Advertisement optimized delivery system and delivery method

Abstract

The invention discloses an advertisement optimized putting system and putting method, and the method comprises the following steps: S10: obtaining user behavior data, the user behavior data comprising historical interaction information and real-time environment data of a user; s20, analyzing the historical interaction information and the real-time environment data of the user, and identifying user features and user interests; s30, based on a combined algorithm of a convolutional neural network and a generative adversarial network, dynamically generating an optimal advertisement content and an optimal advertisement format matched with the user features and the user interests; and S40, performing real-time advertisement putting on a target advertisement platform by using the optimal advertisement content and the optimal advertisement format. According to the method provided by the embodiment of the invention, the personalized advertisement content and format are dynamically generated by acquiring and analyzing the user behavior data, and the advertisement is put on the target platform in real time, so that the accuracy and effect of the advertisement are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet advertising, and in particular to an advertisement optimization delivery system and delivery method. Background Art

[0002] With the rapid development of the Internet and mobile technology, the online advertising market has continued to expand and become an indispensable part of corporate marketing strategies. However, traditional advertising methods often have problems such as insufficient accuracy and poor user experience.

[0003] In the prior art, the content and format of advertisements are usually pre-set and difficult to dynamically adjust according to the real-time status and interests of users. This static advertising delivery method cannot adapt to the rapid changes in user needs, nor can it fully utilize real-time data to optimize advertising effects. In addition, most advertising systems lack in-depth analysis and utilization of user behavior data, and only rely on simple demographic characteristics or keyword matching to deliver advertisements. This method often ignores the personalized needs and potential interests of users.

[0004] On the other hand, with the development of technology, some advanced advertising systems have begun to try to use machine learning algorithms to optimize advertising. However, these systems still have shortcomings in real-time, personalization, and creative generation. In particular, in terms of dynamically generating advertising content and formats, existing technologies cannot fully utilize the potential of deep learning models. Therefore, how to effectively integrate these advanced technologies to improve the personalization and delivery accuracy of advertising has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, an object of the present invention is to provide an advertisement optimization delivery system and delivery method to overcome the deficiencies in the prior art.

[0006] To achieve the above objectives, in a first aspect, an embodiment of the present invention provides an advertisement optimization delivery method, which comprises the following steps: Acquire user behavior data, wherein the user behavior data includes historical interaction information and real-time environment data of the user; Analyze the user's historical interaction information and real-time environment data to identify user characteristics and user interests; Based on the combined algorithm of convolutional neural network and generative adversarial network, the optimal advertising content and format that matches user characteristics and interests are dynamically generated; The optimal advertisement content and advertisement format are used to deliver advertisements in real time on the target advertisement platform.

[0007] In a second aspect, a system for optimizing advertisement delivery is provided, the system comprising: A data collection module is used to obtain user behavior data, wherein the user behavior data includes the user's historical interaction information and real-time environment data; A data processing module, used to analyze the user's historical interaction information and real-time environment data, and identify user characteristics and user interests; Advertisement generation module, based on the combined algorithm of convolutional neural network and generative adversarial network, dynamically generates the optimal advertisement content and advertisement format that matches the user characteristics and interests; The advertisement delivery module is used to deliver advertisements in real time on a target advertisement platform using the optimal advertisement content and advertisement format.

[0008] The above technical solution has the following beneficial technical effects: This method accurately identifies user characteristics and interests by acquiring and analyzing user historical interaction information and real-time environmental data in real time, and dynamically generates optimal advertising content and format using a combination of convolutional neural networks and generative adversarial networks. This method can improve the personalization and delivery accuracy of advertising, thereby improving advertising effectiveness, user experience, and bringing advertisers a higher return on investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention. Figure 1 is a flow chart of an advertisement optimization delivery method according to an embodiment of the present invention; Figure 2 is a flowchart of step S10 of an embodiment of the present invention; Figure 3 is a flowchart of step S20 of an embodiment of the present invention; Figure 4 is a flowchart of step S30 of an embodiment of the present invention; Figure 5 is a flowchart of step S33 of an embodiment of the present invention; Figure 6 is a flowchart of step S34 of an embodiment of the present invention; Figure 7 is a flowchart of step S40 of an embodiment of the present invention; Figure 8 is a flowchart of step S50 of an embodiment of the present invention; Fig. 9 It is a functional block diagram of an advertisement optimization delivery system according to an embodiment of the present invention; Fig.10 It is a schematic diagram of the structure of a computer system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0011] The embodiments of the present invention obtain and analyze user behavior data, dynamically generate personalized advertising content and format, and deliver them in real time on a target platform, thereby improving the accuracy and effectiveness of advertising.

[0012] like Figure 1 As shown, an embodiment of the present invention provides an advertisement optimization delivery method, which includes the following steps: S10: acquiring user behavior data, wherein the user behavior data includes the user's historical interaction information and real-time environment data; S20: analyzing the user's historical interaction information and real-time environment data to identify user characteristics and user interests; S30: dynamically generating optimal advertisement content and advertisement format that matches the user characteristics and user interests based on a combined algorithm of a convolutional neural network and a generative adversarial network; S40: performing real-time advertisement delivery on a target advertisement platform using the optimal advertisement content and advertisement format.

[0013] In the specific implementation process, the user behavior data obtained in step S10 may include historical interactive information such as the user's browsing history, click records, purchase behavior, social media activities, and real-time environmental data such as the user's current geographic location, time, weather, and device type. In step S20, the system uses machine learning algorithms, such as collaborative filtering and cluster analysis, to conduct in-depth analysis of these data to identify the user's hobbies, consumption tendencies, and other characteristics. Step S30 uses a combination algorithm of convolutional neural networks (CNN) and generative adversarial networks (GAN). CNN is used to extract deep representations of user characteristics and interests, while GAN is used to generate advertising content and formats that match them. For example, for a user who often browses sports-related content, the system will generate a dynamic sports shoe advertisement, and the shoe style and color in the advertisement will be dynamically adjusted according to the user's preferences. Finally, in step S40, the system delivers the generated advertising content to the target platform in real time, such as social media, search engines, or mobile applications.

[0014] The technical effects of the method of the embodiment of the present invention are as follows: First, by analyzing user behavior data in real time, the user's immediate needs and interests can be accurately grasped, greatly improving the relevance and accuracy of advertisements. Secondly, by utilizing the combined algorithm of CNN and GAN, not only can highly personalized advertising content be generated, but also the creativity and attractiveness of the advertisements can be ensured. Thirdly, the dynamically generated advertising content and format can adapt to different delivery platforms and user devices, improving the adaptability of the advertisements and user experience. Finally, this method realizes the intelligence of the entire process of advertising delivery, from data collection and analysis to content generation and delivery, which can be completed automatically, greatly improving the efficiency and effectiveness of advertising delivery.

[0015] In an alternative embodiment, step S30 can be replaced by: dynamically generating optimal advertising content and advertising format that matches user characteristics and user interests based on a convolutional neural network, a content matching algorithm, a resource library, and a multimodal content generation method. The processing includes: Obtain user feature data and user interest data. Use convolutional neural networks to extract high-dimensional feature representations from these data, namely user feature vectors. This user feature vector contains various characteristics of the user, such as consumption habit tags or interest tags. At the same time, convert the attributes of the advertising content into vector form to obtain the advertising content vector. These user feature vectors and advertising content vectors will be used in the subsequent content matching and personalized generation process.

[0016] The similarity between the user feature vector and the ad content vector is calculated using a content matching algorithm based on cosine similarity. From the pre-prepared ad content library (as part of the resource library), the ad content with the highest similarity is selected as the basis, which will be used in the subsequent personalized generation process.

[0017] Then, appropriate elements are selected from the resource library for advertisement generation. The resource library includes sub-libraries such as template library, image library, audio library, and video clip library, which together constitute the source of materials for advertisement content. The most suitable advertisement template is selected according to the user's feature vector through the rule engine. Various advertisement elements are dynamically generated using multimodal content generation methods: the copy can use RNN-based language model, simplified Transformer model or template filling combined with natural language processing (NLP) technology; the image can be selected from the image sub-library of the resource library, generated using a simplified conditional generative adversarial network (CGAN) or applying image splicing technology; the audio can use TTS (Text-to-Speech) technology, select matching background music from the audio sub-library of the resource library or apply simple audio mixing technology; the video can be selected and combined from the video clip sub-library of the resource library based on the user's feature vector, use dynamic template technology or apply simple video transitions and special effects; the advertisement format is selected according to the user's device, browsing habits and platform characteristics.

[0018] Finally, the generated ad content is combined with the selected ad format to form a complete personalized ad. This process makes full use of the user feature vectors extracted in the previous steps and the matched ad content to ensure that the generated ad is highly relevant to the user's features and interests.

[0019] like Figure 2 As shown, further, step S10 specifically includes the following sub-steps: S11: obtaining the user's historical interaction information, which includes the user's advertising browsing history, advertising click record and advertising purchase record; S12: obtaining the user's real-time environmental data, which includes the user's geographic location data, device type data and usage period data; S13: processing the historical interaction information and the real-time environmental data to obtain processed user behavior data.

[0020] In sub-step S11, the system collects historical interaction information of users through multiple channels. Ad browsing history records may include information such as the types of ads, ad content, and dwell time that users have browsed on various platforms. For example, the system records that users often browse ads for sports shoes and fitness equipment, and the average dwell time is longer. Ad click records record the details of ads that users have clicked, including the category of the ad, product information, and click time. Ad purchase records further record the actual purchase behavior of users through ads, and these data are particularly important for analyzing users' purchasing intentions and spending power.

[0021] In sub-step S12, the system captures the user's environmental data in real time. Geographic location data can be obtained through the GPS or IP address of the user's device, which helps to deliver localized advertising content. For example, if it is detected that the user is near a commercial area, the system will give priority to delivering promotional ads for nearby stores. Device type data includes whether the user is using a mobile phone, tablet or desktop computer, which affects the display format and size of the advertisement. Usage time period data records the time period when the user is active, such as weekday evenings or weekend mornings, which helps to deliver advertisements at the best time.

[0022] In sub-step S13, the system pre-processes and integrates the collected historical interaction information and real-time environment data. This process may include steps such as data cleaning, standardization, and feature extraction. For example, the system will unify the data from different sources into a unified format, remove outliers, and extract key features such as user interest tags, activity index, etc. The processed data will form a comprehensive user behavior portrait, providing a basis for subsequent analysis and advertising generation.

[0023] This multi-dimensional, real-time data collection and processing method greatly improves the accuracy and timeliness of advertising.

[0024] like Figure 3 As shown, further, step S20 specifically includes the following sub-steps: S21: performing time series analysis on the processed user behavior data, arranging the processed user behavior data in chronological order, and dividing the processed user behavior data into a recent time window, a mid-term time window, and a long-term time window; S22: within each time window, statistically analyzing the user's historical interaction information, identifying the user's interest category within the time window based on the user's ad browsing history, ad click record, and ad purchase record, and generating an interest tag; S23: Combine the user's real-time environmental data and historical interaction information, use the decision tree algorithm to extract and classify features, and obtain the user's feature labels, which include age group labels, gender labels, common device type labels, and active time period labels.

[0025] In sub-step S21, the system first arranges the user behavior data in chronological order, so that the evolution of user interests and behaviors can be clearly shown. Then, the system divides the timeline into three windows: a recent time window (such as the last week), a medium-term time window (such as the last month), and a long-term time window (such as the last three months or longer). This division method allows the system to consider the user's immediate interests, recent trends, and long-term preferences at the same time. For example, a user frequently browsed smartwatch ads in the past week (recent interest), but mainly focused on fitness equipment in the past month (medium-term interest), and in the long run, the user has maintained a high interest in technology products (long-term preference).

[0026] In sub-step S22, the system performs in-depth statistical analysis on the user's historical interaction information within each time window. For the ad browsing history records, the system calculates the number of views and duration of different categories of ads, and identifies the types of ads that the user is most interested in. The analysis of ad click records includes calculating the click-through rate of each type of ad, as well as the dwell time after clicking, which helps to judge the user's deep interest. Ad purchase records directly reflect the user's consumption tendency and purchasing power. By combining these analyses, the system generates a set of interest tags for each time window. For example, the near-term time window generates a "smart watch enthusiast" tag, the medium-term time window generates a "fitness enthusiast" tag, and the long-term time window generates a "technology enthusiast" tag.

[0027] In sub-step S23, the system uses a decision tree algorithm to extract and classify user features. The decision tree algorithm was selected because it can handle multiple types of data (including categorical data and numerical data) and has good interpretability. The system uses the user's real-time environmental data (such as geographic location, the type of device currently used) and historical interaction information (such as purchase behavior, browsing mode) as input features. Through the branches and nodes of the decision tree, the system can gradually infer various characteristic labels of the user. For example, based on the user's frequent browsing and purchasing behavior at night, the system will label the user as a "night active user"; based on the content browsed by the user and the type of goods purchased, the system will infer the user's age group and gender, such as "25-35 years old" and "female" labels; based on the user's most commonly used access device, the system will generate a "mobile user" label.

[0028] Through the above sub-steps, time series analysis ensures that the system can capture the dynamic changes of user interests, and the decision tree algorithm helps the system extract key user features from complex data.

[0029] like Figure 4 As shown, further, step S30 specifically includes: S31: Analyze the user characteristics and the user interests using a convolutional neural network to extract key feature elements; In this step, the system uses a convolutional neural network (CNN) to process user characteristics and interest data. The multi-layer structure of CNN can effectively capture complex patterns and high-level features in the data. For example, for user interest data, CNN will identify deep-level features such as user preferences for specific brands and seasonal demand for certain types of products. For user characteristic data, CNN will extract abstract features such as user spending power and lifestyle. These extracted key feature elements will provide an important basis for subsequent advertising generation.

[0030] S32: Based on the key feature elements, create a plurality of candidate advertising contents and advertising formats using a generative adversarial network architecture including a generator network and a discriminator network; wherein the generator network is used to generate the candidate advertising contents and advertising formats, and the discriminator network is used to evaluate the candidate advertising contents and advertising formats; In this step, the system uses a Generative Adversarial Network (GAN) architecture to create personalized ads. The generator network receives key feature elements as input and then generates a variety of ad content and formats. For example, for a user identified as a "sports enthusiast" who frequently uses a mobile device, the generator will create a series of vertical image ads showing sports shoes. At the same time, the discriminator network is responsible for evaluating whether these generated ads are consistent with user characteristics and interests, and whether they are attractive and persuasive enough.

[0031] The working principle of the generator network to generate candidate advertising content and advertising formats based on key feature elements is as follows: The generator network adopts a multimodal generation method and uses dedicated generation models for different types of advertising content. For advertising text, it uses a language model based on RNN or Transformer, takes key feature elements as input conditions, and generates advertising copy that matches user characteristics and interests. Advertising images are generated through conditional generative adversarial networks (CGANs), which use key feature elements as conditions to create visual content that matches user preferences. Advertising audio uses text-to-speech (TTS) technology to convert the generated copy into speech, and selects appropriate background music for mixing based on feature elements. Advertising videos select and combine relevant clips from a pre-prepared video clip library based on user characteristics, and apply transition effects and filters that match the feature elements. Throughout the process, the generator network continuously refers to key feature elements to ensure that the generated advertising content of various types is highly relevant to user characteristics in terms of theme, style and format, thereby improving the personalization and potential effectiveness of advertising.

[0032] S33: inputting the candidate advertisement content and advertisement format into the discriminator network, evaluating the matching degree between the candidate advertisement content and advertisement format and the key feature elements, and obtaining a matching score; In this step, the discriminator network evaluates each generated candidate ad. It considers whether the ad content matches the user's interests (for example, whether the product category displayed in the ad matches the user's browsing history), whether the ad format is suitable for the user's usage habits (for example, whether it is suitable for display on the user's commonly used devices), and the overall quality and attractiveness of the ad. The discriminator gives each candidate ad a matching score, which reflects the degree of fit between the ad and the user's characteristics.

[0033] S34: Optimizing the generator network according to the matching score; Based on the matching scores given by the discriminator, the system will adjust and optimize the generator network. If certain types of advertising content or formats receive higher matching scores, the generator will tend to generate more similar ads. Conversely, if certain ads perform poorly, the generator will reduce the probability of generating such ads. This process is actually a back-propagation process. Through continuous adjustment and learning, the generator can gradually improve its ability to generate high-quality, high-matching ads.

[0034] S35: Repeat steps S32 to S34 until the generated candidate advertisement content and advertisement format reach a preset quality threshold; This step reflects the iterative optimization process of GAN. The system will continue to repeat the cycle of generation, evaluation, and optimization until the generated ads meet the preset quality standards. This quality threshold includes multiple dimensions, such as the minimum requirement for matching scores, the diversity index of ad content, the creative novelty score, etc. Through this continuous iteration and improvement, the system can continuously improve the quality and relevance of generated ads.

[0035] S36: Selecting the best advertisement content and advertisement format from the candidate advertisement content and advertisement format that meet a preset quality threshold according to the matching score.

[0036] In this final step, the system selects the ad with the highest match score from all candidate ads that meet the quality threshold as the final output. This selection process also takes into account other factors, such as the novelty of the ad (to avoid showing similar ads to users repeatedly), the current marketing strategy, etc. The final selected ad should not only be highly matched to the user's characteristics and interests, but also have the potential to attract user attention and promote conversions.

[0037] In some embodiments, step S36 may specifically include the following steps: setting a preset quality threshold, which is determined based on historical advertising performance data and current market trends; screening all candidate advertising content and advertising formats, and eliminating options with matching scores lower than the preset quality threshold; sorting the remaining candidate advertising content and advertising formats, and arranging them from high to low according to the matching scores; selecting the candidate advertising content and advertising formats that are in the top N positions after sorting, where N is a predetermined value; performing a multi-dimensional evaluation on the selected N candidate items, including but not limited to expected click-through rate, expected conversion rate, budget fit, etc.; based on the multi-dimensional evaluation results, using a weighted scoring method to calculate the comprehensive score of each candidate item; selecting the candidate item with the highest comprehensive score as the optimal advertising content and advertising format.

[0038] Through the above steps, the system is able to generate highly personalized and high-quality advertising content. This method combines the powerful analytical capabilities of deep learning and the creativity of generative adversarial networks, and can continuously learn and adapt to changes in user preferences, thereby continuously providing the most relevant and effective advertising content.

[0039] like Figure 5 As shown, further, step S33 specifically includes the following sub-steps: S331: Converting the candidate advertisement content and advertisement format into standardized candidate advertisement content and advertisement format; Specifically, in this sub-step, the system standardizes the generated candidate ad content and ad formats. The purpose of this process is to ensure that all candidate ads can be compared and evaluated on the same basis. Standardization includes the following aspects: word vector conversion of text content, converting text information into numerical vectors that can be processed by neural networks; resizing and color standardization of image content to ensure that all images have the same resolution and color space; encoding ad formats, such as converting different ad formats (banners, pop-ups, videos, etc.) into one-hot encoding. Through this standardization process, the system can ensure that ads of different types and formats can be fairly evaluated by the discriminator network.

[0040] S332: Inputting the standardized candidate advertisement content and advertisement format, as well as the key feature elements, into the discriminator network; Specifically, in this sub-step, the system inputs the standardized candidate ads and the previously extracted key feature elements into the discriminator network. The key feature elements here include the user's interest tags, demographic characteristics, behavioral patterns, etc. Inputting this information together allows the discriminator to consider both the characteristics of the ad itself and the characteristics of the target user, thereby making a more comprehensive and accurate evaluation. For example, for a user labeled as a "technology enthusiast", the discriminator will pay special attention to whether the ad contains the latest technology product information.

[0041] S333: using a discriminator network to calculate the similarity between the standardized candidate advertisement content and advertisement format and the key feature elements; Specifically, in this sub-step, the discriminator network will conduct an in-depth analysis of the input advertising content, format, and user features, and calculate the similarity or matching degree between them. This process involves comparisons in multiple dimensions: content relevance, which indicates whether the advertising content is highly relevant to the user's interest tags; format adaptability, which indicates whether the advertising format is suitable for the user's device type and usage habits; time relevance, which indicates whether the advertisement is in line with the user's current time period and scenario; historical behavior consistency, which indicates whether the advertisement is consistent with the user's historical interaction behavior. The discriminator network will use methods such as cosine similarity and Euclidean distance to quantify the matching degree of these dimensions.

[0042] S334: Generate a matching score based on the calculated similarity; In this sub-step, the system combines the similarities of each dimension calculated in the previous step to generate an overall matching score. This process involves weighted average of different dimensions. For example, content relevance is given a higher weight because it directly affects the user's interest; while format adaptability is given a lower weight because it mainly affects the user experience but does not directly determine whether the user is interested in the ad. The final matching score should be a value that comprehensively reflects the degree of matching between the ad and the user, ranging from 0 to 1, where 1 indicates a perfect match.

[0043] S335: Repeat steps S331 to S334 for all candidate advertisement contents and advertisement formats to obtain a set of matching scores.

[0044] In this substep, the system performs the above evaluation process on all candidate ad content and formats created by the generator network. Each candidate ad is given a matching score. This set of scores is not only used to select the best ads, but also provides important feedback information for the optimization of the generator network. For example, if a certain type of ad content or format is found to consistently receive high scores, this information will be used to guide the generator to generate more similar ads in the future.

[0045] Through the above steps, the system can comprehensively and deeply analyze the degree of match between each candidate ad and the target user. This method not only takes into account the quality of the ad itself, but also makes full use of the user's personal characteristics and preferences, thereby greatly improving the accuracy and effectiveness of personalized ads. The iterative and adaptive nature of this process enables the system to continuously learn and improve, generating more and more accurate and effective personalized ads over time.

[0046] like Figure 6 As shown, further, step S34 specifically includes the following sub-steps: S341: Calculate the loss function of the generator network based on the matching score. In this sub-step, the system uses the matching score given by the discriminator network to calculate the loss function of the generator network. The loss function is a key indicator to measure the performance of the generator network, which reflects the gap between the generated advertising content and the ideal result. Specifically, the loss function takes the form of L = -log(D(G(z))), where G(z) represents the advertising content generated by the generator network, and D(G(z)) represents the matching score given by the discriminator network to the advertising content. By minimizing this loss function, the goal is to allow the generator to generate advertising content that can obtain a high matching score.

[0047] S342: Using the back-propagation algorithm, calculate the gradient of the parameters of each layer of the generator network based on the loss function. In this sub-step, the system uses the back-propagation algorithm to calculate the gradient of the loss function on each parameter in the generator network. The back-propagation algorithm starts from the output layer and calculates the influence of each parameter on the loss function layer by layer. This process can be expressed as ∂L / ∂θ = ∂L / ∂y * ∂y / ∂θ, where L is the loss function, y is the output of the network, and θ is the parameter of the network. Using the chain rule, the gradient of each parameter in each layer of the network can be calculated. These gradient information indicate how to adjust the parameters to reduce the value of the loss function.

[0048] S343: Using the stochastic gradient descent method, based on the gradient calculated in step S342, update the parameters of the generator network, including the weights and biases of each layer. In this sub-step, the system uses the stochastic gradient descent (SGD) method to update the parameters of the generator network. The basic update rule of SGD is θ_new = θ_old - η * ∂L / ∂θ, where η is the learning rate, which determines the step size of each update. In practical applications, variants of SGD can be used, such as Adam or RMSprop, which optimizers can adaptively adjust the learning rate to accelerate the convergence process. In this way, the weights and biases of the generator network are gradually adjusted to generate advertising content that is more in line with the target.

[0049] S344: Repeat steps S341 to S343 until the performance of the generator network reaches a preset threshold or the number of iterations reaches a predetermined value. This sub-step reflects the iterative nature of generator network optimization. The system will continue to repeat the process of loss calculation, gradient calculation, and parameter update until the stopping condition is met. There are two types of stopping conditions: performance threshold (for example, stop when the average matching score of the generated advertising content reaches 0.9 or above) and number of iterations (for example, stop after 10,000 iterations). This process ensures that the generator network can fully learn and optimize, while also preventing the risk of overfitting. In each iteration, the system monitors the performance of the generator, including indicators such as the quality and diversity of the generated ads, to ensure that the optimization is moving in the right direction.

[0050] S345: Save the optimized parameters of the generator network. In this last sub-step, the system will save the optimized parameters of the generator network. This includes the weights and bias values ​​of all layers in the network. These parameters represent the knowledge and capabilities learned by the generator network. These parameters can be used for rapid deployment, continuous learning, model analysis, and version control. By saving these optimized parameters, the system can quickly generate high-quality personalized ads in the future.

[0051] like Figure 7 As shown, further, step S40 specifically includes the following sub-steps: S41: Based on the user's real-time environmental data, determine the corresponding target advertising platform for advertising delivery. Specifically, the system will analyze the user's real-time environmental data, such as the type of device currently used by the user (smartphone, tablet or desktop computer), the geographical location, the current time and other factors. Based on these data, the system will select the most suitable advertising platform for delivery. For example, if the user is using a mobile device and is located in a business district, the system will select a mobile social media platform or a location-based advertising platform for delivery. If the user is using a desktop computer and is at work, the system will select a professional social network or industry-related website for advertising delivery. This platform selection based on real-time data can increase the exposure and click-through rate of advertisements, thereby improving the overall advertising effect.

[0052] S42: Using a reinforcement learning algorithm, adjust the frequency and time interval of advertising on the target advertising platform in real time. This sub-step introduces a reinforcement learning algorithm to optimize the advertising delivery strategy. The reinforcement learning algorithm continuously tries different delivery frequencies and time intervals, and learns the optimal delivery strategy based on the results of each attempt (such as click-through rate, conversion rate, etc.). Specifically, the system divides each day into multiple time periods and assigns different delivery frequencies to each time period. Over time, the algorithm learns in which time periods increasing the delivery frequency can bring better results, and in which time periods reducing the delivery frequency can save costs. This process is continuous and dynamic, and can adapt to changes in user behavior patterns and changes in the market environment. In this way, the system can optimize the cost-effectiveness of advertising while ensuring advertising effectiveness.

[0053] S43: placing advertisements corresponding to the optimal advertisement content and advertisement format on the target advertisement platform, and adopting the A / B testing method to simultaneously place multiple fine-tuned versions of advertisement content and advertisement format to identify the best performing advertisement variant. Specifically, this sub-step is not just to simply place the optimal advertisement that has been determined, but to adopt a more dynamic and sophisticated method. The system will place multiple versions of advertisements at the same time, and these versions will differ in copywriting, pictures, layout or other details. Through A / B testing, the system can compare the performance of these different versions in real time, such as indicators such as click-through rate, conversion rate, and user stay time. This method allows the system to continuously optimize the advertisement content and quickly identify the most popular advertisement variants. For example, the system will test different titles, different button colors, or different picture layouts. By analyzing the results of these tests, the system can continuously adjust and improve the advertisement content to achieve the best advertising effect. This continuous optimization method ensures that the advertisement content always maintains the highest appeal and effectiveness and can adapt to changing user preferences and market trends.

[0054] Furthermore, the method further comprises the following steps: S50: Monitor the advertising effect in real time to obtain feedback data of the advertising effect; use the feedback data of the advertising effect to update the model parameters of the convolutional neural network and generative adversarial network combination algorithm to improve the accuracy of advertising generation and matching.

[0055] like Figure 8 As shown, further, step S50 specifically includes the following sub-steps: S51: Real-time monitoring of the delivery effects of all advertisements delivered on the target advertising platform, and obtaining feedback data on the delivery effects of the advertisements, which include: quantitative indicators, including click-through rate, conversion rate, user dwell time and user interaction depth; qualitative indicators, which are sentiment tendency keywords extracted from user comments using natural language processing technology. Specifically, quantitative indicators provide objective data support. For example, the click-through rate reflects the attractiveness of the advertisement, the conversion rate indicates the effectiveness of the advertisement, and the user dwell time and interaction depth reflect the user's interest in the content of the advertisement. At the same time, qualitative indicators capture the user's subjective feelings and emotional reactions by analyzing user comments. For example, the system will identify positive keywords such as "interesting", "practical", "innovative", or negative keywords such as "misleading".

[0056] S52: Construct a comprehensive evaluation model, integrate the quantitative indicators and the qualitative indicators, and generate a comprehensive score for advertising effectiveness. Specifically, in this sub-step, the system integrates indicators of different natures into a unified scoring system. This comprehensive evaluation model adopts a weighted average method, in which each indicator is assigned a certain weight. For example, the conversion rate is given a higher weight because it is directly related to the commercial value of the advertisement; and the weight of sentiment-oriented keywords is dynamically adjusted according to their frequency and intensity of occurrence. This model also contains some nonlinear transformations to better capture the interaction between different indicators. The final generated comprehensive score will become a key indicator for measuring the overall effectiveness of the advertisement, providing an important basis for subsequent optimization decisions.

[0057] S53: Based on the comprehensive score of advertising effect, the gradient boosting decision tree algorithm is used to obtain the feature importance ranking, so as to identify the key feature combinations that rank in the top 10% of the impact on advertising effect. Specifically, this sub-step uses machine learning technology to deeply analyze the key factors that affect advertising effect. Gradient boosting decision tree (such as XGBoost or LightGBM) is a powerful ensemble learning algorithm that can effectively handle complex nonlinear relationships and feature interactions. Through this algorithm, the system can evaluate the contribution of each feature to the advertising effect. For example, the system found that specific combinations of advertising color matching, copy length, image position, etc. are particularly effective in improving click-through rate. After identifying these key feature combinations, the system can optimize advertising design more specifically and focus resources on the most influential factors.

[0058] S54: Update the model parameters of the convolutional neural network, specifically including: adjusting the convolution kernel size and the number of convolution kernels of the convolution layer according to the identified key feature combination; adjusting the pooling window size of the pooling layer according to the generated comprehensive score of the advertising effect; and updating the number of neurons and activation function of the fully connected layer according to the obtained user interaction depth.

[0059] Specifically, this sub-step optimizes the structure of the convolutional neural network in a targeted manner. By adjusting the size and number of convolution kernels, the system can better capture key feature combinations. For example, if it is found that small-scale texture features have a greater impact on advertising effects, the number of small-sized convolution kernels will be increased. Adjusting the pooling window size can balance the accuracy and computational efficiency of feature extraction. Updating the fully connected layer can adjust the complexity of the model according to the depth of user interaction to better fit user behavior patterns. These adjustments ensure that the model can more accurately capture the key factors that affect advertising effects.

[0060] In one embodiment, the step S54 specifically includes: based on the identified key feature combination, calculating the spatial scale and complexity of each feature, wherein the spatial scale refers to the size range of the feature in the image, and the complexity refers to the texture detail level of the feature; according to the calculation results, determining the convolution kernel size range that is most suitable for capturing these key features to ensure that the convolution kernel can effectively extract the identified features; within the determined convolution kernel size range, testing different convolution kernel size combinations by a grid search method, and evaluating the feature extraction effect of each combination; selecting the convolution kernel size combination with the best performance, and adjusting the number of convolution kernels of the convolution layer accordingly according to the number and complexity of the features; using the comprehensive score of the advertising effect to construct an adaptive pooling mechanism, which dynamically adjusts the pooling operation according to the advertising effect The strength and mode of the pooling layer are determined; according to the constructed adaptive pooling mechanism, the pooling window size of the pooling layer is dynamically adjusted to strike a balance between the accuracy of feature extraction and computational efficiency; by analyzing the user's behavioral data such as clicks, slides, and stays on the advertisements, the user interaction depth is quantified, and a mapping relationship between the user interaction depth and the complexity of the neural network model is established; based on the established mapping relationship, the number of neurons required for the fully connected layer is determined to match the model complexity with the user interaction depth; different activation functions (such as ReLU, LeakyReLU, Swish, etc.) are compared experimentally to select the activation function that best suits the current user interaction; according to the results of the previous steps, the structure of the fully connected layer is updated, and the newly determined number of neurons and activation function are applied to optimize the model's responsiveness to user interaction.

[0061] S55: Update the model parameters of the generative adversarial network, specifically including: adjusting the upsampling layer parameters of the generator network according to the obtained user residence time; adjusting the convolution layer parameters of the discriminator network according to the extracted sentiment tendency keywords; updating the weight of the adversarial loss function according to the generated comprehensive score of the advertising effect, and balancing the training process of the generator network and the discriminator network.

[0062] Specifically, this sub-step optimizes the performance of the generative adversarial network. By adjusting the upsampling layer parameters, the generator can generate advertising content that is more likely to attract users' attention for a long time. Adjustments to the discriminator network enable it to more accurately identify user preferences. The update of the adversarial loss function weights ensures a balance between the generator and the discriminator, preventing one party from over-dominantly dominating the training process. For example, if the quality of the generated advertising content is found to be insufficient, the weight of the discriminator in the loss function is increased, thereby improving the standard of the generated content. This dynamic balancing approach can continuously improve the quality and relevance of ad generation.

[0063] In one embodiment, the step S55 specifically includes: analyzing the user's dwell time, establishing a relationship model between the user's dwell time and the richness of the content, wherein the richness of the content refers to the amount of information and visual complexity of the advertising content; determining the target output resolution of the upsampling layer of the generator network based on the relationship model; adjusting the upsampling layer parameters of the generator network according to the target resolution, including the upsampling ratio and the convolution kernel size; performing semantic clustering on the extracted sentiment tendency keywords to identify the main sentiment categories; designing convolution filters according to the identified sentiment categories to capture different types of sentiment features; integrating the designed convolution filters into the discriminant In the convolutional layer of the generator network, adjust the parameter configuration of the convolutional layer; based on the comprehensive score of advertising effect, construct a dynamic weight adjustment method, which dynamically calculates the weight adjustment coefficient by analyzing advertising effect indicators (click-through rate, conversion rate, etc.); using this dynamic weight adjustment method, calculate the relative performance of the generator network and the discriminator network in real time during the training process; according to the relative performance, dynamically adjust the weight ratio of the generator loss and the discriminator loss in the adversarial loss function, for example, increase its loss weight when the generator performance is weak, and vice versa; through iterative training, continuously fine-tune the weight of the adversarial loss function until the generator and discriminator networks reach a balanced state. In this process, the target output resolution is used to guide the adjustment of the upsampling layer parameters, the upsampling ratio and the convolution kernel size are directly applied to the structural optimization of the generator network, and the emotional features are used to enhance the recognition ability of the discriminator network, and jointly improve the overall performance of the generative adversarial network. Example

[0064] like Fig. 9As shown, this embodiment provides an advertisement optimization delivery system, which includes: a data acquisition module, used to obtain user behavior data, the user behavior data includes the user's historical interaction information and real-time environment data; a data processing module, used to analyze the user's historical interaction information and real-time environment data, and identify user characteristics and user interests; an advertisement generation module, based on a combination algorithm of a convolutional neural network and a generative adversarial network, dynamically generates optimal advertisement content and advertisement format that matches the user characteristics and user interests; an advertisement delivery module, used to use the optimal advertisement content and advertisement format to deliver real-time advertisements on a target advertisement platform; a central processing unit, used to control the operation of the above modules; a memory, used to store data and programs required for system operation.

[0065] In some embodiments, the data collection module includes: a historical data collection unit, used to obtain the user's historical interaction information, which includes the user's advertising browsing history records, advertising click records and advertising purchase records; a real-time data collection unit, used to obtain the user's real-time environmental data, which includes the user's geographic location data, device type data and usage period data; a data preprocessing unit, used to process the historical interaction information and the real-time environmental data to obtain processed user behavior data.

[0066] In some embodiments, the data processing module includes: a time series analysis unit, which is used to perform time series analysis on the processed user behavior data, arrange the processed user behavior data in chronological order, and divide it into a recent time window, a medium-term time window, and a long-term time window; an interest identification unit, which is used to perform statistical analysis on the user's historical interaction information in each time window, identify the user's interest category in the time window according to the user's ad browsing history, ad click record, and ad purchase record, and generate interest tags; a feature extraction unit, which is used to combine the user's real-time environmental data and historical interaction information, use a decision tree algorithm to perform feature extraction and classification, and obtain the user's feature tags, which include age group labels, gender labels, common device type labels, and active time period labels.

[0067] In some embodiments, the advertisement generation module includes: a feature analysis unit, which uses a convolutional neural network to analyze user features and user interests and extract key feature elements; a generator network, which creates multiple candidate advertisement contents and advertisement formats based on the key feature elements; a discriminator network, which is used to evaluate the degree of match between the candidate advertisement contents and advertisement formats and the key feature elements to obtain a match score; a network optimization unit, which optimizes the generator network according to the match score; and an advertisement selection unit, which selects the optimal advertisement content and advertisement format from the candidate advertisement contents and advertisement formats that meet a preset quality threshold according to the match score.

[0068] In some embodiments, the advertising delivery module includes: a platform selection unit, which determines the corresponding target advertising platform for advertising delivery based on the user's real-time environmental data; a delivery strategy unit, which uses a reinforcement learning algorithm to adjust the frequency and time interval of advertising delivery on the target advertising platform in real time; an A / B testing unit, which delivers advertisements corresponding to the optimal advertising content and advertising format on the target advertising platform, and simultaneously delivers multiple fine-tuned versions of advertising content and advertising formats to identify the best performing advertising variants.

[0069] In some embodiments, the advertising optimization delivery system further includes: an effect monitoring module, which is used to monitor the advertising delivery effect in real time and obtain feedback data on the advertising delivery effect; a model updating module, which uses the feedback data on the advertising delivery effect to update the model parameters of the convolutional neural network and generative adversarial network combination algorithm to improve the accuracy of advertising generation and matching.

[0070] In some embodiments, the effect monitoring module includes: a quantitative indicator collection unit, which is used to collect quantitative indicators including click-through rate, conversion rate, user residence time and interaction depth; a qualitative indicator collection unit, which uses natural language processing technology to extract sentiment tendency keywords from user comments as qualitative indicators; a comprehensive evaluation unit, which constructs a comprehensive evaluation model, integrates the quantitative indicators and the qualitative indicators, and generates a comprehensive score of advertising effect.

[0071] In some embodiments, the model update module includes: a feature importance analysis unit, which uses a gradient boosting decision tree algorithm based on the comprehensive score of advertising effect to obtain a feature importance ranking, thereby identifying the key feature combinations that rank in the top 10% in terms of their impact on advertising effect; a convolutional neural network update unit, which updates the model parameters of the convolutional neural network based on the identified key feature combinations and advertising effect data; and a generative adversarial network update unit, which updates the model parameters of the generative adversarial network based on the obtained user feedback data and the comprehensive score of advertising effect.

[0072] In some embodiments, the advertising optimization delivery system further includes: a graphics processor for accelerating the computing process of convolutional neural networks and generative adversarial networks; a network interface for exchanging data with external advertising platforms and data sources; and a database for storing user behavior data, advertising content, and delivery effect data.

[0073] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here. Example

[0074] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the above methods is implemented.

[0075] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. Of course, there are other forms of readable storage media, such as quantum memory, graphene memory, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunication signals. Example

[0076] The present invention also provides an electronic device. The electronic device of an embodiment of the present invention comprises: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the present invention.

[0077] Reference below Fig.10 , which shows a schematic diagram of the structure of a computer system 800 of an electronic device suitable for implementing an embodiment of the present invention. Fig.10 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0078] like Fig.10 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0079] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed.

[0080] In particular, according to the embodiments disclosed in the present invention, the process described in the main step diagram above can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the main step diagram. In the above embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the central processing unit 801, the above functions defined in the system of the present invention are executed.

[0081] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0082] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0083] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be arranged in a processor, and the names of these units do not constitute limitations on the units themselves in some cases.

[0084] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for optimizing advertisement delivery, characterized in that: It includes the following steps: S10: Acquire user behavior data, where the user behavior data includes historical interaction information and real-time environment data of the user; S20: Analyze the user's historical interaction information and real-time environment data to identify user characteristics and user interests; S30: Based on the combined algorithm of convolutional neural network and generative adversarial network, it dynamically generates the optimal advertising content and advertising format that matches the user characteristics and interests; S40: Using the optimal advertisement content and advertisement format to deliver advertisements in real time on the target advertisement platform.

2. The method for optimizing advertisement delivery according to claim 1, characterized in that: Step S10 specifically includes: S11: Obtaining historical interaction information of the user, which includes the user's ad browsing history, ad click history, and ad purchase history; S12: Acquire real-time environment data of the user, which includes geographic location data, device type data, and usage period data of the user; S13: Process the historical interaction information and the real-time environment data to obtain processed user behavior data.

3. The method for optimizing advertisement delivery according to claim 2, characterized in that: Step S20 specifically includes: S21: performing time series analysis on the processed user behavior data, arranging the processed user behavior data in chronological order, and dividing the processed user behavior data into a recent time window, a mid-term time window, and a long-term time window; S22: within each time window, statistically analyzing the user's historical interaction information, identifying the user's interest category within the time window based on the user's ad browsing history, ad click record, and ad purchase record, and generating an interest tag; S23: Combine the user's real-time environmental data and historical interaction information, use the decision tree algorithm to extract and classify features, and obtain the user's feature labels, which include age group labels, gender labels, common device type labels, and active time period labels.

4. The method for optimizing advertisement delivery according to claim 1, characterized in that: Step S30 specifically includes: S31: Analyze the user characteristics and the user interests using a convolutional neural network to extract key feature elements; S32: Based on the key feature elements, create a plurality of candidate advertising contents and advertising formats using a generative adversarial network architecture including a generator network and a discriminator network; wherein the generator network is used to generate the candidate advertising contents and advertising formats, and the discriminator network is used to evaluate the candidate advertising contents and advertising formats; S33: inputting the candidate advertisement content and advertisement format into the discriminator network, evaluating the matching degree between the candidate advertisement content and advertisement format and the key feature elements, and obtaining a matching score; S34: Optimizing the generator network according to the matching score; S35: Repeat steps S32 to S34 until the generated candidate advertisement content and advertisement format reach a preset quality threshold; S36: Selecting the best advertisement content and advertisement format from the candidate advertisement content and advertisement format that meet a preset quality threshold according to the matching score.

5. The method for optimizing advertisement delivery according to claim 4, characterized in that: Step S33 specifically includes: S331: Converting the candidate advertisement content and advertisement format into standardized candidate advertisement content and advertisement format; S332: Inputting the standardized candidate advertisement content and advertisement format, as well as the key feature elements, into the discriminator network; S333: using a discriminator network to calculate the similarity between the standardized candidate advertisement content and advertisement format and the key feature elements; S334: Generate a matching score based on the calculated similarity; S335: Repeat steps S331 to S334 for all candidate advertisement contents and advertisement formats to obtain a set of matching scores.

6. The method for optimizing advertisement delivery according to claim 5, characterized in that: Step S34 specifically includes: S341: Calculating the loss function of the generator network according to the matching score; S342: Using a back propagation algorithm, calculate the gradient of the parameters of each layer of the generator network based on the loss function; S343: Using the stochastic gradient descent method, based on the gradient calculated in step S342, update the parameters of the generator network, including the weights and biases of each layer; S344: Repeat steps S341 to S343 until the performance of the generator network reaches a preset threshold or the number of iterations reaches a predetermined value; S345: Save the optimized parameters of the generator network.

7. The method for optimizing advertisement delivery according to claim 1, characterized in that: Step S40 specifically includes: S41: Based on the real-time environment data of the user, determine a corresponding target advertising platform for advertising delivery; S42: Using a reinforcement learning algorithm, adjusting the frequency and time interval of advertising on the target advertising platform in real time; S43: delivering the advertisement corresponding to the optimal advertisement content and advertisement format on the target advertisement platform, and adopting an A / B testing method to simultaneously deliver multiple fine-tuned versions of advertisement content and advertisement format to identify the best performing advertisement variant.

8. The method for optimizing advertisement delivery according to claim 1, characterized in that: The method further comprises: S50: Monitor the advertising effect in real time to obtain feedback data of the advertising effect; use the feedback data of the advertising effect to update the model parameters of the convolutional neural network and generative adversarial network combination algorithm to improve the accuracy of advertising generation and matching.

9. The method for optimizing advertisement delivery according to claim 8, characterized in that: Step S50 specifically includes: S51: real-time monitoring of the delivery effects of all advertisements delivered on the target advertising platform, and obtaining feedback data on the delivery effects of the advertisements, which includes: quantitative indicators, including click-through rate, conversion rate, user stay time and user interaction depth; qualitative indicators, which are sentiment tendency keywords extracted from user comments using natural language processing technology; S52: constructing a comprehensive evaluation model, integrating the quantitative index and the qualitative index, and generating a comprehensive score of the advertising effect; S53: Based on the comprehensive score of advertising effect, the gradient boosting decision tree algorithm is used to obtain the feature importance ranking, thereby identifying the key feature combinations that rank in the top 10% in terms of their impact on advertising effect; S54: updating the model parameters of the convolutional neural network, specifically including: adjusting the convolution kernel size and the number of convolution kernels of the convolution layer according to the identified key feature combination; adjusting the pooling window size of the pooling layer according to the generated comprehensive score of the advertising effect; updating the number of neurons and the activation function of the fully connected layer according to the obtained user interaction depth; S55: Update the model parameters of the generative adversarial network, specifically including: adjusting the upsampling layer parameters of the generator network according to the obtained user residence time; adjusting the convolution layer parameters of the discriminator network according to the extracted sentiment tendency keywords; updating the weight of the adversarial loss function according to the generated comprehensive score of the advertising effect, and balancing the training process of the generator network and the discriminator network.

10. An advertisement optimization delivery system, characterized in that: The system comprises: A data collection module is used to obtain user behavior data, wherein the user behavior data includes the user's historical interaction information and real-time environment data; A data processing module, used to analyze the user's historical interaction information and real-time environment data, and identify user characteristics and user interests; Advertisement generation module, based on the combined algorithm of convolutional neural network and generative adversarial network, dynamically generates the optimal advertisement content and advertisement format that matches the user characteristics and interests; The advertisement delivery module is used to deliver advertisements in real time on a target advertisement platform using the optimal advertisement content and advertisement format.

Images