentiment, visual elements and topic classification from advertising content, and use statistical methods to screen out key data that has a decisive impact on advertising effectiveness evaluation; S4. Use a multi-objective genetic algorithm to design an advertising performance optimization model. The advertising performance optimization model combines pre-processed user behavior data, market dynamics, and competitor advertising performance data, uses Pareto frontier analysis to determine the trade-off relationship between different advertising performance indicators, and finds the optimal trade-off point between various indicators through iterative update solutions; S5. Automatically adjust the advertising delivery strategy based on the output results of the multi-objective genetic algorithm. The advertising delivery strategy includes modification of advertising content, optimization of delivery time, and precise positioning of target audiences. S6. Use machine learning models to predict the market response after the adjusted advertising strategy is launched, and adjust the advertising strategy parameters in real time; S7. Re-enter the performance data after the advertisement is released, including the audience's purchase behavior, user feedback and revisit rate, into the system, and continue to perform performance evaluation and strategy adjustment using the optimized multi-objective genetic algorithm to form a closed-loop advertising effect automatic optimization system with dynamic learning and adaptive adjustment; The S3 comprises the following steps: S31, extracting emotional tendency features from the advertisement content using natural language processing technology, wherein the emotional tendency features include emotional score, emotional intensity and emotional change rate: Among them, E i represents the sentiment score of the i-th advertisement, w j Represents the weight of sentiment words, sentiment word score ij represents the score of the jth sentiment word in the i-th advertisement; Among them, I i Indicates the emotional intensity of the i-th advertisement, the emotional intensity word score ik represents the score of the kth emotional intensity word in the ith advertisement, and n represents the total number of emotional intensity words; Among them, C i Indicates the sentiment change rate of the i-th advertisement, sentiment score ij represents the score of the jth sentiment word in the i-th advertisement, sentiment score ij-1 represents the score of the previous sentiment word in the i-th advertisement; S32. Extract visual element features from the advertisement content using image recognition technology, wherein the visual element features include color distribution, image texture and shape features: Among them, D i Represents the color distribution of the i-th advertisement, color value il represents the color value of the lth pixel in the ith advertisement, and p represents the total number of pixels; Among them, T i Represents the image texture and texture features of the i-th advertisement im represents the value of the mth texture feature in the i-th advertisement, and q represents the total number of texture features; Among them, H i Represents the shape characteristics and shape area of ​​the i-th advertisement io represents the area of ​​the oth shape in the ith advertisement, the total shape area i Represents the total area of ​​all shapes in the i-th advertisement, the perimeter of the shape io represents the perimeter of the oth shape in the ith ad; S33, extracting topic classification features from the advertisement content, classifying the advertisement text through a text classification algorithm, and extracting the main topic categories: Among them, F i The main topic score and topic word score of the i-th advertisement are shown in it represents the score of the tth topic word in the ith advertisement, and s represents the total number of topic words; S34. Screen the emotional tendency features, visual element features and topic classification features through correlation analysis to select key data that have a decisive influence on the evaluation of advertising effects; The S34 comprises the following steps: S341, using nonlinear correlation analysis method to perform correlation analysis on sentiment tendency features, sentiment tendency features include sentiment score E i 、Emotional Intensity I i and the emotion change rate C i ,The correlation between the sentiment tendency characteristics and the advertising effect evaluation results is calculated by the following formula: Among them, α, β, and γ are the weights of sentiment score, sentiment intensity, and sentiment change rate, respectively. and are the means of the sentiment tendency feature set and the effect evaluation results, respectively. The nonlinear term in the formula Used to capture the interrelationships of sentiment tendency features; S342, using high-dimensional correlation analysis method to perform correlation analysis on visual element features, the visual element features include color distribution D i , image texture T i and shape feature H i ,The correlation between the visual element features and the advertising effect evaluation results is calculated by the following formula: Among them, δ, ∈, ζ are the weights of color distribution, image texture and shape features respectively. and are the mean of the visual element feature set and the effect evaluation result, respectively. The high-dimensional term in the formula Used to capture the interrelationships of visual element features; S343, using multidimensional correlation analysis method to conduct correlation analysis on topic classification features, the topic classification features include the main topic score F i ,The correlation between the topic classification features and the advertising effect evaluation results is calculated by the following formula: in, and are the mean of the topic classification feature set and the effect evaluation result, respectively. The multidimensional term in the formula Used to capture the interrelationships of topic classification features; S344, conduct correlation analysis on comprehensive emotional tendency features, visual element features and topic classification features, and calculate the correlation between the comprehensive features and the advertising effect evaluation results; The S4 comprises the following steps: S41. Design an advertising performance optimization model using a multi-objective genetic algorithm, which combines pre-processed user behavior data, market dynamics, and competitor advertising performance data; S42, initialize the population, generate an initial population containing multiple individuals, each individual represents an advertising strategy, and the individual is composed of a feature vector x of user behavior data, market dynamics and competitor advertising performance data i constitute; S43. Define a fitness function, taking the cost efficiency, audience coverage and user engagement of advertising performance indicators as targets, and calculate the fitness function by the following formula: Fitness Among them, x i represents the characteristic vector of the i-th individual, w1, w2 and w3 are the weights of advertising performance indicators, α1, α2, α3 and α4 are the weights of click-through rate, viewing time, audience response and interactive data, MT represents market dynamics, SI represents search index, BF represents volatility factor, IG represents industry growth rate, EI represents economic index, MS represents market share, CC represents competitor clicks, CD represents competitor impressions, AI represents advertising influence, and AF represents advertising cost; S44. Use Pareto frontier analysis to determine the trade-offs between different advertising performance indicators, construct a Pareto frontier set, and find the optimal balance point between multiple advertising performance indicators: Pareto front Among them, P is the population, f is the fitness function; S45, selection operation, selects excellent individuals in the population based on the fitness function value. The selection probability is proportional to the fitness value of the individual. The selection probability is calculated by the following formula: Among them, x i represents the i-th individual, N represents the total number of the population, and γ is the selection pressure parameter; S46, crossover operation, generating new individuals by performing crossover operation on the selected individuals, the crossover operation is performed by the following formula: ij =λ·parent1 ij +(1-λ)·parent2 ij +ξ·(Market Dynamics ij +Competitor performance ij ); Among them, the offspring ij represents the jth gene of the offspring individual, the parent 1 ij and parent 2 ij They represent the j-th gene of the parent individual, λ is the crossover coefficient, and ξ is the fusion coefficient of market and competitor data; S47, mutation operation, mutation operation is performed on the newly generated individuals to increase the diversity of the population. The mutation operation is performed using the following formula: Mutant offspring ij = offspring ij +μ·(random number-0.5)·(user behavior data ij +Market Dynamics ij ); Among them, the mutant offspring ij represents the jth gene of the mutated offspring individual, μ is the coefficient of variation, and user behavior data includes click-through rate, viewing time, audience response and interaction data; S48, iterative update, through multiple iterative selection, crossover and mutation operations, gradually optimize the individuals in the population until the preset termination condition is met. The termination condition includes the maximum number of iterations or the population fitness value reaches stability. The iterative update is performed by the following formula: Where k represents the number of iterations, represents the i-th individual in the k-th iteration, η is the fusion coefficient; S49, output the optimal solution, and select the individual with the highest fitness value from the final population as the optimal advertising delivery strategy; The S5 comprises the following steps: S51. According to the output result of the multi-objective genetic algorithm, determine the advertisement content that needs to be modified, including advertisement text, image and video elements. The modification of the advertisement content is performed by the following formula: Modified content ij =λ·Original content ij +(1-λ) Optimization suggestions ij ; Among them, the modified content ij Indicates the modified content of the jth ad element. The original content ij Indicates the original content of the jth ad element, optimization suggestions ij represents the optimization suggestion for the jth advertising element, and λ is the content modification coefficient; S52. According to the output results of the multi-objective genetic algorithm, the advertisement delivery time is optimized, and the optimal delivery time window is calculated by the following formula: Optimal delivery time = argmax t (α·click rate (t) + β·watching time (t) + γ·interaction data (t)); Among them, t represents time, α, β and γ are the weights of click-through rate, viewing time and interactive data respectively; S53. According to the output results of the multi-objective genetic algorithm, the target audience is accurately positioned, and the audience positioning parameters are calculated by the following formula: Among them, audience characteristics i represents the characteristics of the i-th audience, the audience response i represents the reaction of the i-th audience to the advertisement, audience interaction i represents the interaction data of the i-th audience, and N is the total number of audiences; S54, applying the modified advertising content, optimized delivery time and precisely positioned target audience to the actual advertising delivery strategy, forming a dynamic adjustment mechanism, and continuously optimizing the advertising delivery strategy based on real-time feedback; The S6 comprises the following steps: S61. Construct a machine learning model. The machine learning model uses pre-processed user behavior data, market dynamics, and competitor advertising performance data as input to predict the market response to the adjusted advertising strategy. The machine learning model makes predictions using the following formula: Predicting reactions i (y) = f(α·user behavior data + β·market dynamics + γ·competitor advertising performance); Where y represents the predicted market response, α, β, and γ are the weights of user behavior data, market dynamics, and competitor advertising performance, respectively, and f represents the prediction function of the machine learning model; S62. Train the model, use historical advertising data and its actual market response as a training set, optimize the model parameters through the gradient descent algorithm, and calculate the loss function of the model through the following formula: Among them, L represents the loss function, N is the number of training set samples, and the actual response i represents the actual market reaction of the ith sample and the predicted reaction i represents the predicted market reaction of the i-th sample, and R(w) is the regularization term of the model parameters; S63. Use the trained machine learning model to predict the market reaction to the adjusted advertising strategy, input the latest user behavior data, market dynamics and competitor advertising performance data, and the model outputs the predicted market reaction value; S64. Monitor the actual market reaction in real time, compare the actual market reaction data with the market reaction value predicted by the model, and calculate the error using the following formula: Among them, e represents the error, M is the number of monitoring data samples, and the actual response j represents the actual market reaction of the jth sample and the predicted reaction j represents the predicted market reaction of the jth sample; S65. Adjust the advertisement delivery strategy parameters in real time based on the error value, and optimize the parameters using the following formula: Where η is the learning rate, is the partial derivative of the error with respect to the parameter, so that the advertising delivery strategy can maintain the best effect in the dynamic changes of the market.

2. The method for tracking the real-time effect of advertisement delivery based on a multi-objective optimization algorithm according to claim 1, characterized in that: The S1 comprises the following steps: S11. Deploy a data collection module on multiple digital platforms, including a social media platform, a search engine, and a video platform, wherein the data collection module is capable of capturing user behavior data in real time; S12, the captured user behavior data includes user click-through rate CTR, viewing time WT, audience response AR and interaction data ID; S13. Collect market dynamics data and competitor advertising performance data from third-party data providers, including market trends and competitor advertising effectiveness.

Images