Real-time advertisement putting effect prediction and optimization method based on deep learning

Abstract

The invention discloses a real-time advertisement putting effect prediction and optimization method based on deep learning, particularly relates to the technical field of advertisement putting, and aims to analyze historical data of an advertisement putting platform, extract user behavior characteristics and advertisement content characteristics, and construct a deep learning model in a multi-modal fusion mode. And performing feature learning on the model by using a deep learning algorithm to generate a dynamic feature set, and distinguishing high-efficiency and low-efficiency delivery scenes through a classifier. And generating a real-time delivery optimization strategy based on a feature classification result in combination with the user group differentiation model and the delivery scene information. Real-time feedback data is generated after implementation, and the model continuously adapts to continuously changing markets and user interests through multi-level difference analysis and model updating. And predicting the overall effect of the advertisement by using the updated model, and triggering a strategy adjustment notification mechanism if the optimized putting effect or the user interest characteristics deviate, thereby realizing efficient, flexible and sustainable optimization of the advertisement putting.

Description

Technical Field

[0001] The present invention relates to a field, and more specifically, to a real-time advertising delivery effect prediction and optimization method based on deep learning. Background Art

[0002] Existing advertising decisions mostly rely on experience judgment or simple indicator analysis, lacking in-depth exploration of the relationship between user preferences, material characteristics and delivery timing. Traditional strategies are often based on fixed parameters or single-dimensional data, making it difficult to respond to changes in market hotspots and user interest shifts in a timely manner. This leads to problems such as waste of delivery resources, unstable delivery effects, and delayed strategy adjustments. With the rapid increase in the scale of Internet and mobile users, user behavior patterns have become more complex, and multimodal data (such as text, images, and videos) and user feature dimensions have continued to increase. Simple linear analysis and manual optimization methods can no longer meet the requirements of high-precision, real-time delivery. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a real-time advertising delivery effect prediction and optimization method based on deep learning to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A real-time advertising delivery effect prediction and optimization method based on deep learning, comprising the following steps:

[0006] Based on the historical data of the advertising platform, extract user behavior characteristics and advertising content characteristics, and build a deep learning model;

[0007] Performing feature learning on the deep learning model through a deep learning algorithm to generate a dynamic feature set for advertisement delivery;

[0008] Using a classifier to classify the dynamic feature set into categories of efficient delivery and inefficient delivery, and outputting feature classification results;

[0009] Generate a real-time delivery optimization strategy based on the feature classification results, combined with the user group differentiation model and delivery scenario;

[0010] Based on the real-time delivery optimization strategy, obtain the real-time feedback data of the optimized delivery, perform multi-level difference analysis with the feature classification results, and update the deep learning model;

[0011] Based on the updated deep learning model, the overall advertising effect is predicted and the optimized real-time delivery effect is generated;

[0012] According to the optimized real-time delivery effect and the changing trend of user interest characteristics, an advertising strategy adjustment notification mechanism is triggered.

[0013] In a preferred embodiment, based on the historical data of the advertising delivery platform, user behavior characteristics and advertising content characteristics are extracted to construct a deep learning model, including:

[0014] Conduct statistical analysis on the click-through rate, dwell time and conversion frequency of users in historical delivery records to extract the user behavior characteristics;

[0015] Decomposing and representing the characteristics of text, image or multimedia content of historically placed advertising materials, and extracting the characteristics of the advertising content;

[0016] Perform multimodal fusion of the user behavior characteristics and the advertising content characteristics through a feature mapping network to generate feature input for deep learning model training;

[0017] An original deep learning model with a multi-layer neural network structure is constructed, and the original deep learning model is trained using the feature input after multimodal fusion to obtain the deep learning model.

[0018] In a preferred embodiment, the deep learning model is subjected to feature learning through a deep learning algorithm to generate a dynamic feature set for advertisement delivery, including:

[0019] Iteratively update and optimize the feature input using a recursive neural network, a convolutional neural network, or a structure based on an attention mechanism to extract potential temporal patterns in the process of user-ad interaction;

[0020] Incremental data in the current advertising cycle is embedded in real time, and new user behaviors and advertising response patterns are integrated into the model feature space to form an initial dynamic feature set;

[0021] The initial dynamic feature set is processed by feature regularization and dimension reduction strategies to obtain the dynamic feature set; the stability and distinguishability of the dynamic feature set in multi-dimensional space are better than those of the initial dynamic feature set.

[0022] In a preferred embodiment, the dynamic feature set is classified into efficient delivery and inefficient delivery using a classifier, and the feature classification result is output, including:

[0023] Input the dynamic feature set into a pre-trained classifier model, and automatically classify the advertisement instances according to delivery effect evaluation indicators to obtain classification results; the delivery effect indicators at least include click-through conversion rate and user interaction depth;

[0024] The classification results are scored and filtered according to their confidence, and classification samples with too low confidence levels are eliminated to obtain the feature classification results.

[0025] In a preferred embodiment, the feature classification results include efficient delivery samples and inefficient delivery samples; based on the feature classification results, combined with the user group differentiation model and delivery scenario, a real-time delivery optimization strategy is generated, including:

[0026] Performing cluster analysis on the characteristic distribution of the user group to which the highly efficient delivery samples belong, and identifying user preference characteristics and key attributes;

[0027] Perform feature correction on the delivery scenarios corresponding to the inefficient delivery samples to discover potential improvement links; the delivery scenarios include region, time period, and device type;

[0028] According to the mapping relationship between the user group differentiation model and the delivery scenario, the bidding strategy, advertisement exposure frequency and material combination are dynamically adjusted to form the real-time delivery optimization strategy.

[0029] In a preferred embodiment, based on the real-time delivery optimization strategy, real-time feedback data of the optimized delivery is obtained, multi-level difference analysis is performed with the feature classification results, and the deep learning model is updated, including:

[0030] Collecting user interaction data generated after the implementation of the real-time delivery optimization strategy to obtain real-time feedback data; the real-time feedback data at least includes clicks, bounce rates or secondary visit behaviors;

[0031] Compare the real-time feedback data with the feature classification results in various dimensions to obtain difference analysis results; the difference analysis results are used to describe the effectiveness and defects of the real-time delivery optimization strategy adjustment; the dimensions at least include user group characteristics, advertising material special training or time series variables;

[0032] According to the difference analysis results, the deep learning model is retrained in parameters or the model structure is fine-tuned.

[0033] In a preferred embodiment, based on the updated deep learning model, the overall advertising effect is predicted to generate an optimized real-time delivery effect, including:

[0034] Use the updated deep learning model to predict the effect of the upcoming advertising instances and output the expected click-through rate and conversion rate;

[0035] A multi-dimensional evaluation is performed on the expected click rate and conversion rate to generate the optimized real-time delivery effect; the multi-dimensional evaluation includes at least one dimension of the evaluation of target audience fit, cost-benefit ratio and brand polish.

[0036] In a preferred embodiment, according to the optimized real-time delivery effect and the changing trend of user interest characteristics, triggering the advertising strategy adjustment notification mechanism includes:

[0037] When real-time monitoring detects that user interest characteristics are shifting, or the optimized real-time delivery effect is lower than the expected threshold, the strategy adjustment notification is triggered through the preset threshold judgment logic;

[0038] Send strategic adjustment prompts to advertising operations personnel or automated decision-making systems to guide the modification of delivery plans. The modification plan at least includes replacing creatives, redefining user groups, or adjusting pricing strategies.

[0039] The technical effects and advantages of the real-time advertising delivery effect prediction and optimization method based on deep learning of the present invention are as follows:

[0040] Through multimodal feature learning and classification, this method can quickly identify efficient and inefficient delivery scenarios, and flexibly adjust delivery strategies based on user group differentiation models and delivery scenarios. Compared with the traditional approach that relies on experience and static rules, the method of the present invention can perform multi-level difference analysis and model fine-tuning at any time according to real-time feedback data during the delivery process, so that the delivery strategy closely fits changes in user interests and market environment changes, thereby effectively improving the utilization rate of delivery resources and reducing unnecessary costs. Ultimately, the present invention significantly improves the accuracy and real-time performance of the prediction of the overall effect of advertising delivery, achieving continuous optimization of strategies and significant improvement in delivery results. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of a real-time advertising delivery effect prediction and optimization method based on deep learning according to the present invention; DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Reference Figure 1 .

[0044] The embodiment provides a real-time advertising delivery effect prediction and optimization method based on deep learning of the present invention, which includes the following steps:

[0045] Based on the historical data of the advertising platform, extract user behavior characteristics and advertising content characteristics, and build a deep learning model;

[0046] Performing feature learning on the deep learning model through a deep learning algorithm to generate a dynamic feature set for advertisement delivery;

[0047] Using a classifier to classify the dynamic feature set into categories of efficient delivery and inefficient delivery, and outputting feature classification results;

[0048] Generate a real-time delivery optimization strategy based on the feature classification results, combined with the user group differentiation model and delivery scenario;

[0049] Based on the real-time delivery optimization strategy, obtain the real-time feedback data of the optimized delivery, perform multi-level difference analysis with the feature classification results, and update the deep learning model;

[0050] Based on the updated deep learning model, the overall advertising effect is predicted and the optimized real-time delivery effect is generated;

[0051] According to the optimized real-time delivery effect and the changing trend of user interest characteristics, an advertising strategy adjustment notification mechanism is triggered.

[0052] Specifically, based on the historical data of the advertising platform, user behavior characteristics and advertising content characteristics are extracted to build a deep learning model, including:

[0053] Statistical analysis is performed on the click-through rate, dwell time and conversion frequency of users in historical delivery records to extract the user behavior characteristics; specifically, click-through rate: the ratio of the number of times a user clicks on an ad to the number of times the ad is displayed; dwell time: the average length of time a user stays on the ad target page, which can reflect the depth of the user's interest in the ad content; conversion frequency: the number and frequency of times a user actually purchases, registers or downloads from clicking on an ad;

[0054] By conducting statistical analysis on these indicators, we can find out the preferences and habits of users when facing different advertisements, and thus extract the characteristics that can represent the interactive behavior patterns of users. For example, if a certain type of advertisement has a high click-through rate and a high dwell time among young female users, then these behavioral characteristics can be used to identify the advertising preferences of this group.

[0055] Decompose and represent the characteristics of text, image or multimedia content of historical advertising materials, and extract the characteristics of the advertising content; specifically, the advertising content characteristics are information extracted from the materials of the advertisement itself (such as pictures, videos, title copy, description text, etc.). Extract text features such as keyword distribution and emotional tendency through natural language processing (NLP) technology; analyze image features such as color composition, main objects, and clarity and position of brand logos through computer vision (CV) technology; extract key image features and audio features in time series frames for video or multimedia content. If the advertisement is a picture showing sports shoes, its color (mainly red), shoe shape features (running shoes), brand logo location and clarity, etc. can be extracted through image recognition.

[0056] If the ad has a text description, such as "lightweight and breathable running shoes, suitable for outdoor sports in summer", the text features can be extracted through word segmentation, word embedding and other methods, such as the vector representation of keywords such as "lightweight", "breathable", "summer", and "outdoor sports";

[0057] The user behavior characteristics and the advertising content characteristics are multimodally fused through a feature mapping network to generate feature inputs for deep learning model training; specifically, user behavior characteristics belong to behavioral pattern data, and advertising content characteristics belong to media content characteristics, which are different modalities (behavioral data, visual data, text data). Multimodal fusion technology refers to the use of a specific neural network structure (feature mapping network) to map features of different modalities into the same feature space, so that the model can simultaneously understand the correlation between user preferences and advertising content attributes.

[0058] Through this fusion, we finally get a unified feature input, which is a high-dimensional vector that combines user behavior preferences and advertising material attributes, providing a basis for subsequent deep model training. For example, the user click-through rate is fused with the color features in the advertising image. Assuming that the user behavior feature is a vector (such as [0.1 (click-through rate), 30 seconds (stay), 0.05 (conversion rate)]), the advertising content feature is another vector (such as color vector, keyword vector), these two parts of data are merged into a feature input through a fusion layer (such as a simple connection layer or an attention mechanism fusion layer);

[0059] Construct an original deep learning model with a multi-layer neural network structure, and use the feature input after multimodal fusion to train the original deep learning model to obtain the deep learning model; specifically, use the above-mentioned fused feature input to build a deep learning model (such as a multi-layer perceptron, a network containing convolutional layers, RNN or Transformer structure). The model can learn complex mapping relationships through multiple layers of nonlinear transformations, thereby providing deep representation for subsequent feature extraction, classification, and prediction. The final output is a trainable model framework that can take multimodal features as input and adapt to the optimization needs of advertising delivery scenarios through parameter training.

[0060] Specifically, the deep learning model is subjected to feature learning through a deep learning algorithm to generate a dynamic feature set for advertisement delivery, including:

[0061] Using recursive neural networks, convolutional neural networks or structures based on attention mechanisms, the feature inputs are iteratively updated and optimized to extract potential temporal patterns in the process of user-ad interaction; specifically, RNN (such as LSTM or GRU) can be used to capture temporal features: for example, the pattern of how user clicks on ads change with date in the past week. CNN can be used to extract local patterns of features, such as extracting local regularities of continuous feature changes. The use of attention mechanisms allows the model to focus on specific time points or specific types of features, so as to better extract feature combinations that have an important impact on ad effect prediction. Assuming that we collect user interaction data on a certain type of ad every day, RNN can input data from the past 7 days and iteratively update the hidden state to learn the temporal trend of user interest changes within 7 days. If a user is unusually active on an ad on a certain day, RNN can capture this peak change and consider it in subsequent predictions;

[0062] The incremental data in the current advertising delivery cycle is embedded in real time, and the newly added user behaviors and advertising response patterns are integrated into the model feature space to form an initial dynamic feature set; specifically, in actual delivery, data is continuously generated. For example, a new advertisement that has just been launched today will gradually generate click and stay data. By embedding the newly generated data (incremental data) into the model in real time, that is, updating the input features of the model online, the feature representation of the model can be continuously updated. This is combined with the basic model constructed in claim 2, so that the model is no longer fixed after training, but changes in real time with the feedback data of the actual delivery, thereby forming a dynamic feature set;

[0063] The initial dynamic feature set is processed by feature regularization and dimensionality reduction strategies to obtain the dynamic feature set; the dynamic feature set ensures that the stability and distinguishability of the dynamic feature set in multi-dimensional space are better than the initial dynamic feature set; specifically, the dynamic feature set may be very high in dimension and constantly updated. In order to avoid computational complexity problems caused by feature redundancy, overfitting, and high dimensionality, the model needs to regularize the features (such as L2 regularization, dropout) and reduce the dimensionality (such as using PCA or autoencoder for embedding and dimensionality reduction). Through these strategies, the dynamic feature set can adapt to real-time changes while maintaining a certain degree of stability and interpretability, and will not affect classification or prediction performance due to high dimensionality or excessive noise.

[0064] Specifically, the dynamic feature set is classified into efficient delivery and inefficient delivery using a classifier, and the feature classification result is output, including:

[0065] The dynamic feature set is input into a pre-trained classifier model, and the advertising instances are automatically classified according to the delivery effect evaluation indicators to obtain the classification effect; the delivery effect evaluation indicators at least include click-through conversion rate and user interaction depth; specifically, the classifier (such as a random forest, a gradient boosting decision tree, or a fully connected neural network classification layer) will automatically classify these feature inputs, with the goal of distinguishing advertisements into two categories: "efficient delivery" (i.e., already performing well) and "inefficient delivery" (poor performance).

[0066] Effect evaluation indicators such as click-through rate (CTR) and user interaction depth (such as multiple clicks, favorites, and secondary visits) provide reference standards for classification. If the CTR and interaction depth of an ad are continuously higher than the average of the industry or ads of the same category, it tends to be classified as an efficient delivery ad; assuming that the CTR of an ad in the current cycle is significantly higher than the average level, and the user's stay time and secondary visit rate for the ad are also high, then the classifier will mark the ad as efficient delivery; otherwise, it is inefficient delivery;

[0067] The classification results are scored and filtered for confidence, and the classification samples with low confidence are eliminated to obtain the feature classification results; specifically, when the classifier gives the classification result, it usually provides a confidence score, which indicates the degree of certainty of the model in the classification judgment. If the confidence is too low, it means that the classification result of the sample is not reliable enough.

[0068] Filtering samples with low confidence can reduce the impact of misjudgment on subsequent strategy formulation. When the model is unclear about an advertisement, it will not easily classify it as efficient or inefficient, so as not to mislead the generation of subsequent optimization strategies. For example, if an advertisement has special features and the classifier predicts that the advertisement is efficiently delivered, but the confidence level is only 55%, and the minimum confidence level set by the system is 80%, then the advertisement will not be immediately classified as efficient and may require more data or manual review.

[0069] Specifically, the feature classification results include efficient delivery samples and inefficient delivery samples; based on the feature classification results, combined with the user group differentiation model and delivery scenario, a real-time delivery optimization strategy is generated, including:

[0070] Perform cluster analysis on the characteristic distribution of the user group to which the efficient delivery samples belong, and identify user preference characteristics and key attributes; specifically, the classified efficient advertising instances correspond to a group of user groups that have achieved good delivery results. By performing cluster analysis on these groups, it is possible to discover the common preference characteristics of this group of users. For example, they may be more interested in advertisements during the same time period (such as after 8 p.m.), or may be more likely to make purchases for a certain type of product. The results of cluster analysis can be used for subsequent targeted delivery optimization; for example, clustering of user groups for efficient advertising found that users aged 30-35, living in first-tier cities, and having a high interest in technology products browse shopping apps late at night and are easy to convert. These characteristics can guide us to strengthen the exposure of this type of advertising in similar user groups;

[0071] For the delivery scenarios corresponding to the inefficient delivery samples, feature correction is performed to find potential improvement links; the delivery scenarios include region, time period, and device type; specifically, inefficiently delivered ads often indicate that the current delivery strategy is not feasible in certain scenarios. For example, users in a specific region have a low response to ads, or the ad display effect is poor on a specific device (such as a tablet).

[0072] Analyzing and correcting the features of these scenarios can help us identify problems, such as the need for materials that are more suitable for local users, or optimizing materials for mobile devices, thereby improving inefficient delivery. For example, if we find that a certain ad has a very low click-through rate in a certain province, it may be because users in that province have low interest in this type of product, or the ad loads slowly. After feature correction, we can identify the need for lighter materials to reduce loading delays, thereby increasing click-through rates;

[0073] According to the mapping relationship between the user group differentiation model and the delivery scenario, the bidding strategy, advertising exposure frequency and material combination are dynamically adjusted to form the real-time delivery optimization strategy; specifically, the core role of the user group differentiation model is to identify different user groups by analyzing the user's behavioral characteristics, interest preferences, etc., and formulate personalized advertising delivery strategies for each group; through this model, advertisers can accurately locate the target user group, improve the relevance and interaction rate of advertisements, thereby optimizing the delivery effect and improving ROI (return on investment); the user group differentiation model is constructed by deep learning and clustering analysis of the historical data of the advertising platform; features are extracted based on the user's behavioral data (such as click records, browsing history, purchase behavior, etc.), and the algorithm is used to divide the users into multiple groups; each group is personalized predicted and optimized according to its characteristics (such as activity, points of interest, etc.), so as to ensure that the advertising content is highly matched with the needs of the target group, and the model is dynamically updated according to market changes; personalized strategies are formulated for advertisements according to different user group characteristics and different delivery scenarios. For example, for efficient groups, you can appropriately increase bids and exposure frequency; for inefficient scenarios, lower bids, reduce exposure or change creatives; for example, for efficient user groups, moderately increase bids, increase exposure frequency, and try different creatives to further improve results; reduce bids for inefficient areas, give priority to high-quality creatives for efficient groups, and achieve optimal resource allocation.

[0074] Specifically, based on the real-time delivery optimization strategy, real-time feedback data of the delivery after optimization is obtained, multi-level difference analysis is performed with the feature classification results, and the deep learning model is updated, including:

[0075] The user interaction data generated after the implementation of the real-time delivery optimization strategy is collected to obtain real-time feedback data; the user interaction data at least includes clicks, bounce rate or secondary visit behavior; specifically, click data: whether the user clicks on the new advertising material; bounce rate: whether the user leaves immediately after clicking the advertisement without further interaction; secondary visit behavior: whether the user visits the advertisement or related pages again in a short period of time;

[0076] The real-time feedback data and the feature classification results are compared in various dimensions to obtain a difference analysis structure; the difference analysis results are used to describe the effectiveness and shortcomings of the real-time delivery optimization strategy adjustment; the dimensions include at least user group characteristics, advertising material characteristics or time series variables; specifically, user group dimension: whether the previously defined high-efficiency group still performs well; whether the optimization strategy is also effective on the new user group; advertising material characteristic dimension: whether the interaction is improved after the material is adjusted; time series dimension: whether the effect of the strategy adjustment continues to improve over time or is just a short-term phenomenon;

[0077] According to the difference analysis results, the parameters of the deep learning model are retrained or the model structure is fine-tuned to improve the adaptability and generalization ability of the model to subsequent delivery scenarios; specifically, when the difference analysis shows that the model does not predict certain newly emerging scenarios well, or the effect of a specific user group and material combination does not meet expectations, we can retrain or fine-tune the deep learning model itself. For example, increase the amount of training data, optimize the learning rate, and adjust the network structure (such as adding an attention layer, changing the network depth) to make the model better adapt to the ever-changing delivery environment. By continuously updating the model, online learning and adaptive optimization can be achieved to cope with dynamic changes in the advertising delivery environment.

[0078] Specifically, based on the updated deep learning model, the overall advertising effect is predicted and the optimized real-time delivery effect is generated, including:

[0079] The updated deep learning model is used to predict the effect of the upcoming advertising instances and output the expected click-through rate and conversion rate. Specifically, for new advertising instances (advertisements that have not yet been actually launched or have just started to be launched), the model can predict future performance, such as click-through rate and conversion rate. This helps operators understand the effect of advertising before or in the early stages of the launch, making it easier to formulate strategies in advance.

[0080] A multi-dimensional evaluation is performed on the expected click rate and conversion rate to generate the optimized real-time delivery effect; the multi-dimensional evaluation includes at least one dimension of target audience fit, cost-benefit ratio and brand exposure; specifically, target audience fit: whether the advertisement is delivered to the user group that is most likely to be interested; cost-benefit ratio: whether the advertising cost invested matches the revenue generated; brand exposure: even if the conversion rate is not high, can the advertisement increase brand awareness and potential consumer cognition; taking these dimensions into consideration, a more comprehensive evaluation of the advertising delivery effect is formed, which is called the optimized real-time delivery effect.

[0081] Specifically, according to the optimized real-time delivery effect and the changing trend of user interest characteristics, the advertising strategy adjustment notification mechanism is triggered, including:

[0082] When real-time monitoring shows that user interest characteristics are shifted, or the optimized real-time delivery effect is lower than the expected threshold, the strategy adjustment notification is triggered through the preset threshold judgment logic; specifically, user interest characteristics shift:

[0083] User interest characteristics (such as browsing behavior, purchase intention, keyword preference) may change over time, with hot events or seasonal changes; for example, during the holidays, users may pay more attention to discounted products rather than high-end brands;

[0084] These changes can be detected by tracking user behavior data (such as search keywords, types of ads recently clicked); once the interest feature shift exceeds the set threshold (for example, the user's click rate on a certain type of product suddenly drops by more than 20%), a notification will be triggered;

[0085] If the optimized delivery results, such as click-through rate and conversion rate, do not reach the target value (threshold) set by the system, it indicates that there may be problems with the current delivery strategy and it needs to be adjusted. For example, if the click-through rate target is set to 5%, but the actual click-through rate is only 3%, then an adjustment will be triggered.

[0086] Thresholds are a set of predefined conditions in the system that are used to determine whether delivery effects or user interest characteristics deviate from the normal range. Through logical judgment (such as "if click rate < threshold or conversion rate < threshold, trigger notification"), anomalies are automatically identified.

[0087] Sending strategy adjustment prompts to advertising operators or automated decision-making systems to guide modification of delivery plans, which may include at least replacing creatives, redefining user groups, or adjusting pricing strategies; specifically, sending notifications: after detecting user interest shifts or abnormal delivery effects, adjustment suggestions will be sent to relevant personnel (such as the advertising operation team) in the form of notifications;

[0088] Suggestions for strategy adjustment: Based on the actual situation, the following adjustment strategies are proposed:

[0089] Replace creatives: If the existing ad creatives lose their appeal, it is recommended to replace them with new creatives that are more in line with current user interests; for example, change autumn product promotions to winter new product promotions;

[0090] Redefine the user group: If the interest characteristics of the original user group have changed, it is recommended to redefine the target audience. For example, adjust the key users from 30-40 year old men to 20-30 year old women;

[0091] Adjust pricing strategy: For auction ads, you may be advised to increase bids to increase exposure, or lower bids to reduce cost waste.

[0092] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0093] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0096] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0097] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0098] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0099] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0100] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time advertising effect prediction and optimization method based on deep learning, characterized in that: The following steps are involved: Based on the historical data of the advertising platform, extract user behavior characteristics and advertising content characteristics, and build a deep learning model; Performing feature learning on the deep learning model through a deep learning algorithm to generate a dynamic feature set for advertisement delivery; Using a classifier to classify the dynamic feature set into categories of efficient delivery and inefficient delivery, and outputting feature classification results; Generate a real-time delivery optimization strategy based on the feature classification results, combined with the user group differentiation model and delivery scenario; Based on the real-time delivery optimization strategy, obtain the real-time feedback data of the optimized delivery, perform multi-level difference analysis with the feature classification results, and update the deep learning model; Based on the updated deep learning model, the overall advertising effect is predicted and the optimized real-time delivery effect is generated; According to the optimized real-time delivery effect and the changing trend of user interest characteristics, an advertising strategy adjustment notification mechanism is triggered.

2. The method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 1, characterized in that: Based on the historical data of the advertising platform, we extract user behavior characteristics and advertising content features and build a deep learning model, including: Conduct statistical analysis on the click-through rate, dwell time and conversion frequency of users in historical delivery records to extract the user behavior characteristics; Decomposing and representing the characteristics of text, image or multimedia content of historically placed advertising materials, and extracting the characteristics of the advertising content; Perform multimodal fusion of the user behavior characteristics and the advertising content characteristics through a feature mapping network to generate feature input for deep learning model training; An original deep learning model with a multi-layer neural network structure is constructed, and the original deep learning model is trained using the feature input after multimodal fusion to obtain the deep learning model.

3. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 2, characterized in that: Through the deep learning algorithm, feature learning is performed on the deep learning model to generate a dynamic feature set for advertising delivery, including: Iteratively update and optimize the feature input using a recursive neural network, a convolutional neural network, or a structure based on an attention mechanism to extract potential temporal patterns in the process of user-ad interaction; Incremental data in the current advertising cycle is embedded in real time, and new user behaviors and advertising response patterns are integrated into the model feature space to form an initial dynamic feature set; The initial dynamic feature set is processed by feature regularization and dimension reduction strategies to obtain the dynamic feature set; the stability and distinguishability of the dynamic feature set in multi-dimensional space are better than those of the initial dynamic feature set.

4. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 3, characterized in that: The dynamic feature set is classified into efficient delivery and inefficient delivery by using a classifier, and feature classification results are output, including: Input the dynamic feature set into a pre-trained classifier model, automatically classify the advertisement instance according to the delivery effect evaluation index, and obtain a classification result; the delivery effect evaluation index at least includes click-through conversion rate and user interaction depth; The classification results are scored and filtered according to their confidence, and classification samples with too low confidence levels are eliminated to obtain the feature classification results.

5. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 4, characterized in that: The feature classification results include efficient delivery samples and inefficient delivery samples; Based on the feature classification results, combined with the user group differentiation model and delivery scenario, a real-time delivery optimization strategy is generated, including: Performing cluster analysis on the characteristic distribution of the user group to which the highly efficient delivery samples belong, and identifying user preference characteristics and key attributes; Perform feature correction on the delivery scenarios corresponding to the inefficient delivery samples to discover potential improvement links; the delivery scenarios include region, time period, and device type; According to the user group differentiation model and the mapping relationship between the delivery scenarios, the bidding strategy, advertisement exposure frequency and material combination are dynamically adjusted to form the real-time delivery optimization strategy.

6. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 5, characterized in that: Based on the real-time delivery optimization strategy, real-time feedback data of the optimized delivery is obtained, multi-level difference analysis is performed with the feature classification results, and the deep learning model is updated, including: Collecting user interaction data generated after the implementation of the real-time delivery optimization strategy to obtain real-time feedback data; the user interaction data at least includes clicks, bounce rates or secondary visit behaviors; Compare the real-time feedback data with the feature classification results in various dimensions to obtain difference analysis results; the difference analysis results are used to describe the effectiveness and defects of the real-time delivery optimization strategy adjustment; the dimensions at least include user group characteristics, advertising material characteristics or time series variables; According to the difference analysis results, the deep learning model is retrained in parameters or the model structure is fine-tuned.

7. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 6, characterized in that: Based on the updated deep learning model, the overall advertising effect is predicted and optimized real-time delivery effects are generated, including: Use the updated deep learning model to predict the effect of the upcoming advertising instances and output the expected click-through rate and conversion rate; A multi-dimensional evaluation is performed on the expected click rate and conversion rate to generate the optimized real-time delivery effect; the multi-dimensional evaluation includes at least an evaluation of any one of the dimensions of target audience fit, cost-benefit ratio and brand exposure.

8. A method for predicting and optimizing real-time advertising delivery effects based on deep learning according to claim 7, characterized in that: According to the optimized real-time delivery effect and the changing trend of user interest characteristics, an advertising strategy adjustment notification mechanism is triggered, including: When real-time monitoring detects that user interest characteristics are shifting, or the optimized real-time delivery effect is lower than the expected threshold, the strategy adjustment notification is triggered through the preset threshold judgment logic; Sending a strategy adjustment prompt to the advertising operation personnel or the automated decision-making system to guide the modification of the delivery plan, wherein the modification of the delivery plan at least includes replacing the material, redefining the user group, or adjusting the pricing strategy.

9. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the real-time advertising delivery effect prediction and optimization method based on deep learning as described in any one of claims 1 to 8 is implemented.

10. A computer storage medium, characterized in that: It includes computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the real-time advertising delivery effect prediction and optimization method based on deep learning as described in any one of claims 1 to 8.

Images