A digital advertising system

Abstract

The present invention relates to the field of advertising delivery technology, and specifically, to a digital advertising system. It includes a data acquisition preprocessing module, a deep learning model reasoning module, a scenario adaptation decision module, a dynamic frequency adjustment module, and a performance monitoring and feedback loop module. The data acquisition preprocessing module collects and cleans multi-dimensional user behavior data in real time to ensure high-quality data input; the deep learning model reasoning module constructs and updates the context-aware graph to capture the user's immediate needs and complex contextual relationships, and provides an accurate personalized recommendation basis; the scenario adaptation decision module combines real-time context information to intelligently judge the user's immediate state and determine the best time to display advertisements. The present invention introduces context-aware graph technology into the deep learning model reasoning module to achieve accurate capture of user instant behavior and complex contextual relationships, and efficiently associates the two, thereby providing more personalized and timely advertising content.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement placement, and in particular to a digital advertisement system. Background Art

[0002] In today's digital age, the popularity of the Internet and mobile devices has made digital advertising systems an important tool for companies and brands to attract potential customers. Marketers hope to display the most relevant advertising content to target users at the most appropriate time and place through precise advertising, so as to increase user engagement and conversion rate. If advertising is not handled properly, such as frequent or inappropriate advertising displays, it will not only reduce the user experience, but may also cause users to be disgusted or even uninstall the application, thereby weakening the advertising effect and the profitability of the application.

[0003] Although marketers are aware of the importance of accurate advertising, existing technologies still have obvious deficiencies in dealing with in-app advertising, especially in optimizing user experience. Existing technologies have a problem of insufficient relevance when dealing with deep user behavior patterns. Specifically, it is difficult to accurately capture the relationship between users' immediate behavior and complex contexts and associate the two together, resulting in inaccurate timing and content of advertising display, affecting user experience and advertising effectiveness. This results in that although advertisements can classify users and adjust frequency based on certain historical characteristics, they fail to accurately match the user's current status in actual effect, and thus fail to effectively impress the target audience. Summary of the invention

[0004] The purpose of the present invention is to provide a digital advertising system to solve the problem that it is difficult to accurately associate users' immediate behaviors with specific situations in existing advertising systems. Specifically, the deep learning model reasoning module dynamically updates CAGS based on a high-quality multi-dimensional user behavior spectrum, where each node represents a specific user behavior or state, and the edge represents the correlation between behaviors. By continuously integrating real-time data streams and historical data, CAGS not only captures long-term behavior patterns, but also incorporates external contextual information such as geographic location and timestamps to ensure a more comprehensive and accurate understanding of user scenarios. This dynamic update mechanism enables CAGS to efficiently reflect changes in user behavior, thereby providing a precise and timely basis for advertising content recommendations.

[0005] To achieve the above purpose, a digital advertising system is provided, including a data collection preprocessing module, which is used to collect multi-dimensional user behavior spectra from multiple channels in real time, and clean and extract features from these data, and also includes:

[0006] Deep learning model inference module: The deep learning model inference module receives the multi-dimensional user behavior spectrum processed by the data collection preprocessing module, calculates the output through the forward propagation of the multi-layer neural network, and converts it into a probability distribution through the softmax function to represent the user's interest in different advertisements;

[0007] The deep learning model inference module builds and updates the context-aware graph based on the multi-dimensional user behavior spectrum. The context-aware graph is used to capture the user's immediate needs and complex contextual relationships and provide a precise basis for personalized recommendations. Each node in the context-aware graph represents a specific user behavior or state, and the edge represents the correlation between behaviors. The context-aware graph continuously integrates real-time data streams and historical data from the data acquisition and preprocessing module, performs forward propagation calculations through a multi-layer neural network, and generates feature representations. The feature representations are then passed to the softmax function in the deep learning model inference module and converted into a new probability distribution. The probability distribution is further passed to the scenario adaptation decision module to determine the most appropriate advertising content and deliver it.

[0008] As a further improvement of the present technical solution, the context-aware graph dynamically updates node weights and edge connection strengths so that the model can reflect the user's latest behavior patterns in real time.

[0009] As a further improvement of the present technical solution, the context-aware graph incorporates geographic location, timestamp, and weather conditions into node weights to enhance the accuracy of understanding the user's specific context.

[0010] As a further improvement of the present technical solution, the deep learning model reasoning module transmits the processed and analyzed multi-dimensional user behavior spectrum and the context-aware graph to the scenario adaptation decision module for further scenario adaptation decision.

[0011] As a further improvement of the present technical solution, the scenario adaptation decision module combines the current geographical location and time and determines the most appropriate advertisement content and its display strategy through decision logic.

[0012] As a further improvement of the present technical solution, the scenario adaptation decision module analyzes the real-time performance data and user feedback from the display strategy, and decides whether to allow advertising push by monitoring the user's various behavioral signals and device status. If the user is determined to be in a non-idle state, a blocking mechanism is adopted.

[0013] As a further improvement of the present technical solution, if the scenario adaptation decision module determines that the user is in an idle state, the multi-dimensional user behavior spectrum is transmitted to the dynamic frequency adjustment module.

[0014] As a further improvement of the present technical solution, the dynamic frequency adjustment module identifies the prime time of each user based on the user's behavior pattern, and optimizes the advertisement push frequency in a personalized manner according to the user's interaction history and current status.

[0015] As a further improvement of the present technical solution, the scenario adaptation decision module transmits the real-time performance data and user feedback to the performance monitoring and feedback loop module.

[0016] As a further improvement of the present technical solution, the performance monitoring and feedback loop module monitors and analyzes the click-through rate, conversion rate and other key performance indicators of the advertisement in real time, and feeds them back to the scenario adaptation decision module and the dynamic frequency adjustment module.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. In this digital advertising system, the accuracy and timeliness of advertising delivery are significantly improved by building and updating the context-aware graph in the deep learning model inference module. This process incorporates external contextual information such as geographic location, timestamp, and weather conditions, making the understanding of user scenarios more comprehensive and accurate. Specifically, geographic location information can help the system determine whether the user is at home or at work; timestamp information can identify the user's daily activity patterns, such as commuting time and rest periods; weather conditions can infer the user's current emotional state and possible interest preferences. Through the integration of these external contextual information, CAGS can more accurately infer the user's immediate needs, provide more personalized and timely advertising content recommendations, avoid inappropriate advertising interruptions, and thus significantly improve user satisfaction and advertising conversion rates.

[0019] 2. In this digital advertising system, the frequency of advertising push is effectively optimized and the user experience is significantly improved through the collaborative work of the scenario adaptation decision module and the dynamic frequency adjustment module. The scenario adaptation decision module combines real-time context information to intelligently judge the user's immediate status and determine the best time to display the advertisement. For those users who frequently interact with advertisements or show a high interest in a specific type of advertisement, the system appropriately increases the number of advertisement displays; for users who show signs of advertising fatigue or are indifferent to advertisements, the frequency of advertisements is reduced to avoid excessive disturbance. In addition, the module identifies the "golden hours" of each user, moderately increases the advertising exposure rate during these hours, and reduces the frequency of advertising push during the time periods when the user is less active. This personalized frequency optimization mechanism ensures that advertisements are only displayed to users in the most appropriate context, maximizing the advertising effect while also improving the overall user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1It is a schematic diagram of the digital advertising system flow of the present invention;

[0021] Figure 2 This is a schematic diagram of the deep learning model reasoning module flow of the present invention;

[0022] Figure 3 It is a flow chart of the scenario adaptation decision module of the present invention.

[0023] The meaning of each number in the figure is:

[0024] Among them: 100, data acquisition preprocessing module; 200, deep learning model reasoning module; 300, scenario adaptation decision module; 400, dynamic frequency adjustment module; 500, performance monitoring and feedback loop module. DETAILED DESCRIPTION

[0025] 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.

[0026] At the same time, some technical terms are explained here:

[0027] CAGS refers to Context-Aware Graph Spectrum, which is a multi-dimensional network based on graph theory;

[0028] The Softmax function is a mathematical function that is usually used to convert a vector containing K real values ​​into a probability distribution vector of the same length. It ensures that the output values ​​are between 0 and 1, and the sum of all elements is equal to 1. This makes the softmax function very suitable for probability prediction in multi-classification problems, which refers to P ( );

[0029] Click-through rate (CTR) refers to the ratio of the number of times an ad is displayed that users click on it, and is used to measure the effectiveness of an ad in attracting users to click on it.

[0030] Conversion rate (CVR) refers to the proportion of users who complete the expected goal (such as purchase, registration or download) after clicking on the ad, reflecting the effectiveness of the ad in guiding users to take action;

[0031] Other key performance indicators (KPIs) include but are not limited to return on investment, cost per acquisition, page dwell time, etc., which comprehensively evaluate the overall performance and efficiency of the advertising campaign.

[0032] See also Figure 1 As shown, a digital advertising system is provided, which aims to solve the problem of inaccurate advertising delivery and insufficient user experience caused by the difficulty in accurately linking users' immediate behaviors and specific situations in existing advertising systems. Specifically, more accurate and personalized advertising delivery is achieved through a data collection preprocessing module 100, a deep learning model reasoning module 200, a scenario adaptation decision module 300, and a dynamic frequency adjustment module 400.

[0033] First, the data collection preprocessing module 100 is responsible for real-time collection of multi-dimensional user behavior spectra from multiple channels (such as click streams, browsing history, purchase records, in-app operations, etc.), and data cleaning and feature extraction of the collected multi-dimensional user behavior spectra. This module collects real-time user behavior data through multiple methods such as API interfaces, log files, and sensor data, and ensures the quality of the multi-dimensional user behavior spectra by removing noise and invalid data, and extracts key features from it, such as the user's active time period, preferred content category, and recent operation path. In order to ensure the quality of the multi-dimensional user behavior spectra, we use a comprehensive data processing formula that combines outlier detection and dimensionality reduction technology:

[0034] ;

[0035] In the formula, It is a data set after outliers have been removed;

[0036] is the observed value in the original data;

[0037] and are the mean and standard deviation respectively;

[0038] is the transformation matrix composed of principal components;

[0039] is the data matrix after dimensionality reduction.

[0040] Through the above comprehensive formula, not only the quality of the multi-dimensional user behavior spectrum is improved, but also redundant information is reduced, so that the subsequent deep learning model inference module 200 can run more efficiently. High-quality multi-dimensional user behavior spectrum input is crucial for the deep learning model, which helps the model to more accurately capture the user's immediate behavior and specific context, thereby providing more accurate and personalized recommendations for the advertising system. In this way, the advertising delivery of the entire system is not only more accurate, but also can significantly improve the user experience.

[0041] In order to further improve the accuracy and timeliness of advertising, we noticed that traditional user portraits are often static and cannot reflect the rapid changes in user behavior in a timely manner. Although existing methods can provide certain analysis results when processing multi-dimensional user behavior spectra, they are obviously insufficient in capturing users' immediate behaviors and specific situations. In particular, when faced with complex and changing user behavior patterns, traditional methods often cannot update user behavior patterns in real time, resulting in inaccurate advertising recommendations and affected user experience.

[0042] Therefore, we need to pass the processed multi-dimensional user behavior spectrum to a more powerful module to dynamically analyze and predict the user's immediate behavior and specific context. To this end, we introduced a deep learning model reasoning module 200. This module not only analyzes based on preprocessed high-quality data, but also builds and updates context-aware graphs in it. In this way, the deep learning model can reflect the user's latest behavior patterns in real time and capture their latest interests and needs. This dynamic adjustment capability enables the advertising system to display the most relevant advertising content at the most appropriate time and place, avoiding inappropriate advertising interruptions, thereby greatly improving user satisfaction and advertising conversion rates.

[0043] See also Figure 2 As shown, the deep learning model inference module 200 uses advanced deep neural network technology to perform complex pattern recognition and prediction based on preprocessed high-quality data. This module can not only process a large amount of multi-dimensional user behavior data, but also calculate the output through the forward propagation of a multi-layer neural network, and finally convert it into a probability distribution through a softmax function to represent the user's interest in different advertisements. Specifically, the module uses the following formula to describe its working principle:

[0044] ;

[0045] ;

[0046] and They are The weight matrix and bias vector of the layer;

[0047] It is The activation value of the layer;

[0048] No. The activation value of the layer;

[0049] is the activation function;

[0050] Indicates that given the input x, the category The probability of the user's interest in the i-th advertisement;

[0051] Represents the label or output of the i-th category;

[0052] x represents the input vector, usually containing eigenvalues;

[0053] Represents the model's original score for the i-th category, usually the output of the last layer of the neural network;

[0054] express The exponential function value of , used to convert the raw score to a positive number.

[0055] Through the above formula, the deep learning model can efficiently process the multi-dimensional user behavior spectrum and perform real-time analysis in combination with contextual information. This dynamic adjustment capability enables the advertising system to display the most relevant advertising content at the most appropriate time and place, avoiding inappropriate advertising interruptions, thereby greatly improving user satisfaction and advertising conversion rates.

[0056] In the context-aware graph constructed by the deep learning model reasoning module 200, each node represents a specific user behavior or state, and the edge represents the correlation between behaviors. CAGS can not only capture the long-term behavior patterns of users, but also dynamically reflect the immediate behavior and specific situations of users, making advertising delivery more accurate and personalized. The construction and update mechanism of CAGS is dynamic and constantly adjusted based on real-time data and historical data. By combining real-time data streams (such as click streams, browsing history, purchase records, in-app operations, etc.), CAGS can update the latest behavior patterns of users in real time. For example, when users frequently visit sports news websites in a certain period of time, CAGS will update the user's interest weight in this type of content accordingly. At the same time, CAGS not only pays attention to the user's behavior patterns, but also considers external context information (such as geographic location, timestamp, weather conditions, etc.). These context information are integrated into CAGS through the edge and node attributes in the graph structure, which enhances the accuracy of understanding the user's scenario. For example, users may prefer entertainment advertisements on weekend nights, while paying more attention to career development-related content during the day on weekdays. Through the fusion of context information, CAGS can more accurately infer users' immediate needs.

[0057] Based on the constructed and updated CAGS, the deep learning model can perform efficient dynamic reasoning. The output is calculated through forward propagation and finally converted into a probability distribution through the softmax function, indicating the user's interest in different advertisements. Specifically, this module uses a deep neural network to reason about the cleaned multi-dimensional user behavior data to generate a personalized advertising recommendation list or specific advertising content. Specifically, the following formula is used to describe its working principle:

[0058] ;

[0059] ;

[0060] Refers to user behavior characteristics that are updated in real time through context-aware graphs;

[0061] is the update function used to fuse new and old data and generate a new feature vector ;

[0062] It is a historical behavioral characteristic;

[0063] It is a real-time behavioral feature;

[0064] It is a comprehensive feature vector that combines behavioral features and external context information in the context-aware graph. It represents the features that can reflect the user's immediate behavior and specific situation after processing;

[0065] is a function used to fuse behavioral features and external context information;

[0066] It is the user's behavioral feature vector, which contains the user's real-time behavioral data, such as click stream, browsing history, purchase history, etc. These data reflect the user's immediate interests and behavior patterns;

[0067] It is an external context information vector, which contains external information related to user behavior, such as geographic location, timestamp, weather conditions, etc. This information helps to understand the user's specific situation more accurately.

[0068] When new user behavior data When it arrives, we first update the node Weight and edge Connection strength , and by incorporating external contextual information Further adjust the node weights, and the final feature vector For personalized recommendation decisions:

[0069] ;

[0070] ;

[0071] ;

[0072] ;

[0073] in, Represents the value or attribute of node i, and node i represents a specific node in the graph;

[0074] is the node at time t The weight of

[0075] At time t-1, the node The weight of

[0076] is the strength of the edge between node i and node j at time t;

[0077] is the strength of the edge between node i and node j at time t-1;

[0078] is the user behavior data collected at the current time t;

[0079] is the edge between node i and node j, which represents the association or interaction strength between the two nodes;

[0080] is the strength and state of the node at time t;

[0081] and is the attenuation factor , used to balance the impact of new and old information;

[0082] and They are functions for calculating node weights and edge connection strengths, respectively;

[0083] Is a function that calculates the node At time t, the context the extent of the impact;

[0084] is a parameter that controls the degree of contextual influence;

[0085] It is the feature vector that is finally input into the deep learning model, including the updated node weights, edge connection strengths, and external context information.

[0086] Through the design of the above formula, the deep learning model reasoning module 200 can overcome its limitations in capturing changes in users' immediate behavior and integrating multi-dimensional contextual information. CAGS dynamically updates node weights and edge connection strengths, ensuring that the model can reflect the user's latest behavior patterns in real time, rather than relying solely on static historical data. In addition, it also incorporates external contextual information such as geographic location, timestamp, weather conditions, etc. into node weights, improving the accuracy of understanding the user's specific situation. These improvements make personalized recommendation decisions based on CAGS more accurate, and can provide the most suitable advertising content based on the user's current most realistic state, solving the problem that traditional models are difficult to accurately capture users' immediate behaviors and complex contextual relationships.

[0087] In order to further improve the accuracy and timeliness of advertising, the system not only needs to capture the user's immediate interests and specific situations, but also needs a dedicated decision-making mechanism to determine when, where, and how to display the most appropriate advertising content. This is because even if you have high-quality user feature representations, if you cannot make the best advertising recommendation decisions based on these features, you still cannot maximize the advertising effect and user experience. Therefore, we introduce a scenario adaptation decision module 300, which makes the final advertising recommendation decision based on the processed feature vectors.

[0088] Specifically, CAGS generates a more comprehensive multi-dimensional user behavior spectrum through dynamic updating and integration, and transmits the updated multi-dimensional user behavior spectrum to the scenario adaptation decision module 300. This process ensures that the scenario adaptation decision module 300 receives the latest data that best reflects the user's immediate status, so that it can more accurately select the most suitable advertising content and optimize the time and location of advertising display. In this way, not only the accuracy of advertising delivery is improved, but also the user experience is significantly enhanced, ensuring that the advertisement is displayed to the user in the most appropriate context.

[0089] See also Figure 3 As shown, the scenario adaptation decision module 300 is responsible for making the best advertising display decision based on the user's status and specific situation. This module receives the updated multi-dimensional user behavior spectrum from CAGS, combines it with real-time context information (such as geographic location, timestamp, etc.), and determines the most appropriate advertising content and its display strategy through a series of decision logic. To achieve this, the module uses the following formula to perform user status judgment and advertising decision:

[0090] ;

[0091] refers to the optimal set of advertisements;

[0092] S refers to the set of all available ads;

[0093] N refers to the number of different scenarios that the system considers during its decision-making process;

[0094] Refers to the probability of a user's interest in the i-th advertisement, i.e., the value passed by the above softmax function;

[0095] Refers to advertisement A in the context The following relevance score.

[0096] In order to more accurately control advertising push, the scenario adaptation decision module 300 has an advanced user status recognition function, which determines whether to allow advertising push by monitoring the user's various behavioral signals and device status. The module detects screen usage to determine whether the user is actively using the device, such as browsing the web, playing games, or watching videos; at the same time, it analyzes application usage patterns to understand the type of application currently used by the user to determine whether he is working or entertaining. In addition, using device sensor data such as accelerometers and gyroscopes, it can be determined whether the user is moving or stationary, and combined with time period and location information, it can infer the user's possible behavior patterns, such as commuting time or resting at home.

[0097] Based on these data, the module sets corresponding thresholds to evaluate whether the user is in an active state. When the screen is not operated or locked for a long time, and the user is at home or at a permanent location and the time is not working hours, and the device is stationary, the system believes that the user may be resting or waiting, which is defined as the user's idle state, allowing relevant advertisements to be pushed. On the contrary, if the screen is continuously active, indicating that the user is concentrating on using the device, or the user is using a specific application such as office software or video conferencing tools, and the device sensor shows that the user is moving such as walking or driving, it is considered that the user is in an active state, and the blocking mechanism is activated to avoid inappropriate advertising interruptions. In this way, the scenario adaptation decision module 300 can intelligently judge the user's immediate state and decide whether to push advertisements based on the preset threshold, thereby ensuring that the advertisements are presented to the user in the most appropriate context, maximizing the advertising effect and user experience.

[0098] In order to further optimize the timing and frequency of advertising push, the scenario adaptation decision module 300 transmits the multi-dimensional user behavior spectrum to the dynamic frequency adjustment module 400. By analyzing the data in these behavior spectra, such as the user's active time period, application usage mode and device status, the dynamic frequency adjustment module 400 can intelligently adjust the frequency of advertising display. For example, for those users who frequently interact with advertisements or show a high interest in a particular type of advertisement, the system may appropriately increase the number of advertisement displays; while for those users who show signs of advertising fatigue or are indifferent to advertisements, the frequency of advertisements will be reduced to avoid excessive disturbance. In addition, the module will also consider the user's active time period, identify each user's "golden time", moderately increase the advertising exposure rate during these time periods, and reduce the frequency of advertising push during the user's less active time period. In this way, it ensures that advertisements are displayed to users at the most appropriate frequency in the most appropriate context, maximizing advertising effects and user experience.

[0099] In order to ensure the efficient operation and continuous optimization of the advertising push system, the scenario adaptation decision module 300 transmits real-time performance data and user feedback to the performance monitoring and feedback loop module 500. This module monitors the effect of advertising push in real time, including click-through rate (CTR), conversion rate (CVR) and other key performance indicators (KPI), and optimizes itself based on these data. After each advertisement is pushed, the system collects user feedback, such as whether the advertisement is clicked, how much time is spent on the advertisement link, etc., to evaluate the relevance and attractiveness of the advertisement content. Based on the collected data, the performance monitoring and feedback loop module 500 fine-tunes the scenario adaptation decision and dynamic frequency adjustment strategy. If it is found that certain types of advertisements perform particularly well in specific situations, the system will give priority to these advertisements for push; on the contrary, if some advertisements continue to perform poorly, their push frequency will be automatically reduced or even removed from the recommendation list. In addition, the module will also pay attention to the changing trends of user behavior, update user portraits and preference settings in a timely manner, and ensure that advertising push is always kept up to date and most relevant. Ultimately, this helps build a healthier, more sustainable advertising ecosystem, increasing both marketer return on investment (ROI) and improving the overall experience for users.

[0100] In summary, by introducing context-aware graph technology, this system can accurately capture users' immediate behaviors and complex contextual relationships and efficiently associate the two. This innovation enables advertising recommendations to not only reflect users' long-term interests, but also dynamically adapt to their current status and specific situations, thereby providing more personalized and timely advertising content.

[0101] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A digital advertising system, comprising a data collection preprocessing module (100), wherein the data collection preprocessing module (100) is used to collect multi-dimensional user behavior spectra from multiple channels in real time, and clean and extract features from the data, wherein: Also includes: A deep learning model inference module (200), wherein the deep learning model inference module (200) receives the multi-dimensional user behavior spectrum processed by the data collection preprocessing module (100), calculates the output through forward propagation of a multi-layer neural network, and converts it into a probability distribution through a softmax function, so as to represent the user's interest in different advertisements; The deep learning model inference module (200) constructs and updates a context-aware graph based on the multi-dimensional user behavior spectrum. The context-aware graph is used to capture the user's immediate needs and complex contextual relationships. Each node in the context-aware graph represents a specific user behavior, and the edge represents the correlation between behaviors. The context-aware graph continuously integrates the real-time data stream and historical data from the data acquisition preprocessing module (100), performs forward propagation calculations through the multi-layer neural network, and generates feature representations. The feature representations are then passed to the softmax function in the deep learning model inference module (200) and converted into a new probability distribution. The probability distribution is further passed to the scenario adaptation decision module (300) for determining advertising content and delivering the determined advertising content; in: The context-aware graph dynamically updates node weights and edge connection strengths to allow the model to reflect the user's latest behavior patterns in real time; The context-aware graph converts external context information It is integrated into the node weight to enhance the accuracy of understanding the user's specific situation. The specific formula is as follows: ； Represents the value or attribute of node i, and node i represents a specific node in the graph; is the weight of node i at time t; is the updated weight of node i at time t; is the external context information of node i at time t the extent of the impact; is a parameter that controls the degree of context influence.

2. The digital advertising system according to claim 1, characterized in that: The model that reflects the user's latest behavior pattern in real time is: ； ； in, is the weight of node i at time t-1; is the strength of the edge between node i and node j at time t; is the strength of the edge between node i and node j at time t-1; is the user behavior data collected at the current time t; is the edge between node i and node j, which represents the association or interaction strength between the two nodes; and is the attenuation factor , used to balance the impact of new and old information; and They are functions for calculating node weights and edge connection strengths, respectively.

3. The digital advertising system according to claim 1, characterized in that: The deep learning model inference module (200) transmits the processed and analyzed multi-dimensional user behavior spectrum and the context perception map to the scenario adaptation decision module (300) for further scenario adaptation decision.

4. The digital advertising system according to claim 3, characterized in that: The scenario adaptation decision module (300) combines the current geographical location and time and uses decision logic to determine the most appropriate advertising content and its display strategy.

5. The digital advertising system according to claim 4, characterized in that: The scenario adaptation decision module (300) analyzes the real-time performance data and user feedback from the display strategy, and determines whether to allow advertisement push by monitoring various user behavior signals and device status. If the user is determined to be in a non-idle state, a blocking mechanism is adopted.

6. The digital advertising system according to claim 5, characterized in that: If the scenario adaptation decision module (300) determines that the user is in an idle state, the multi-dimensional user behavior spectrum is transmitted to the dynamic frequency adjustment module (400).

7. The digital advertising system according to claim 6, characterized in that: The dynamic frequency adjustment module (400) identifies the prime time of each user based on the user's behavior pattern, and optimizes the advertisement push frequency in a personalized manner according to the user's interaction history and current status.

8. The digital advertising system according to claim 7, characterized in that: The scenario adaptation decision module (300) transmits the real-time performance data and user feedback to the performance monitoring and feedback loop module (500).

9. The digital advertising system according to claim 8, characterized in that: The performance monitoring and feedback loop module (500) monitors and analyzes the click-through rate, conversion rate and other key performance indicators of the advertisement in real time, and feeds them back to the scenario adaptation decision module (300) and the dynamic frequency adjustment module (400).

Images