Advertisement layered delivery bidding method and system based on user behaviors
By constructing user behavior feature vectors and deep neural network models, combined with an improved K-means++ clustering algorithm, user segmentation and dynamic bidding were achieved. This solved the problems of inaccurate user activation probability judgment and low resource allocation efficiency in the advertising system, and improved the ROI and cost control of advertising.
Patent Information
- Application Number
- CN202511239703.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-01
- Publication Date
- 2025-12-09
AI Technical Summary
In existing advertising systems, the audience scoring is coarse-grained, the bidding mechanism lacks flexibility, and the cost control methods are crude. This leads to inaccurate judgment of user activation probability, low resource allocation efficiency, and an inability to achieve the optimal bidding strategy of high bids for high-potential users and low bids for low-potential users.
By acquiring user behavior data in real time, constructing user behavior feature vectors, using deep neural network models to predict user activation probabilities, combining improved K-means++ clustering algorithms to segment users, and calculating bid adjustment coefficients through multi-objective optimization functions to achieve dynamic bidding.
It improved the accuracy of user activation probability prediction, reduced user acquisition costs, increased ROI, and achieved more efficient allocation of advertising resources.
Smart Images
Figure BDA0005576176800000071 
Figure BDA0005576176800000081 
Figure BDA0005576176800000091
Abstract
Description
Technical Field
[0001] This application relates to the field of real-time bidding technology, and more specifically, to a method and system for tiered advertising bidding based on user behavior. Background Technology
[0002] In the era of fierce competition for existing users in the mobile internet sector, advertising has become a core means for apps to acquire new users and reactivate churned users. However, the industry faces a series of challenges in actual operation, especially the increasingly prominent issues of high customer acquisition costs and inconsistent user engagement quality. In-depth analysis reveals that the root causes of these problems mainly include the following:
[0003] Coarse-grained user scoring: Most traditional scoring models rely on lightweight algorithms, which are difficult to fully depict user behavior patterns and potential value, resulting in inaccurate judgment of user activation probability and inability to guarantee user activation effect.
[0004] The bidding mechanism lacks flexibility: existing bidding rules are generally rigid and cannot flexibly respond to changes in budget rhythm, CPA control requirements, and ROI improvement targets, resulting in low resource allocation efficiency.
[0005] The cost control methods are primitive and extensive: the advertising system lacks a refined perception of user value, making it difficult to achieve the optimal bidding strategy of "high bid for high-potential users and low bid for low-potential users".
[0006] Against this backdrop, there is an urgent need for an intelligent bidding mechanism that balances the accuracy of behavioral profiling with real-time responsiveness, in order to support more efficient and ROI-oriented advertising. Summary of the Invention
[0007] To address the aforementioned technical problems, this invention proposes a user behavior-based tiered advertising bidding method and system.
[0008] The technical solution of this invention is as follows:
[0009] This invention proposes a user behavior-based tiered advertising bidding method, comprising the following steps:
[0010] Real-time acquisition of user behavior data in the App and advertising delivery chain, and real-time storage of user behavior datasets through message queues and databases;
[0011] Transform the behavioral data from the original behavioral dataset into user behavioral feature vectors;
[0012] The user behavior feature vector is used as the input to the deep neural network (DNN) model, and the output is a user activation probability score, which is used to measure the user's potential to respond to advertising.
[0013] Based on user behavior feature vectors and user activation probability scores, users are segmented using an improved K-means++ clustering algorithm to obtain user segmentation results. The user segmentation is used to divide users into different potential levels to support differentiated bidding strategies.
[0014] Based on the user segmentation results, a multi-objective optimization function is used to calculate the bid adjustment coefficient for each segment, and dynamic bidding is carried out according to the bid adjustment coefficient for each segment.
[0015] Preferably, the user's behavioral data in the App and advertising delivery chain includes: active timestamps, ad click feedback records, in-App registration behavior, device attribute information, and ad interaction records.
[0016] Preferably, the behavioral data is collected through front-end tracking, server-side log collection, and third-party advertising platform APIs, and the data is warehoused using Apache Pulsar message queues and ODPS data warehouses.
[0017] Preferably, the DNN model uses an MLP (Multilayer Perceptron) to capture the nonlinear relationships between user behavior features through multiple fully connected layers and nonlinear activation functions; and uses binary cross-entropy as the loss function and the Adam optimizer for model training.
[0018] Preferably, the improved K-means++ clustering algorithm strengthens the weight of user activation probability scores in the clustering process by introducing weighted Euclidean distance; and confirms the hierarchical levels by weighted contour coefficients and business requirements.
[0019] Preferably, the multi-objective optimization function includes:
[0020] Definition of the comprehensive objective: Construct a comprehensive loss function, which consists of a weighted sum of four sub-objectives. The four sub-objectives include:
[0021] The binary cross-entropy loss sub-objective is used to optimize the accuracy of user conversion rate prediction.
[0022] The cost penalty sub-objective aims to suppress samples with excessively high actual customer acquisition costs.
[0023] The ROI optimization sub-objective encourages higher bids for users with high ROI.
[0024] The sub-objective of suppressing customer acquisition costs limits the entry of samples with costs exceeding the budget into the high-potential layer;
[0025] The weight parameters of each sub-target are dynamically adjusted based on the real-time delivery results;
[0026] Tiered bidding coefficient generation: Based on the user segmentation results, the average conversion rate, average return on investment, and average customer acquisition cost are calculated for each tier, and then the bidding adjustment coefficient for that tier is generated through a linear combination formula.
[0027] Final bidding decision: Multiply the base bid value by the bid adjustment factor of the current user's tier to obtain the final bidding price.
[0028] Preferably, the method further includes a closed-loop optimization mechanism, specifically:
[0029] Exposure data collection: Real-time recording of ad exposure events through client-side tracking. Collected fields include exposure timestamp, ad placement identifier, hash value of device identifier, and user segmentation tags;
[0030] Behavioral attribution alignment: By using an attribution system, clicks, registrations, and activations after user exposure are linked to exposure records to form a complete exposure-behavior chain;
[0031] Stratified indicator statistics: The statistics for each layer are aggregated and calculated based on stratified labels, and the core indicators are normalized. These core indicators include:
[0032] Conversion rate: The number of users who converted at this level divided by the total number of impressions at this level;
[0033] Return on investment: Total revenue generated by users in this tier divided by the total advertising cost of this tier;
[0034] Customer acquisition cost: Total advertising cost of this layer divided by the number of users converted in this layer;
[0035] Strategy updates dynamically:
[0036] Regularly updated: The bid adjustment coefficients and objective function weights for each layer are recalculated at a fixed time every day;
[0037] Threshold-triggered update: When the return on investment or customer acquisition cost of a certain layer exceeds a preset threshold, a weight adjustment is immediately triggered;
[0038] Model Iteration and Optimization: The collected exposure-conversion samples are fed back into the deep neural network model for incremental training, and the auxiliary prediction module is updated simultaneously.
[0039] On the other hand, the present invention also provides a user behavior-based tiered advertising bidding system, comprising:
[0040] The dataset construction module acquires user behavior data in real time on the App and in the advertising delivery chain, and builds user behavior datasets by loading them into the warehouse in real time through message queues and databases;
[0041] The feature construction module transforms the behavioral data from the original behavioral dataset into user behavioral feature vectors.
[0042] The user rating module takes user behavior feature vectors as input to a deep neural network (DNN) model and outputs user activation probability ratings.
[0043] The user segmentation module performs user segmentation based on user behavior feature vectors and user activation probability scores using an improved K-means++ clustering algorithm to obtain user segmentation results.
[0044] The bidding adjustment module calculates the bid adjustment coefficient for each level based on the user segmentation results using a multi-objective optimization function, thereby achieving dynamic bidding.
[0045] In another aspect, the present invention also provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements an advertising tiered bidding method based on user behavior as described in any embodiment of the present invention.
[0046] In another aspect, the present invention also provides a computer-readable medium for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement a user behavior-based tiered advertising bidding method as described in any embodiment of the present invention.
[0047] The present invention has the following beneficial effects:
[0048] 1. The DNN model improves the AUC of activation probability prediction by 15%+ by mining nonlinear relationships compared to the traditional logistic regression model. The improved K-means++ clustering algorithm is used to segment users and strengthens the influence of user activation probability scores through weighted clustering, reducing the segmentation error by 25%.
[0049] 2. The multi-objective function adjusts the bidding coefficient in real time to achieve "high bid for high potential users and low bid for low potential users". In actual application, the user group activation cost is reduced by 10%, and the average daily campaign cost is reduced by more than 20,000 yuan.
[0050] 3. The exposure tracking and attribution system provides hourly feedback, replacing traditional A / B testing (daily response), and incremental training of the DNN model improves ROI by 15% over the long term. Detailed Implementation
[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.
[0053] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0054] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.
[0055] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.
[0056] Example 1:
[0057] To address the problems of existing technologies, this invention provides a user behavior-based tiered advertising bidding method, comprising the following steps:
[0058] Real-time acquisition of user behavior data in the App and advertising delivery chain, and real-time storage of user behavior datasets through message queues and databases;
[0059] In this embodiment, user behavior data in the App and ad delivery chain includes:
[0060] Active timestamp: The timestamp of the user's most recent active moment;
[0061] Ad click return records, showing return and activity levels after ad clicks;
[0062] In-app registration behavior, such as initial registration time, path, and content publishing behavior;
[0063] Device attribute information, including device model, operating system version, and region of origin;
[0064] Ad interaction logs are records of user interactions with ad creatives, such as clicks and impressions.
[0065] The user behavior data above is aggregated through front-end tracking, server-side log collection, and third-party advertising platform APIs, and then fed into a warehouse (such as ODPS) in real time via message queues (such as Apache Pulsar) and database binlog to build a unified user behavior dataset.
[0066] The original behavior dataset is transformed into user behavior feature vectors; the specific process is as follows:
[0067] User behavior data is transformed into structured user behavior feature vectors, covering static features, dynamic behavior features, and derived statistical features.
[0068] Example vectors are shown below:
[0069] U = [f1, f2, f3, ..., f n ];
[0070] In the formula: U is the user behavior feature vector; f1 is the number of hours since the last active timestamp; f2 is the number of times ads have been clicked in the last 7 days; f3 is the average retention probability within 1 hour after activation, estimated through historical samples; f n This is the nth eigenvalue in the user behavior feature vector, and its physical meaning is defined by the business logic.
[0071] All features will be normalized using Min-Max or Z-score based on their distribution characteristics to ensure good numerical stability and learning efficiency when input into subsequent DNN models.
[0072] The user behavior feature vector is used as the input to the deep neural network (DNN) model, and the output is a user activation probability score, which is used to measure the user's potential to respond to advertising.
[0073] In this embodiment, the DNN model uses an MLP (Multilayer Perceptron) with multiple fully connected layers and nonlinear activation functions (such as ReLU) to fully capture the nonlinear relationships between user behavior features.
[0074] The model takes a normalized user behavior feature vector as input and outputs a user activation probability score in the range [0,1]. Model training is based on historical real conversion data, using binary cross-entropy as the loss function and the Adam optimizer for backpropagation optimization.
[0075] Deployed in an online service, the model has millisecond-level response capabilities and can provide real-time scoring and prediction for new users or devices, providing a strong discriminative signal for subsequent hierarchical clustering.
[0076] Based on user behavior feature vectors and user activation probability scores, users are segmented using an improved K-means++ clustering algorithm to obtain user segmentation results. The user segmentation is used to divide users into different potential levels to support differentiated bidding strategies.
[0077] In this embodiment, after obtaining the user activation probability score, a more granular user segmentation is achieved through an improved K-means++ clustering algorithm. This algorithm uses user behavior feature vectors and user activation probability scores as clustering inputs, solving problems in traditional methods such as "scores cannot participate in clustering" and "initial centers are prone to getting trapped in local optima".
[0078] The input vector for the improved K-means++ clustering algorithm is:
[0079] X user =[f1,f2,f3,…,f n ,s];
[0080] In the formula: X user s is the input vector for the improved K-means++ clustering algorithm; s is the user activation probability score output by the DNN model.
[0081] To enhance the influence of user activation probability scores on grouping, we introduce a "weighted Euclidean distance" metric, defined as follows:
[0082]
[0083] In the formula: a i b i S represents the values of users A and B on the i-th behavioral feature dimension, respectively; A S B α and β are the user activation probability scores for users A and B, respectively; α and β are the weighting coefficients of the user behavior feature vector and the user activation probability score, respectively. In this embodiment, α = 0.7 and β = 0.3. α and β are adjustable parameters.
[0084] Furthermore, when initializing centroids, high-density, high-scoring representative samples are prioritized to improve clustering stability and interpretability. The optimal number of clustering layers is automatically selected using weighted silhouette coefficients, supporting arbitrary number of layers for grouping according to business needs, far exceeding the coarse granularity of the traditional "three-layer grouping" approach.
[0085] The following is an example of the structure output by the improved K-means++ clustering algorithm:
[0086]
[0087] Based on the user segmentation results, a multi-objective optimization function is used to calculate the bid adjustment coefficient for each segment, and dynamic bidding is carried out according to the bid adjustment coefficient for each segment.
[0088] In this embodiment, the clustering stratification results are pushed by Pulsar and subject to real-time bidding. The specific bidding formula is as follows:
[0089] Bid final=BaseBid×α c ;
[0090] In the formula: Bid final The final bid; BaseBid is the base bid; α c For the bid adjustment coefficient of users in layer c, a multi-objective function is introduced for dynamic adjustment.
[0091] The multi-objective optimization function includes:
[0092] Definition of the comprehensive objective: Construct a comprehensive loss function, which consists of a weighted sum of four sub-objectives. The four sub-objectives include:
[0093] The binary cross-entropy loss sub-objective is used to optimize the accuracy of CVR prediction.
[0094] The cost penalty sub-objective aims to suppress samples with excessively high actual customer acquisition costs.
[0095] The ROI optimization sub-objective encourages higher bids for users with high ROI.
[0096] The sub-objective of suppressing customer acquisition costs limits the entry of samples with costs exceeding the budget into the high-potential layer;
[0097] The weight parameters of each sub-target are dynamically adjusted based on the real-time delivery results;
[0098] The specific comprehensive loss function is as follows:
[0099] L total =λ1·L bce +λ2·L cost +λ3·L roi +λ4·L cac ;
[0100] In the formula: L total The value of the comprehensive loss function; L bce Focusing on CVR accuracy, a binary cross-entropy loss based on DNN output ratings is used; L cost This is a cost penalty item to prevent the score from being too high while the actual CPA is too high; L roi To encourage high-ROI users to increase their bids; L cac To suppress CAC over-budget user samples from entering the high-potential layer; λ1~λ4 are weight parameters for each sub-objective that are dynamically adjusted based on real-time feedback.
[0101] When combining the results of hierarchical clustering, we introduce a hierarchy-aware bidding objective function:
[0102] α c =f(CVR) c ROI c CACc );
[0103] For example, a feasible linear combination model is:
[0104]
[0105] γ1 represents the average of standardized indicators within the c-th layer group; γ2, γ3 are adjustable parameters on the business side, supporting strategy control.
[0106] This formula feeds back the actual performance of each group after clustering to the bidding weight configuration, realizing hierarchical dynamic pricing with multi-objective perception.
[0107] To replace traditional A / B testing, this invention proposes a closed-loop optimization mechanism of "online sample tracking + automatic feedback optimization," which improves the bidding model through real-time feedback. Specifically:
[0108] Exposure data collection: Real-time recording of ad exposure events through client-side tracking. Collected fields include exposure timestamp, ad placement identifier, hash value of device identifier, and user segmentation tags;
[0109] Behavioral attribution alignment: By using an attribution system, clicks, registrations, and activations after user exposure are linked to exposure records to form a complete exposure-behavior chain;
[0110] Stratified indicator statistics: The statistics for each layer are aggregated and calculated based on stratified labels, and the core indicators are normalized. These core indicators include:
[0111] Conversion rate: The number of users who converted at this level divided by the total number of impressions at this level;
[0112] Return on investment: Total revenue generated by users in this tier divided by the total advertising cost of this tier;
[0113] Customer acquisition cost: Total advertising cost of this layer divided by the number of users converted in this layer;
[0114] The above metrics will be dynamically calculated according to a specified window period (e.g., the past 24 hours) and standardized using Min-Max / Z-score to obtain... These are input features used for subsequent bidding function optimization.
[0115] Strategy updates dynamically:
[0116] Regularly updated: The bid adjustment coefficients and objective function weights for each layer are recalculated at a fixed time every day;
[0117] Threshold-triggered update: When the return on investment or customer acquisition cost of a certain layer exceeds a preset threshold, a weight adjustment is immediately triggered;
[0118] Model iterative optimization: The collected exposure-conversion samples are fed back into the deep neural network model for incremental training, and the auxiliary prediction module is updated simultaneously, specifically as follows:
[0119] The collected tracking samples are also fed back as training data for:
[0120] Incremental fine-tuning of the DNN user rating model (e.g., fine-tuning or online learning);
[0121] Used to update auxiliary prediction modules (such as ROI predictor and CVR calibrator);
[0122] The relative contribution of each indicator in the multi-objective loss function is dynamically evaluated, and the weight combination is optimized.
[0123] By combining the incremental sample management mechanism of streaming data platforms (such as Flink+ODPS), high-quality training data can be continuously accumulated, ensuring the long-term robustness of the model and the stability of the policy effect.
[0124] Example 2:
[0125] This embodiment provides a user behavior-based tiered advertising bidding system, including:
[0126] The dataset construction module acquires user behavior data in real time on the App and in the advertising delivery chain, and builds user behavior datasets by loading them into the warehouse in real time through message queues and databases;
[0127] The feature construction module transforms the behavioral data from the original behavioral dataset into user behavioral feature vectors.
[0128] The user rating module takes user behavior feature vectors as input to a deep neural network (DNN) model and outputs user activation probability ratings.
[0129] The user segmentation module performs user segmentation based on user behavior feature vectors and user activation probability scores using an improved K-means++ clustering algorithm to obtain user segmentation results.
[0130] The bidding adjustment module calculates the bid adjustment coefficient for each level based on the user segmentation results using a multi-objective optimization function, thereby achieving dynamic bidding.
[0131] Example 3:
[0132] This embodiment provides an electronic device that stores a computer program. When the computer program is executed by a processor, it implements a user behavior-based tiered advertising bidding method as described in any embodiment of the present invention.
[0133] Example 4:
[0134] This embodiment provides a computer-readable medium for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement a user behavior-based tiered advertising bidding method as described in any embodiment of the present invention.
[0135] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.
[0136] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0138] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code. The above descriptions are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made using the present invention specification, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A user behavior-based tiered advertising bidding method, characterized in that, Includes the following steps: Real-time acquisition of user behavior data in the App and advertising delivery chain, and real-time storage of user behavior datasets through message queues and databases; Transform the behavioral data from the original behavioral dataset into user behavioral feature vectors; The user behavior feature vector is used as the input to the deep neural network (DNN) model, and the output is a user activation probability score, which is used to measure the user's potential to respond to advertising. Based on user behavior feature vectors and user activation probability scores, users are segmented using an improved K-means++ clustering algorithm to obtain user segmentation results. The user segmentation is used to divide users into different potential levels to support differentiated bidding strategies. Based on the user segmentation results, a multi-objective optimization function is used to calculate the bid adjustment coefficient for each segment, and dynamic bidding is carried out according to the bid adjustment coefficient for each segment.
2. The user behavior-based tiered advertising bidding method according to claim 1, characterized in that: The user behavior data in the App and advertising delivery chain includes: active timestamps, ad click feedback records, in-app registration behavior, device attribute information, and ad interaction records.
3. The user behavior-based tiered advertising bidding method according to claim 2, characterized in that: The behavioral data is collected through front-end tracking, server-side log collection, and third-party advertising platform APIs. The data is then stored in the Apache Pulsar message queue and ODPS data warehouse.
4. The user behavior-based tiered advertising bidding method according to claim 1, characterized in that: The DNN model uses an MLP (Multilayer Perceptron) to capture the nonlinear relationships between user behavior features through multiple fully connected layers and nonlinear activation functions. The model was trained using binary cross-entropy as the loss function and the Adam optimizer.
5. The user behavior-based tiered advertising bidding method according to claim 1, characterized in that: The improved K-means++ clustering algorithm strengthens the weight of user activation probability scores in the clustering process by introducing weighted Euclidean distance; and confirms the hierarchical levels by using weighted contour coefficients and business requirements.
6. The user behavior-based tiered advertising bidding method according to claim 1, characterized in that: The multi-objective optimization function includes: Definition of the comprehensive objective: Construct a comprehensive loss function, which consists of a weighted sum of four sub-objectives, including: The binary cross-entropy loss sub-objective is used to optimize the accuracy of user conversion rate prediction. The cost penalty sub-objective aims to suppress samples with excessively high actual customer acquisition costs. The ROI optimization sub-objective encourages higher bids for users with high ROI. The sub-objective of suppressing customer acquisition costs limits the entry of samples with costs exceeding the budget into the high-potential layer; The weight parameters of each sub-target are dynamically adjusted based on the real-time delivery results; Tiered bidding coefficient generation: Based on the user segmentation results, the average conversion rate, average return on investment, and average customer acquisition cost are calculated for each tier, and then the bidding adjustment coefficient for that tier is generated through a linear combination formula. Final bidding decision: Multiply the base bid value by the bid adjustment factor of the current user's tier to obtain the final bidding price.
7. The user behavior-based tiered advertising bidding method according to claim 1, characterized in that: The method also includes a closed-loop optimization mechanism, specifically: Exposure data collection: Real-time recording of ad exposure events through client-side tracking. Collected fields include exposure timestamp, ad placement identifier, hash value of device identifier, and user segmentation tags; Behavioral attribution alignment: By using an attribution system, clicks, registrations, and activations after user exposure are linked to exposure records to form a complete exposure-behavior chain; Stratified indicator statistics: The statistics for each layer are aggregated and calculated based on stratified labels, and the core indicators are normalized. These core indicators include: Conversion rate: The number of users who converted at this level divided by the total number of impressions at this level; Return on investment: Total revenue generated by users in this tier divided by the total advertising cost of this tier; Customer acquisition cost: Total advertising cost of this layer divided by the number of users converted in this layer; Strategy updates dynamically: Regularly updated: The bid adjustment coefficients and objective function weights for each layer are recalculated at a fixed time every day; Threshold-triggered update: When the return on investment or customer acquisition cost of a certain layer exceeds a preset threshold, a weight adjustment is immediately triggered; Model Iteration and Optimization: The collected exposure-conversion samples are fed back into the deep neural network model for incremental training, and the auxiliary prediction module is updated simultaneously.
8. A user behavior-based tiered advertising bidding system, characterized in that, include: The dataset construction module acquires user behavior data in real time on the App and in the advertising delivery chain, and builds user behavior datasets by loading them into the warehouse in real time through message queues and databases; The feature construction module transforms the behavioral data from the original behavioral dataset into user behavioral feature vectors. The user rating module takes user behavior feature vectors as input to a deep neural network (DNN) model and outputs user activation probability ratings. The user segmentation module performs user segmentation based on user behavior feature vectors and user activation probability scores using an improved K-means++ clustering algorithm to obtain user segmentation results. The bidding adjustment module calculates the bid adjustment coefficient for each level based on the user segmentation results using a multi-objective optimization function, thereby achieving dynamic bidding.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a user behavior-based tiered advertising bidding method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a user behavior-based tiered advertising bidding method as described in any one of claims 1 to 7.