Advertisement bidding system and device

Abstract

The embodiment of the invention provides an advertisement bidding system and equipment. The advertisement bidding system comprises a steady-state traffic price module which is used for simulating an association relationship among advertisement traffic, advertisement bidding and advertisement cost in a first advertisement traffic scene, and giving an advertisement budget of a next time slice according to input advertisement feature data; the unsteady flow price module is used for simulating an association relationship among the advertisement flow, the advertisement bid and the advertisement cost in a second advertisement flow scene, and giving an advertisement budget of a next time slice according to the input advertisement feature data; and the bid regulation and control module is used for sending an advertisement quotation to an advertisement platform according to the advertisement budget and specified target data provided by the steady-state traffic price module or the unsteady-state traffic price module. According to the advertisement bidding system and device, the adaptability of the advertisement bidding system to various advertisement traffic scenes can be improved to a certain extent.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of software development technology, and more particularly to an advertising bidding system and device. Background Art

[0002] In the field of Internet advertising, the advertising bidding system is an important tool for advertising platforms to achieve accurate delivery and cost optimization. However, when applied to different advertising traffic scenarios, the existing advertising bidding system often lacks sufficient adaptability and flexibility, and it is difficult to effectively cope with the complex changes in different traffic scenarios.

[0003] Therefore, the existing advertising bidding system has the problem of being difficult to apply to various advertising traffic scenarios. Summary of the invention

[0004] In view of this, multiple implementations of the present application are dedicated to providing an advertising bidding system and device, which can improve the adaptability of the advertising bidding system to various advertising traffic scenarios to a certain extent.

[0005] One embodiment of the present application provides an advertising bidding system, which includes: a steady-state traffic price module, which is used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in a first advertising traffic scenario, and provide an advertising budget for the next time slice based on input advertising feature data; a non-steady-state traffic price module, which is used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in a second advertising traffic scenario, and provide an advertising budget for the next time slice based on input advertising feature data; wherein the stability of advertising traffic in the second advertising traffic scenario is weaker than the stability of advertising traffic in the first advertising traffic scenario; and a bid regulation module, which is used to send advertising quotations to an advertising platform based on the advertising budget and specified target data provided by the steady-state traffic price module or the non-steady-state traffic price module.

[0006] One embodiment of the present application provides a computer device, which includes a memory and a processor. The memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the aforementioned advertising bidding system.

[0007] In multiple implementations provided in this application, the association between advertising traffic, advertising bids, and advertising expenses in the first advertising traffic scenario and the second advertising traffic scenario is simulated through a steady-state traffic price module and a non-steady-state traffic price module, respectively, and the advertising budget for the next time slice is provided according to the input advertising feature data; at the same time, in combination with the bid regulation module, an advertising quotation is sent to the advertising platform according to the budget and designated target data of the steady-state traffic price module or the non-steady-state traffic price module, so that the advertising bidding system can effectively adapt to the fluctuating characteristics of advertising traffic in different advertising traffic scenarios. In this way, the adaptability of the advertising bidding system to a variety of advertising traffic scenarios is achieved, further improving the flexibility and effectiveness of advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A module diagram of an advertising bidding system provided for one embodiment of the present application.

[0009] Figure 2 A module diagram of an advertisement bidding system provided for one embodiment of the present application; wherein a non-steady-state uncertainty estimation submodule is provided.

[0010] Figure 3 A module diagram of an advertising bidding system provided for one embodiment of the present application; wherein a steady-state uncertainty estimation submodule is provided.

[0011] Figure 4 A schematic diagram of the modules of an advertisement bidding system provided for one embodiment of the present application; wherein a steady-state optimization submodule is provided.

[0012] Figure 5 A module diagram of an advertisement bidding system provided for one embodiment of the present application; wherein a non-steady-state optimization submodule is provided.

[0013] Figure 6 A module diagram of an advertising budget-bidding model in a steady-state traffic price module provided for one embodiment of the present application.

[0014] Figure 7 A module diagram of a bid control module provided for one embodiment of the present application.

[0015] Figure 8 A confidence curve diagram is provided for one embodiment of the present application.

[0016] Fig. 9 A schematic diagram of a computer device provided for one embodiment of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0018] In the description of the embodiments of the present application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0019] In the related technology, in the field of Internet advertising, traditional advertising bidding systems usually use historical advertising data and fixed rules to predict and regulate advertising budgets. Specifically, these advertising bidding systems model the correlation between advertising traffic, advertising bids, and advertising expenses, and predict future advertising budgets based on historical advertising feature data. Although this method can meet the advertising budget needs in steady-state scenarios to a certain extent, in non-steady-state scenarios with large traffic fluctuations, due to the significant changes in the characteristics of advertising traffic, traditional methods are difficult to adapt to the dynamic changes of complex scenarios. In addition, the rule settings of traditional advertising bidding systems are often fixed and single, and lack the ability to flexibly respond to a variety of advertising scenarios, resulting in unsatisfactory budget accuracy and bidding effects in different advertising scenarios.

[0020] With the development of Internet advertising technology, the distinction between steady-state and non-steady-state advertising scenarios has gradually gained attention. Steady-state scenarios refer to situations where advertising traffic is relatively stable and traffic fluctuations are small, while non-steady-state scenarios refer to situations where advertising traffic is highly volatile and has greater uncertainty. In non-steady-state scenarios, the range of advertising traffic changes is wide, and the modeling method based on the steady-state assumption in traditional systems is difficult to effectively capture these changing characteristics, resulting in deviations in budget allocation and bidding strategies.

[0021] In addition, the advertising bidding system in the related art can usually only generate and regulate budgets for a single scenario, and lacks the ability to flexibly adapt to the needs of multiple scenarios. This limitation is particularly prominent when the types of advertisements are diverse, especially when it is necessary to distinguish between short-lifecycle advertisements (such as promotional campaign advertisements) and long-term brand advertisements. The singleness and inflexibility of the advertising bidding system in the related art limits the efficiency and effectiveness of the advertising platform.

[0022] In summary, the related technology still has the problem that the advertising bidding system is difficult to adapt to different advertising traffic scenarios, and further improvements are needed to enhance its adaptability to multiple traffic scenarios.

[0023] In multiple implementations provided in this application, the advertising bidding system can be applied to a computer device with certain computing power and network access capabilities. The computer device can be a desktop computer, a laptop computer, etc., or a server. The server can also be a distributed server, including multiple processors, memories, network communication modules, etc. that work together to achieve various functions. Alternatively, the server can also be a server cluster formed by several servers, with higher computing and data processing capabilities. With the development of science and technology, the server can also be implemented using new forms of technical means, such as a new type of "server" based on quantum computing.

[0024] An embodiment of the present application provides an example of an application scenario of an advertisement bidding system. The advertisement bidding system can be applied to a variety of advertisement traffic scenarios through the coordinated operation of a steady-state traffic price module, a non-steady-state traffic price module and a bid control module.

[0025] For example, in an advertising scenario on an e-commerce platform, advertisers hope to use the advertising bidding system to achieve budget allocation and bid regulation in stable brand promotion scenarios and limited-time promotions with large traffic fluctuations.

[0026] In a stable brand promotion scenario (the first advertising traffic scenario), the steady-state traffic price module can generate an advertising budget based on historical data and advertising feature data. Taking a brand's long-term promotion advertisement as an example, the advertising bidding system can use a multi-task learning model to process the advertising feature data. The tasks include analysis of the correlation between conversion numbers and bids, optimization of exposure and bids, etc. Through the advertising features extracted by the shared network, the steady-state traffic price module derives advertising budget forecasts for different time slices. For example, the advertising budgets that the advertising bidding system can derive are 80 yuan, 90 yuan, and 100 yuan, respectively. On this basis, the steady-state optimization submodule combines the click volume data to further optimize the budget allocation between time slices to maximize the overall click volume. The optimized budget allocation is: the budget for time slice 1 is 85 yuan, the budget for time slice 2 is 95 yuan, and the budget for time slice 3 is 110 yuan.

[0027] For example, in a limited-time promotion scenario (the second advertising traffic scenario), the non-steady-state traffic price module analyzes advertising traffic changes in real time and generates advertising budget forecasts based on the characteristics of drastic traffic fluctuations. For example, in a limited-time promotion, the advertising traffic peak may occur in the middle and late stages of the event. The non-steady-state traffic price module predicts based on historical traffic and real-time traffic data, and derives the advertising budget sequence of the event as follows: the budget for time slice 1 is 50 yuan, the budget for time slice 2 is 70 yuan, and the budget for time slice 3 is 120 yuan. In order to adapt to the rapidly changing advertising traffic, the module ensures the monotonicity of the budget and dynamically adjusts the budget cap. In the case of a sharp increase in advertising traffic, the budget cap is dynamically set to 150 yuan to meet the budget requirements during high-traffic periods.

[0028] In the non-steady-state traffic scenario, in order to cope with the uncertainty of budget forecasting, the non-steady-state uncertainty estimation submodule is enabled. The system analyzes historical sample data similar to the current advertising characteristics to obtain the first uncertainty coefficient reflecting data fluctuations, and uses the actual bid data of the previous bidding cycle to calculate the distribution uncertainty and obtain the second uncertainty coefficient. Combining the two, the system calculates the target uncertainty coefficient. For example, during the delivery of a promotional advertisement, the target uncertainty coefficient was determined to be 0.7, indicating that there is a moderate degree of uncertainty in the budget forecast. Based on this, the bid control module reduces the proportion of budget usage and adopts a more conservative bidding strategy to reduce the risk of delivery.

[0029] During the advertising process, the bid control module performs weighted smoothing calculations on historical ROI data and PV data to generate target ROI data and target PV data. For example, the target ROI obtained by system analysis is 1.5 and the target PV is 8000. The bid control module dynamically adjusts the advertising quotation based on these target data, as well as the budget cap and conversion target set by the advertiser. In addition, the bid control module generates a control error and inputs it into the corresponding controller by analyzing the real-time feedback ROI and PV data. Assuming that the current ROI control error is 0.2 and the PV control error is 500, the controller dynamically optimizes the advertising quotation to gradually converge the error.

[0030] In a limited-time promotion, the system uses the non-steady-state optimization submodule to adjust the budget based on the difference between the actual click volume and the target click volume. For example, in the promotion, the target click volume is 1000, 2000 and 3000, while the actual click volume is 900, 1800 and 2800. After the non-steady-state optimization submodule adjusts the budget allocation, the difference in click volume is further reduced, and the actual click volume after optimization is increased to 950, 1950 and 2950 respectively.

[0031] Through this scenario example, the advertising bidding system uses the coordinated work of the steady-state traffic price module and the non-steady-state traffic price module, combined with the optimization sub-module and uncertainty evaluation mechanism, to accurately adapt to different advertising traffic scenarios and dynamically adjust the budget and advertising quotation. In this way, the advertising bidding system can effectively improve the advertising effect and budget utilization efficiency of advertisers in a complex and changing advertising environment.

[0032] See also Figure 1 The embodiment of the present application provides an advertisement bidding system, which includes a steady-state traffic price module, a non-steady-state traffic price module and a bid regulation module.

[0033] The steady-state traffic price module is used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in the first advertising traffic scenario, and to provide the advertising budget for the next time slice based on the input advertising feature data.

[0034] The non-steady-state traffic price module is used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in a second advertising traffic scenario, and to provide an advertising budget for the next time slice based on input advertising feature data; wherein the stability of advertising traffic in the second advertising traffic scenario is weaker than the stability of advertising traffic in the first advertising traffic scenario.

[0035] The bid regulation module is used to send an advertisement quotation to an advertisement platform according to the advertisement budget and designated target data provided by the steady-state traffic price module or the non-steady-state traffic price module.

[0036] In this embodiment, the advertising bidding system achieves efficient adaptation to different advertising traffic scenarios through the coordinated work of the steady-state traffic price module, the non-steady-state traffic price module and the bid regulation module.

[0037] In some embodiments, it is assumed that the advertiser regularly interacts with the advertising bidding system within a specified time range, and the specified time range may include T time slices, each time slice being a cycle. For example, each time slice is 1 hour. In each time slice t, the advertiser specifies a cpx bid x t , to inform the ad bidding system of course, and then the ad bidding system automatically performs ad purchases. When the time slice t ends, advertisers can observe the aggregated ad performance, such as pv t ,cost t ,click t ,gmv t Etc. For each time slice, the goal of the ad bidding system is to find the value of Cpx to maximize the overall revenue (e.g., clicks, total merchandise volume) while satisfying multiple constraints. These constraints can be related to the efficiency and scale of ad purchases, and there are two constraints that can be followed in the ad bidding system. For example, the goal of the ad bidding system is to maximize clicks, and Cpx is specified as the maximum cost per click (CPC). The constraint constructed with the goal of maximizing clicks can be the following formula.

[0038]

[0039] where click t,i ,cost t,i Given cpc=x t,i When , the expected click volume and cost of , M is the total number of advertising campaigns, B is the total budget, and θ is the highest affordable CPC for the advertiser.

[0040] In this embodiment, the steady-state traffic price module can be used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in the first advertising traffic scenario. The first advertising traffic scenario has high stability, that is, within a certain time range, the fluctuation of its advertising traffic is small, such as the search advertising scenario, which is suitable for building a steady-state traffic model through historical data and advertising feature data. The steady-state traffic price module predicts the advertising budget for the next time slice based on the input advertising feature data (such as target audience characteristics, historical click-through rate, historical exposure, etc.). Specifically, the advertising feature data may include information such as advertising delivery areas, delivery time periods, user behavior preferences, etc. The steady-state traffic price module combines these features and uses modeling methods such as regression analysis and time series prediction to provide a high-accuracy estimate of the advertising budget.

[0041] In this embodiment, the non-steady-state traffic price module can be used to simulate the correlation between advertising traffic, advertising bids and advertising expenses in the second advertising traffic scenario. The traffic volatility of the second advertising traffic scenario is significantly higher than that of the first advertising traffic scenario. For example, in short video scenarios, major promotional activities, and traffic scenarios driven by hot topic events, the rapid change characteristics of traffic put forward higher requirements for the prediction of advertising budgets. The non-steady-state traffic price module constructs a non-steady-state traffic model through reinforcement learning, dynamic feedback and other technologies to cope with the instability of traffic, and predicts the advertising budget for the next time slice in combination with advertising feature data. The non-steady-state traffic price module can also adjust the budget prediction strategy in real time to adapt to the characteristics of rapid changes in traffic. In a specific embodiment, the non-steady-state traffic price module can construct an advertising budget-bid relationship model based on GBRT (GradientBoosting Regression Tree).

[0042] In this embodiment, the bid control module is used to convert the advertising budget of the steady-state traffic price module and the non-steady-state traffic price module into a specific advertising bidding strategy, and send advertising quotations to the advertising platform according to the specified target data (such as the advertiser's delivery budget upper limit, expected conversion rate, target audience coverage, etc.). The bid control module can adopt a PID control algorithm or other dynamic control strategies, combined with the error between the advertiser's goal expected to be achieved by the advertising budget and the actual delivery effect, gradually optimize the advertising bid, and improve the consistency between the advertising delivery effect and the advertiser's goal. Specifically, the bid control module can adjust the bid in a short period of time according to the real-time feedback of the advertising platform to improve core indicators such as advertising click-through rate and exposure.

[0043] In some implementations, in typical applications of the advertising bidding system, the steady-state traffic price module is suitable for stable delivery scenarios such as long-term brand advertising, such as the long-term display of a product's brand promotion advertisement among the target user group; while the non-steady-state traffic price module is more suitable for short-life cycle advertising delivery scenarios, such as short videos, holiday promotional advertisements, limited-time discount event advertisements and other high-traffic fluctuation scenarios. In actual operation, the advertising bidding system can select the applicable traffic price module according to the type of advertising traffic scenario, and the bid control module connects the budget with the specific bidding strategy to form a complete advertising delivery closed loop.

[0044] In multiple implementations provided in this application, the association between advertising traffic, advertising bids, and advertising expenses in the first advertising traffic scenario and the second advertising traffic scenario is simulated through a steady-state traffic price module and a non-steady-state traffic price module, respectively, and the advertising budget for the next time slice is provided according to the input advertising feature data; at the same time, in combination with the bid regulation module, an advertising quotation is sent to the advertising platform according to the budget and designated target data of the steady-state traffic price module or the non-steady-state traffic price module, so that the advertising bidding system can effectively adapt to the fluctuating characteristics of advertising traffic in different advertising traffic scenarios. In this way, the adaptability of the advertising bidding system to a variety of advertising traffic scenarios is achieved, further improving the flexibility and effectiveness of advertising delivery.

[0045] In some implementations, the advertising budget provided by the non-steady-state traffic price module to the bid regulation module complies with a monotonic constraint and has a variable upper limit.

[0046] In this embodiment, the advertising budget provided by the non-steady-state traffic price module to the bid regulation module has a monotonic constraint and a variable upper limit to adapt to the significant traffic fluctuations in non-steady-state traffic scenarios, so that the prediction of the advertising budget can meet the requirements of real-time and rationality.

[0047] Specifically, in non-steady-state traffic scenarios, advertising traffic is highly volatile, such as short video ads, limited-time promotional ads, or ads driven by sudden hot events. Advertising budgets need to respond quickly to changes in traffic to optimize delivery effects. The non-steady-state traffic price module can build an advertising budget prediction model that meets monotonic constraints through dynamic optimization strategies based on comprehensive analysis of historical data and real-time data.

[0048] In this embodiment, the monotonic constraint can be implemented by making the advertising budget present a monotonically increasing or unchanged trend in continuous time slices, thereby avoiding the decrease in delivery efficiency caused by the drastic fluctuation of the advertising budget. In addition, the non-steady-state traffic price module combines the traffic characteristics and delivery goals of the advertising scene to set a variable upper limit for the advertising budget.

[0049] In this embodiment, the non-steady-state traffic price module can adjust the budget according to the real-time advertising feature data by combining the reinforcement learning model or regression model through a dynamic feedback mechanism. For example, based on the main indicators such as the click-through rate and conversion rate of the advertisement and reference indicators such as the number of favorites and the number of additional purchases. The non-steady-state traffic price module can dynamically correct the budget forecast results through weight calculation and distribution fitting, and gradually optimize the budget allocation under the condition of satisfying the monotonic constraint.

[0050] In this embodiment, the non-steady-state traffic price module combines the above monotonic constraints and the budget adjustment mechanism of the variable upper limit to output the advertising budget to the bidding control module, thereby ensuring the improvement of the operating stability of the advertising bidding system in advertising scenarios with drastic traffic fluctuations.

[0051] See also Figure 2 . In some embodiments, the advertising bidding system further includes a non-steady-state uncertainty estimation submodule; the non-steady-state uncertainty estimation submodule is used to obtain advertising sample data similar to the advertising feature data of the non-steady-state traffic price module, and generate a first uncertainty coefficient based on a specified coefficient of variation in combination with the advertising sample data; and, based on the result data in the previous time slice provided by the advertising platform, calculate the uncertainty of the distribution to obtain a second uncertainty coefficient; combine the first uncertainty coefficient and the second uncertainty coefficient to derive a target uncertainty coefficient; wherein the target uncertainty coefficient is used to affect the actual use of the advertising budget by the bidding control module.

[0052] In this embodiment, the advertising bidding system further includes a non-steady-state uncertainty estimation submodule, which is used to process the uncertainty factors of the non-steady-state traffic price module in the advertising budget prediction process, thereby improving the reliability of the budget prediction results and having an effective impact on the actual budget usage process of the bidding control module. Specifically, the non-steady-state uncertainty estimation submodule can first obtain historical advertising sample data similar to the current advertising feature data based on the advertising feature data of the non-steady-state traffic price module, and generate a first uncertainty coefficient in combination with the specified coefficient of variation. For example, at the beginning of time slice t, the non-steady-state uncertainty estimation submodule can generate a first uncertainty coefficient for the advertising feature data s t,i , get k advertising sample data The first uncertainty coefficient is calculated based on the following formula.

[0053]

[0054] in is the prediction result of the non-steady-state traffic price module, and σ(·) is the standard deviation function.

[0055] In addition, the non-steady-state uncertainty estimation submodule also calculates the uncertainty of the distribution based on the bid result data in the previous bidding cycle provided by the advertising platform to obtain the second uncertainty coefficient. The second uncertainty coefficient is calculated by analyzing the difference between the actual bid and the corresponding budget distribution, from the largest to the time slice t-1 in the advertising platform feedback. t―1,i The true result (i.e. the result Y of time slice t-1) t―1,i ), the uncertainty of the distribution can be calculated using the following formula 5.

[0056]

[0057] To better assess the distribution uncertainty of the most recent period, an exponential moving average (EMA) algorithm is used, as shown in Equation 6.

[0058]

[0059] The hyperparameter α=0.9.

[0060] In this embodiment, the non-steady-state uncertainty estimation submodule combines the first uncertainty coefficient and the second uncertainty coefficient to generate a target uncertainty coefficient. The calculation of the target uncertainty coefficient can be achieved by a weighted method. In a specific embodiment, the target uncertainty coefficient can be obtained based on the following formula 7.

[0061]

[0062] In this embodiment, the generated target uncertainty coefficient is used to influence the actual use of the advertising budget by the bid control module. For example, when the target uncertainty coefficient is large, the bid control module can reduce the dependence on the advertising budget and adopt a more conservative bidding strategy; and when the target uncertainty coefficient is small, the bid control module can increase the use of the advertising budget and optimize the advertising effect.

[0063] See also Figure 3 In some embodiments, the steady-state traffic price module can be constructed based on a multi-task learning model, which includes an expert network corresponding to each task and a shared network shared by multiple tasks; the multiple tasks for training the multi-task learning model include multiple of the following: conversion number-bid task, cost-bid task, exposure-bid task or click-bid task.

[0064] In this implementation, the steady-state traffic price module is built based on a multi-task learning model to fully explore the relationship between different advertising features and budgets, and improve the accuracy and robustness of advertising budget prediction. The multi-task learning model achieves adaptation to the needs of various advertising scenarios through the synergy of a shared network and multiple expert networks.

[0065] Specifically, the multi-task learning model of the steady-state traffic price module includes two core parts: a shared network and an expert network. The shared network is used to extract global information from advertising feature data, such as target audience characteristics, historical click-through rate, exposure, etc., to ensure information sharing between different tasks. The expert network performs deep modeling for specific tasks, such as conversion number-bid tasks, cost-bid tasks, exposure-bid tasks, and click-bid tasks. The expert network of each task independently processes the advertising data features related to it, thereby improving the task-specific performance of the model. In a specific manner, the steady-state traffic price module can be based on PLE (Progressive Layered Extraction) as the main network to construct an advertising budget-bid relationship model. During the training process, the four subtasks of "conversion number-bid", "cost-bid", "exposure-bid" and "click-bid" can be trained, and the loss function RMSE can be selected. In some embodiments, the monotonic layer mentioned in CMNN (Constrained Monotonic Neural Networks) can be used to improve the monotonicity of the model.

[0066] In this embodiment, the steady-state traffic price module combines the characteristics of the shared network and the expert network, and can handle multiple advertising budget-related tasks at the same time. For example, in the conversion number-bid task, the steady-state traffic price module inputs specific advertising feature data and predicts the impact of advertising budget on conversion number; in the cost-bid task, the steady-state traffic price module predicts the effect of budget changes on overall cost by analyzing historical cost data. In addition, through the global information extracted by the shared network, each expert network can avoid repeated modeling between tasks, thereby improving the overall modeling efficiency.

[0067] Specifically, the steady-state traffic price module generates the advertising budget forecast results for the next time slice through a multi-task learning model, and provides the optimized advertising budget to the bid control module. For example, for a brand promotion advertisement, the steady-state traffic price module can simultaneously predict the budget requirements corresponding to multiple indicators such as click volume, exposure volume, and conversion number, thereby providing advertisers with a comprehensive budget optimization solution.

[0068] In this embodiment, the steady-state traffic price module realizes accurate prediction of the advertising budget in the steady-state scenario and information sharing between tasks through a multi-task learning model, so that the advertising bidding system can effectively adapt to different advertising delivery scenarios, optimize advertising delivery effects and achieve cost control.

[0069] See also Figure 4In some embodiments, the steady-state traffic price module further includes a steady-state uncertainty estimation submodule; the steady-state uncertainty estimation submodule is used to provide an uncertainty coefficient corresponding to the advertising budget provided by the steady-state traffic price module; wherein the steady-state traffic price module performs a specified number of forward propagations of the steady-state traffic price module to obtain a specified number of advertising budgets provided by the steady-state traffic price module, and obtains the uncertainty coefficient based on the specified number of advertising budgets.

[0070] In this embodiment, the steady-state traffic price module further includes a steady-state uncertainty estimation submodule, which is used to evaluate the uncertainty of the advertising budget prediction in the steady-state scenario, thereby providing the bid control module with more valuable budget information and improving the stability of advertising delivery. The steady-state uncertainty estimation submodule obtains a set of advertising budget results by performing multiple forward propagation operations of the steady-state traffic price module, and calculates the uncertainty coefficient based on the result set.

[0071] Specifically, in this embodiment, the steady-state uncertainty estimation submodule performs N forward propagation operations on the steady-state traffic price module, and each propagation generates an advertising budget result set advertising sample data based on the same input advertising feature data These budget results are calculated using the following formula 8 to obtain the uncertainty coefficient

[0072]

[0073] At the beginning of time slice t, for each advertisement feature data s t,i ,in, is the output of the steady-state traffic price module, and σ(·) is the standard deviation function. Then, referring to the uncertainty calculation method of the non-steady-state scenario, we first calculate the distribution uncertainty. The total uncertainty can be obtained by weighting the two uncertainties.

[0074] The steady-state uncertainty estimation submodule can reflect the reliability of the budget forecast through the above uncertainty coefficient. When the uncertainty is high, it means that the credibility of the steady-state traffic price module's advertising budget forecast for the input advertising feature data is low; conversely, it indicates that the advertising budget forecast result is relatively stable and usable. In some embodiments, the steady-state uncertainty estimation submodule not only calculates the uncertainty coefficient, but also further generates a weighted uncertainty coefficient based on the distribution characteristics of the budget result.

[0075] In this embodiment, the uncertainty coefficient generated by the steady-state uncertainty estimation submodule will be used as feedback information to affect the advertising budget usage strategy of the bid control module. For example, when the uncertainty coefficient is high, the bid control module can adopt a more conservative advertising quotation strategy to reduce the fluctuation of the delivery effect caused by budget uncertainty; when the uncertainty coefficient is low, the bid control module can rely more on the advertising budget forecast results and implement a more active advertising quotation strategy.

[0076] Specifically, for example, in a steady-state advertising scenario of a certain brand, the steady-state uncertainty estimation submodule found that the fluctuation range of the budget distribution is small and the uncertainty coefficient is low through multiple propagation analysis of the budget results, indicating that the credibility of the budget forecast results is high. At this time, the bid control module can directly bid according to the advertising budget provided by the steady-state traffic price module; on the contrary, in another scenario, if the steady-state uncertainty coefficient is high, the advertising budget can be appropriately revised or reduced to ensure the stability of the advertising effect.

[0077] In this embodiment, the steady-state uncertainty estimation submodule provides reliable data support for the strategy adjustment of the bid control module by evaluating the stability of the budget prediction results in the steady-state scenario, thereby further optimizing the advertising delivery effect, improving the advertiser's budget utilization efficiency and the overall performance of the advertising bidding system.

[0078] See also Figure 5 In some embodiments, the steady-state traffic price module further includes a steady-state optimization submodule; the steady-state optimization submodule is used to obtain the advertising budget corresponding to the specified number of time slices based on the advertising budget of the steady-state traffic price module, with the goal of obtaining the overall maximum click volume for the specified number of time slices, and use the advertising budget of the first time slice in the time sequence as the optimized advertising budget of the steady-state traffic price module.

[0079] In this embodiment, the steady-state traffic price module further includes a steady-state optimization submodule, which is used to optimize the budget allocation scheme within a specified number of time slices based on the advertising budget of the steady-state traffic price module, so as to maximize the total number of clicks of the advertising bidding system within these time slices, thereby improving the efficiency of advertising delivery. The steady-state optimization submodule generates a final optimized advertising budget by jointly optimizing the budgets of multiple time slices, and uses it for the budget output of the steady-state traffic price module.

[0080] Specifically, the steady-state optimization submodule takes the overall maximization of the number of ad clicks as the optimization goal. Combined with the characteristics of the ad budget in the steady-state scenario, formulas 1 to 3 can be rewritten as the following formulas.

[0081]

[0082] Where B is the full-day budget, u q represents the advertising budget for the qth hour, Indicates the spending limit for the qth hour (used to deal with a traffic decline, when the plan still hopes to spend the full day's budget to avoid continuous price increases).

[0083] In the generated time slice budget, the steady-state optimization submodule outputs the advertising budget of the first time slice as the optimized advertising budget to the steady-state traffic price module for the next time slice delivery decision. In this way, the steady-state optimization submodule can not only balance the budget allocation of multiple time slices, but also effectively improve the delivery effect of the advertising system.

[0084] In some embodiments, the steady-state optimization submodule can further optimize the budget allocation strategy in combination with uncertainty information. For example, based on the uncertainty coefficient generated by the steady-state uncertainty estimation submodule, the budget allocation weight is adjusted so that a more conservative budget is allocated in time slices with higher uncertainty, and more budget is allocated in time slices with lower uncertainty.

[0085] For example, in a steady-state advertising scenario for a certain brand, the steady-state optimization submodule optimizes the allocation of the advertising budget based on the advertiser's budget target, combined with the historical click-through rate data and delivery effect statistics of multiple time slices, so as to maximize the number of clicks while limiting the total budget. After optimization, the steady-state optimization submodule feeds back the optimized budget value of the first time slice to the steady-state traffic price module to ensure that the advertising system can efficiently use budget resources.

[0086] In this embodiment, the steady-state optimization submodule can not only improve the overall effect of advertisement clicks by optimizing the advertisement budget of multiple time slices, but also enable the advertisement system to use the advertisement budget more efficiently in a steady-state scenario, thereby further enhancing the applicability of the advertisement bidding system.

[0087] See also Figure 6 In some embodiments, the non-steady-state traffic price module further includes a non-steady-state optimization submodule; the non-steady-state optimization submodule is used to adjust the advertising budget to minimize the difference between the actual click volume and the target click volume under the condition of meeting the specified optimization constraints.

[0088] In this embodiment, the non-steady-state traffic price module further includes a non-steady-state optimization submodule, which is used to optimize and adjust according to the advertising budget in the non-steady-state scenario to minimize the difference between the actual click volume and the target click volume under the condition of satisfying the specified optimization constraints, thereby improving the accuracy and adaptability of advertising delivery. The non-steady-state optimization submodule generates an advertising budget that meets the optimization target by optimizing the budget allocation and the click volume difference.

[0089] Specifically, the non-steady-state optimization submodule takes minimizing the difference between the actual click volume and the target click volume as the optimization goal, combines the fluctuation characteristics of advertising traffic and real-time requirements in non-steady-state scenarios, and uses the constraint optimization method to dynamically adjust the advertising budget. Specifically, formulas 1 to 3 can be rewritten as the following formulas.

[0090] min k (15)

[0091] sB q =B q-1 +u q (16)

[0092]

[0093] B q ≤B (19)

[0094]

[0095]

[0096] Among them, τ represents the target value of the benefit, k is the decision variable used to solve CPC, and represents the system vulnerability (or the distance from the optimal benefit). Represents the predicted value P means CPC = x t,i is a possible value for feedback, and Δ() is the L1 distance measurement. Given the uncertainty value Afterwards, P can be obtained by The hyperparameter τ is set to 0.95. In some implementations, if the current scenario is a steady-state scenario, the result of the steady-state traffic price module has good accuracy, which can be (Minimizing k means minimizing the upper limit of -click, which means maximizing the lower limit of click, because model uncertainty is not considered, that is, the click expectation is maximized).

[0097] In specific applications, for example, in the non-steady-state scenario of short video ads, the non-steady-state optimization submodule can dynamically adjust the advertising budget according to the real-time changes in the target click volume and traffic. If the actual click volume is significantly different from the target click volume in a certain time slice, the non-steady-state optimization submodule will prioritize allocating more resources by adjusting the advertising budget to narrow the click volume difference; in other time slices, if the click volume difference is small, the budget allocation can be appropriately reduced to optimize the overall delivery effect.

[0098] In this embodiment, the non-steady-state optimization submodule dynamically optimizes the advertising budget to ensure high consistency between actual clicks and target clicks in non-steady-state scenarios, thereby improving the advertising delivery effect and enhancing the adaptability of the advertising bidding system to complex scenarios.

[0099] See also Figure 7 In some implementations, the bid control module performs smoothing operations based on historical ROI data and PV data to obtain target ROI data and target PV data, and sets an advertising bid based on the target ROI data, the target PV data, and the specified target data.

[0100] In this embodiment, the bid control module generates target ROI data and target PV data through smoothing calculation based on historical ROI data and PV data, and further sets the advertising quotation in combination with the specified target data, thereby achieving optimization and adjustment of advertising delivery. Through this process, the bid control module can balance the return on investment and exposure effect in real-time changing advertising scenarios, ensuring that the advertiser's delivery goals are achieved.

[0101] Specifically, the bid control module performs smoothing operations on historical ROI data and PV data, and can follow the following formula.

[0102]

[0103] For example, using exponential smoothing with p=0.1, the upper limit of the roi data is 20.

[0104] In some embodiments, the bid control module can also combine the dynamic feedback mechanism to adjust the target ROI data and target PV data in real time according to the actual delivery effect. For example, when the actual ROI deviates from the target ROI, the bid control module can adjust the quotation strategy to gradually converge to the target ROI; when the actual PV deviates from the target PV, the bid control module can increase the exposure-oriented advertising delivery budget to ensure that the advertisement achieves the expected exposure effect.

[0105] Through this implementation, the bid control module can realize dynamic optimization and balanced adjustment of advertising quotations, which not only improves the flexibility of advertising delivery, but also further improves the advertiser's budget utilization efficiency and delivery effect, allowing the advertising bidding system to better adapt to complex and changeable advertising traffic scenarios.

[0106] In some embodiments, when the bid control module performs smoothing operations on historical ROI data and PV data to obtain target ROI data and target PV data, the historical ROI data and PV data closer to the current time have greater weights.

[0107] In this embodiment, the bid control module applies time decay weights to historical data during the process of performing weighted smoothing operations based on historical ROI data and PV data to generate target ROI data and target PV data, so that historical data closer to the current time has a greater weight. This method can more accurately reflect the impact of recent advertising effects on current bidding decisions and improve the timeliness of advertising quotations.

[0108] Specifically, the bid control module extracts ROI data and PV data from historical advertising data, which represent the return on investment and exposure of the advertisement respectively. When calculating the target ROI data and target PV data, the time decay weight w is introduced to assign different weights to data in different time slices. The closer the time is to the current time, the greater the weight. Through this weight distribution method, the newer historical data has a greater impact on the calculation of the target data, while the impact of the older data gradually decays. By weighted smoothing the historical data through the time decay weight, the target ROI and target PV can better reflect the dynamic characteristics of advertising at the current time point.

[0109] After the target data is generated, the bid control module combines the specified target data (such as the advertiser's budget cap, expected conversion rate, exposure target, etc.) to generate an advertising quote through an optimization model.

[0110] In some embodiments, the bid control module combines a dynamic feedback mechanism to make real-time adjustments to the calculation of the target data according to the actual delivery effect. For example, when the real-time ROI and PV deviate from the target ROI and target PV, the bid control module can adjust the weight parameter or time decay coefficient to reallocate the weight of the historical data to ensure that the target data can better reflect the actual delivery situation.

[0111] In some embodiments, the bid regulation module generates a ROI control error and a PV control error corresponding to the target ROI data and the target PV data, respectively, and inputs the ROI control error into the ROI controller, and inputs the PV control error into the PV controller, so that the bid regulation module can output an advertising quotation that reduces the ROI control error and the PV control error.

[0112] In this embodiment, the bid control module generates ROI control error and PV control error respectively by combining the target ROI data and the target PV data, and inputs these two types of control errors into the ROI controller and the PV controller respectively. In this way, the bid control module can dynamically adjust the advertising price, thereby gradually reducing the ROI control error and the PV control error, and realizing the accurate optimization of the advertising price.

[0113] Specifically, the bid control module calculates the corresponding ROI control error and PV control error based on the target ROI data and target PV data, combined with the real-time ROI data and PV data obtained during the actual delivery process. The ROI control error is used to reflect the degree of deviation between the target ROI and the actual ROI, and the PV control error reflects the degree of deviation between the target PV and the actual PV. This process ensures that the advertising bid adjustment can accurately respond to the difference between the target data and the actual delivery effect, and optimize the delivery effect.

[0114] After generating the control error, the bid control module inputs the ROI control error and the PV control error to the ROI controller and the PV controller respectively. The ROI controller and the PV controller can use PID control algorithms or other dynamic control algorithms to achieve precise control of the advertising quotation. For example, the ROI controller calculates and outputs the adjusted ROI control amount based on the ROI control error; similarly, the PV controller calculates and outputs the adjusted PV control amount based on the PV control error. The bid control module combines the control amounts of the two to generate the final advertising quotation, thereby gradually reducing the control error.

[0115] In some embodiments, the ROI controller and the PV controller can be further dynamically adjusted in combination with historical delivery data and specified target data. For example, when the ROI control error deviates greatly within a certain period of time, the ROI controller can increase the control intensity and give priority to adjusting the advertising price to reduce the ROI control error; and when the PV control error deviates greatly, the PV controller gives priority to adjusting the advertising price to optimize the advertising exposure effect.

[0116] In addition, the bid control module can optimize controller parameters based on actual delivery feedback information while dynamically adjusting the advertising price. For example, in scenarios where the advertising click-through rate fluctuates greatly, the bid control module can appropriately increase the controller's response sensitivity to speed up the error convergence; in scenarios where advertising delivery is relatively stable, the controller's sensitivity can be reduced to smooth the adjustment process of the advertising price.

[0117] In some implementations, the bid control module generates a confidence level corresponding to an ROI control error using historical click volume; the bid control module corrects the ROI control error using the confidence level and inputs the result to an ROI controller.

[0118] In this embodiment, the bid control module generates the confidence level corresponding to the ROI control error by using the historical click volume, and after correcting the confidence level, inputs the corrected ROI control error into the ROI controller to optimize the advertising bidding strategy and reduce the delivery deviation. This process can improve the accuracy and robustness of the advertising bid, thereby more effectively achieving the delivery goals of advertisers.

[0119] Specifically, the bid control module can first extract click information from the historical data of advertising, and generate ROI control error by combining the target ROI data and the real-time ROI data. ROI control error is used to measure the degree of deviation between the target ROI and the actual ROI, and its numerical value can reflect the degree of consistency between the advertising effect and the expected target. However, since the historical click data may have a certain degree of uncertainty, the bid control module generates a corresponding confidence level for the ROI control error to quantify its reliability.

[0120] In this embodiment, the confidence level is calculated based on the distribution characteristics of the historical click volume data. The more stable the historical click volume data is and the smaller the deviation is, the higher the confidence level is; conversely, if the historical click volume fluctuates greatly or the data distribution is discrete, the confidence level is low. The bid control module further uses the confidence level to correct the ROI control error, so that the corrected ROI control error can more accurately reflect the actual deviation of the delivery effect.

[0121] The corrected ROI control error is then input into the ROI controller. The ROI controller can use a proportional integral derivative (PID) control algorithm or other dynamic control algorithms to adjust the advertising price in real time according to the corrected ROI control error, gradually reducing the deviation between the target ROI and the actual ROI. Specifically, the ROI controller dynamically adjusts the advertising price so that the ROI control error gradually approaches zero in the closed-loop feedback system, thereby improving the advertising effect and being consistent with the advertiser's expected goals.

[0122] In some implementations, the bid control module dynamically adjusts the calculation method of the confidence level in combination with the historical click volume and the real-time delivery data. For example, when the distribution of the historical click volume data in multiple time slices is relatively stable, the bid control module can assign a higher weight to the confidence level to enhance the trust in the correction process; when the historical click volume fluctuates greatly or the delivery environment changes dramatically, the bid control module can reduce the impact of the confidence level on the correction to ensure the flexibility and adaptability of the delivery strategy.

[0123] In some embodiments, for example, the bid regulation module can construct a confidence curve such as Figure 8 In this embodiment, the confidence is related to the historical click volume. It is assumed that the confidence is 0.5 when the historical click volume is 30. For example, it is assumed that the average conversion rate is 0.03, that is, every 30 clicks correspond to 1 transaction, and then the confidence is 0.1 when there are 3 clicks. The formula for calculating the confidence is as follows.

[0124]

[0125] Among them, p conf represents the confidence level, and x can be the historical click volume.

[0126] Furthermore, the confidence can be applied to the roi control error, and the formula is as follows.

[0127] err roi =(roi t -roi t-1 )*(p conf,t +p conf,t-1 ) / 2 (25)

[0128] Among them, err roi is the ROI control error, and t is the time slice. In some embodiments, the ROI data of the ROI controller can be set to a minimum value of 0.2. For the PV controller, when the total number of historical clicks does not exceed 50, or the average number of clicks is <= 9, the cardinality is too small, and it is difficult to obtain a good result based on the PV control error feedback price adjustment. Therefore, a lower limit can be set when calculating the error feedback cardinality.

[0129]

[0130] Among them, err pv is the pv control error. d can be the time slice.

[0131] In some implementations, a cascaded PID controller may be used to execute the price adjustment strategy. Specifically, the cascaded PID controller may adopt a control strategy such as the following formula.

[0132]

[0133] According to the previous statistical relationship between traffic change and bid change, we can get Therefore, the original pid formula is adjusted to:

[0134]

[0135] Among them, bid d Bid for the current time slot. d+1 Bid for the next time slot ad, bid d-1 Bid for the advertisement in the previous time slice of the current time slice, pid pv PV control parameters, pid roi is the roi control parameter, sig(bid d >bid d-1 ) is a logical symbol function used to determine whether the bid is increasing. d >bid d-1 ,sig(bid d >bid d-1) is 1, if sig(bid d ≤bid d-1 ) is 0. 1 / p and 1 / q are weight parameters used to adjust the sensitivity of PID. P，pv Is the proportional factor used to control the pv control error. pv is the pv control error. I，pv is an integration factor used to accumulate historical errors to correct long-term deviations. is the cumulative value of the pv control error. d is the exposure of the dth time slice, pv d--1 is the exposure of the d-1th time slice. avg(err) is the average value of the historical PV control error. By regulating the PV and ROI indicators in stages and levels, the stability of advertising bids can be improved.

[0136] In some embodiments, for the first advertising traffic scenario, it includes a data accumulation stage; in the data accumulation stage, the steady-state traffic price module adjusts the advertising quotation according to the feedback of the main indicator and the reference indicator; wherein the main indicator includes ROI data and PV data, and the reference indicator includes at least one of the following: collection volume, purchase volume, click volume or completion volume.

[0137] In this embodiment, for the first advertising traffic scenario, the advertising bidding system further includes a data accumulation stage. In the data accumulation stage, the steady-state traffic price module adjusts the advertising quotation according to the feedback of the main index and the reference index, thereby realizing dynamic optimization of advertising budget allocation and ensuring that the effect of advertising delivery can meet the expected goals of the advertiser.

[0138] Specifically, during the data accumulation stage, the steady-state traffic price module receives and processes the data of the main indicators and reference indicators. The main indicators include ROI data (return on investment) and PV data (clicks), which are used to measure the economic benefits and coverage of advertising. ROI data reflects the conversion revenue brought by unit advertising expenditure, and PV data indicates the total number of times an advertisement is displayed on the delivery platform, which is an important indicator of the advertising coverage effect. Reference indicators include at least one of the collection volume, add-to-cart volume, click volume, or completion volume, which are used to further refine the evaluation of advertising effectiveness. The collection volume is the behavioral data that users show interest in the advertised product, the add-to-cart volume refers to the number of times users add the advertised product to the shopping cart, the click volume is used to measure the degree of interaction between users and advertisements, and the completion volume indicates the number of times the video advertisement is watched in full.

[0139] In this embodiment, the steady-state traffic price module combines the main indicator with the reference indicator, and can use weighted analysis, dynamic feedback and other technologies to calculate the adjustment value of the advertising quotation. Specifically, the steady-state traffic price module will obtain the main indicator data and reference indicator data generated during the advertising delivery process in real time, and calculate the priority of the advertising quotation adjustment based on these data. For example, when the ROI data is lower than the advertiser's expected target, the steady-state traffic price module can give priority to increasing the budget weight related to the ROI indicator; when the PV data or reference indicators indicate that the advertising coverage is insufficient, it will give priority to increasing the budget allocation related to the coverage.

[0140] In this embodiment, the steady-state traffic price module generates the advertising quotation for the next time slice by comprehensively processing the feedback data of the main index and the reference index. The specific processing method includes the following steps: First, the steady-state traffic price module extracts ROI data and PV data from the advertising platform or historical data, as well as reference index data of the number of favorites, the number of purchases, the number of clicks or the number of completed broadcasts; then, the time decay model is used to perform weighted smoothing on these data to ensure that the index data closer to the current time has a higher weight in the advertising quotation calculation; finally, combined with the advertiser's budget target and delivery requirements, the steady-state traffic price module generates an optimized advertising quotation.

[0141] In some embodiments, the advertising bidding system can calculate the ROI data based on the following formula. In this way, the ROI data can be used to analyze user behavior more comprehensively.

[0142]

[0143] Among them, gmv today is the total transaction amount today, indicating the sales amount directly brought by the advertisement. α*(coll+cart) is the sum of the collection amount and the purchase amount, which is used to indicate the user's interest in the product. β*click is the weighted value of the click amount. γ*vpo is the weighted value of the completion amount. cost is the total cost of the advertisement. θ*(pv+1) is the weighted value of the exposure amount. Among them, θ is the adjustment weight, and pv+1 avoids the denominator being zero. The hyperparameters α, β, γ, and θ are searched through the CMA-ES algorithm, and the initial value is obtained by calculating the market proportion of the advertising platform.

[0144] In some embodiments, for the first advertising traffic scenario, it also includes a personalized regulation stage; in the personalized regulation stage, the steady-state traffic price module constructs the ROI upper limit value, ROI lower limit value, PV upper limit value and PV lower limit value according to the data accumulated in the data accumulation stage; the steady-state traffic price module controls the advertising quotation based on the calculated current ROI data and current PV data, and the comparison result with the ROI upper limit value, the ROI lower limit value, the PV upper limit value and the PV lower limit value.

[0145] In this embodiment, for the first advertising traffic scenario, the advertising bidding system further includes a personalized regulation stage. In this personalized regulation stage, the steady-state traffic price module constructs the ROI upper limit value, ROI lower limit value, PV upper limit value and PV lower limit value based on the data accumulated in the data accumulation stage, and dynamically controls the advertising quotation by comparing the calculated current ROI data and current PV data with these upper and lower limits to optimize the flexibility and effect of advertising delivery.

[0146] Specifically, in the personalized regulation stage, the steady-state traffic price module first calculates the current ROI data and the current PV data based on the historical data and real-time feedback data in the data accumulation stage. The current ROI data is used to measure the return on investment of the advertisement, and the current PV data is used to indicate the exposure effect of the advertisement. On this basis, the steady-state traffic price module generates the ROI upper limit and ROI lower limit, as well as the PV upper limit and PV lower limit through regression analysis, historical distribution fitting and other methods. The calculation of the above upper and lower limits can be based on the following formula:

[0147]

[0148] in, is the upper limit value of pv, Set limits for pv, is the upper limit of roi, is the lower limit of roi. avg() is the average value. sigma() is the standard deviation. pv base is the accumulated pv data set, roi base is the accumulated roi data set.

[0149] In this embodiment, the steady-state traffic price module compares the calculated current ROI data with the ROI upper limit and ROI lower limit, and compares the current PV data with the PV upper limit and lower limit. The specific control rules are as follows: When the current ROI data is higher than the ROI upper limit, the steady-state traffic price module lowers the advertising quotation to avoid wasting advertising resources due to excessively high ROI. When the current ROI data is lower than the ROI lower limit, the steady-state traffic price module increases the advertising quotation to enhance the competitiveness of advertising. When the current PV data is higher than the PV upper limit, the steady-state traffic price module reduces the exposure-oriented budget allocation to optimize the delivery cost. When the current PV data is lower than the PV lower limit, the steady-state traffic price module increases the exposure-oriented budget allocation to increase the coverage of the advertisement. The steady-state traffic price module dynamically adjusts the advertising quotation according to the formula by combining the above-mentioned control rules:

[0150] In some embodiments, the advertising bidding system clusters the advertising plans into multiple advertising clusters based on feature data of the advertising plans. Advertisements in the same advertising cluster use the same hyperparameter control controller, and at least some hyperparameters are different between different advertising clusters.

[0151] In this embodiment, the advertising bidding system clusters the advertising plans into multiple advertising clusters based on the characteristic data of the advertising plans, thereby realizing efficient management and dynamic optimization control of advertising resources. In this process, advertisements belonging to the same advertising cluster share the same hyperparameter control controller, while at least some hyperparameters are different between different advertising clusters to adapt to the specific delivery requirements of each advertising cluster and improve the accuracy and flexibility of advertising delivery.

[0152] Specifically, the advertising bidding system first performs clustering operations based on the characteristic data of the advertising plan. The characteristic data of the advertising plan may include but is not limited to the following information: target audience characteristics, delivery area, delivery time period, advertising type, budget cap, click-through rate, conversion rate, etc. The system analyzes and models these characteristic data, combined with clustering algorithms (such as density-based clustering, K-means clustering or spectral clustering, etc.), to divide advertising plans with similar characteristics into the same advertising cluster. The advertisements in each advertising cluster have high characteristic similarity, so they can share a unified hyperparameter control controller.

[0153] In this embodiment, the hyperparameter control controller is used to manage the delivery behavior of advertisements within the advertising cluster, and its functions include but are not limited to: controlling budget allocation, adjusting bidding strategies, optimizing delivery effects, etc. Due to the high similarity of features, advertisements in the same advertising cluster can use the same hyperparameter control controller, such as a shared delivery budget cap, target ROI, and target PV value, to simplify the control logic and improve computing efficiency. For different advertising clusters, due to the large differences in their features, the advertising bidding system configures at least some different hyperparameters for them to adapt to the specific delivery needs of each advertising cluster. For example, for short video advertising clusters, the system may configure a higher budget fluctuation range to cope with non-steady-state traffic scenarios, while for long-term brand promotion advertising clusters, a stricter budget control strategy will be configured to achieve a stable delivery effect.

[0154] See also Fig. 9 The present embodiment may provide a computer device, the computer device comprising: a memory, and one or more processors in communication with the memory; the memory stores instructions executable by the one or more processors, the instructions are executed by the one or more processors, so that the one or more processors implement the advertising bidding system as described above.

[0155] In some embodiments, the computer device may include a processor, a non-volatile storage medium, an internal memory, a communication interface, a display device, and an input device connected by a system bus. The non-volatile storage medium may store an operating system and related computer programs.

[0156] The user information or user account information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, etc.) involved in multiple implementation methods of the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0157] It should be understood that the specific examples in this article are only intended to help those skilled in the art to better understand the embodiments of the present application, rather than to limit the scope of the present invention.

[0158] It can be understood that in the various implementations of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the implementation methods of the present application.

[0159] It can be understood that the various embodiments described in this application can be implemented individually or in combination, and the embodiments of this application are not limited to this.

[0160] Unless otherwise stated, all technical and scientific terms used in the embodiments of the present application have the same meaning as those generally understood by those skilled in the art of the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more related listed items. The singular forms of "a kind of", "above" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0161] It can be understood that the processor of the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method implementation can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (DigitalSignal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiment of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0162] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (programmable ROM, PROM), an erasable programmable read-only memory (erasable PROM, EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory may be a random access memory (RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0163] Those of ordinary skill in the art will appreciate that the units 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.

[0164] 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 units described above can refer to the corresponding processes in the aforementioned method implementation methods and will not be repeated here.

[0165] 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 implementation described above is only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units 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 units, which can be electrical, mechanical or other forms.

[0166] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0167] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0168] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the 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, and the computer software product is stored in a storage medium, including a number of 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 method 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.

[0169] The above is only a specific implementation of the present application, but the protection scope of the present invention 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 invention should be based on the protection scope of the claims.

Claims

1. An advertising bidding system, characterized in that: The advertising bidding system comprises: The steady-state traffic price module is used to simulate the correlation between advertising traffic, advertising bid and advertising expenditure in the first advertising traffic scenario, and to provide the advertising budget for the next time slice according to the input advertising feature data; A non-steady-state traffic price module, used to simulate the correlation between advertising traffic, advertising bid and advertising expenditure in a second advertising traffic scenario, and to provide an advertising budget for the next time slice according to the input advertising feature data; wherein the stability of advertising traffic in the second advertising traffic scenario is weaker than the stability of advertising traffic in the first advertising traffic scenario; The bid regulation module is used to send an advertisement quotation to an advertisement platform according to the advertisement budget and designated target data provided by the steady-state traffic price module or the non-steady-state traffic price module.

2. The system according to claim 1, characterized in that The advertising budget provided by the non-steady-state traffic price module to the bid regulation module complies with a monotonic constraint and has a variable upper limit.

3. The system according to claim 1, characterized in that The non-steady-state traffic price module also includes a non-steady-state uncertainty estimation submodule; the non-steady-state uncertainty estimation submodule is used to obtain advertising sample data similar to the advertising feature data of the non-steady-state traffic price module, and generate a first uncertainty coefficient based on a specified coefficient of variation combined with the advertising sample data; and, based on the result data of the bid in the previous bidding cycle provided by the advertising platform, calculate the uncertainty of the distribution to obtain a second uncertainty coefficient; combine the first uncertainty coefficient and the second uncertainty coefficient to derive a target uncertainty coefficient; wherein, the target uncertainty coefficient is used to influence the actual use of the advertising budget by the bid control module.

4. The system according to claim 1, characterized in that The steady-state traffic price module is constructed based on a multi-task learning model, which includes an expert network corresponding to each task and a shared network shared by multiple tasks; the multiple tasks for training the multi-task learning model include multiple of the following: conversion number-bid task, cost-bid task, exposure-bid task or click-bid task.

5. The system according to claim 1, characterized in that The steady-state traffic price module also includes a steady-state uncertainty estimation submodule; the steady-state uncertainty estimation submodule is used to give an uncertainty coefficient corresponding to the advertising budget provided by the steady-state traffic price module; wherein the steady-state traffic price module executes a specified number of forward propagations of the steady-state traffic price module to obtain a specified number of advertising budgets provided by the steady-state traffic price module, and derives the uncertainty coefficient based on the specified number of advertising budgets.

6. The system according to claim 1, characterized in that The steady-state traffic price module also includes a steady-state optimization submodule; the steady-state optimization submodule is used to obtain the advertising budget corresponding to the specified number of time slices based on the advertising budget of the steady-state traffic price module, with the goal of obtaining the overall maximum click volume for the specified number of time slices, and use the advertising budget of the first time slice in time sequence as the optimized advertising budget of the steady-state traffic price module.

7. The system according to claim 1, characterized in that The non-steady-state traffic price module also includes a non-steady-state optimization submodule; the non-steady-state optimization submodule is used to adjust the advertising budget to minimize the difference between the actual click volume and the target click volume under the condition of meeting the specified optimization constraints.

8. The system according to claim 1, characterized in that The bid control module performs smoothing calculation based on historical roi data and PV data to obtain target roi data and target PV data, and sets an advertising bid based on the target roi data, the target PV data and the specified target data.

9. The system according to claim 8, characterized in that In the process of performing weighted smoothing operation on historical ROI data and PV data to obtain target ROI data and target PV data, the historical ROI data and PV data closer to the current time have greater weights.

10. The system according to claim 8, characterized in that The bid regulation module generates a ROI control error and a PV control error corresponding to the target ROI data and the target PV data respectively, and inputs the ROI control error into the ROI controller, and inputs the PV control error into the PV controller, so that the bid regulation module can output an advertising quotation that reduces the ROI control error and the PV control error.

11. The system according to claim 10, characterized in that The bid regulation module generates a confidence level corresponding to the ROI control error using the historical click volume; the bid regulation module corrects the ROI control error using the confidence level and then inputs the result to the ROI controller.

12. The system according to claim 1, characterized in that For the first advertising traffic scenario, it includes a data accumulation stage; in the data accumulation stage, the steady-state traffic price module adjusts the advertising quotation according to the feedback of the main indicators and the reference indicators; wherein the main indicators include ROI data and PV data, and the reference indicators include at least one of the following: collection volume, add-to-cart volume, click volume or completion volume.

13. The system according to claim 12, characterized in that For the first advertising traffic scenario, it also includes a personalized regulation stage; in the personalized regulation stage, the steady-state traffic price module constructs the ROI upper limit value, ROI lower limit value, PV upper limit value and PV lower limit value according to the data accumulated in the data accumulation stage; the steady-state traffic price module controls the advertising quotation based on the calculated current ROI data and current PV data, and the comparison result with the ROI upper limit value, the ROI lower limit value, the PV upper limit value and the PV lower limit value.

14. The system according to claim 1, characterized in that In the advertising bidding system, the advertising plans are clustered into multiple advertising clusters based on the characteristic data of the advertising plans. Advertisements in the same advertising cluster are controlled by the same hyperparameter controller, and at least some hyperparameters are different between different advertising clusters.

15. A computer device, characterized in that: The computer device includes a memory and a processor, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the system according to any one of claims 1 to 14.

Images