Advertisement automatic bidding method and system based on predictive delivery planning

Abstract

The invention relates to the technical field of advertisement putting, in particular to an automatic advertisement bidding method and system based on predictive putting planning, and the method comprises the steps: constructing a strategic putting picture composed of multi-dimensional advertisement activity measurement vectors based on historical advertisement putting data, and extracting a high-quality strategic putting picture; the advertising demand of the advertiser is coded into a standardized commercial intention code; a forward disturbance and conditional reverse recovery process training conditional picture synthesizer is constructed by using a high-quality strategic release picture and commercial intention coding, and in an actual bidding stage, starting from an initial strategic release picture, an optimal strategic release picture meeting a release requirement is gradually recovered and generated through the conditional picture synthesizer; and constructing and training a non-backtracking bid command derivation device based on the optimal strategic release picture, and generating a bid command in real time. According to the method, the multi-target and multi-constraint long-term delivery requirement proposed by an advertiser can be met, so that a stable, flexible and optimal-effect advertisement bidding strategy is difficult to realize on the whole.

Description

Technical Field

[0001] The present application relates to the technical field of advertising delivery, and in particular to an automatic advertising bidding method and system based on predictive delivery planning. Background Art

[0002] With the rapid development of digital marketing, online advertising has become a crucial means for businesses to acquire user traffic and improve conversion efficiency. Online advertising platforms typically provide advertisers with bidding services for ad placement. Advertisers can set budgets and strategies based on their marketing goals, and the platform will make specific bidding and display decisions throughout the ad delivery cycle. To improve ad delivery efficiency and reduce manual intervention costs, platforms have gradually introduced automated bidding technology. Using algorithmic models, they calculate and execute optimal bid instructions in real time, and this has become a fundamental capability module in intelligent delivery systems.

[0003] Currently, online advertising platforms typically use real-time responsive models to intelligently optimize advertisers' bidding behavior. Specifically, during the advertising campaign cycle, the model obtains the current status parameters of the advertising account at regular intervals, such as the remaining budget, amount spent, real-time return on investment (ROI), and average cost per click (CPC). Based on this current state information, the model calculates the corresponding bid adjustment instructions and issues them for execution in real time. This process is repeated continuously throughout the campaign cycle, forming a bidding mechanism that continuously makes decisions based on the current state. The core feature of this type of model is that it responds immediately to the state at each moment and makes locally optimal bidding decisions based on this information.

[0004] However, the decision-making process of an immediate-response model primarily revolves around the current state, using immediate-response logic to incrementally control bidding. This limitation becomes apparent as the ad delivery cycle lengthens and the number of decision points increases. First, ad delivery cycles often last 24 hours or longer, encompassing hundreds of decision points. The optimal decision at the moment often requires a foresight into the future. For example, a conservative bidding strategy in the morning might reserve sufficient budget for the more competitive afternoon hours. Immediate-response models lack the ability to comprehensively plan the entire delivery cycle and are unable to capture these global, non-local dependencies, resulting in suboptimal delivery results. Second, the online advertising environment is characterized by significant temporal fluctuations in factors such as the supply and demand structure of ad impressions, user behavior, and bidding competitor strategies. In this context, even a small deviation in the current decision can be amplified in subsequent decision chains, leading to significant fluctuations in actual delivery results. This results in unstable performance of strategies based on local optimization, reducing the robustness of the delivery system. Finally, advertisers often set long-term campaigns with multiple objectives and constraints, such as maximizing gross merchandise volume (GMV) within a limited budget and CPC, ensuring even budget consumption, or improving the smoothness of the campaign curve. Because instant response models cannot plan the path to goal achievement from a global perspective, they require extensive engineering configuration to forcibly break down complex objectives into the bidding logic at each moment. This increases system complexity and reduces adaptability to advertisers' personalized needs.

[0005] In summary, existing instant-responsive models make local decisions based only on the current state and lack the ability to plan global bidding paths for the entire delivery cycle. Furthermore, they are unable to withstand the strategic instability caused by random fluctuations in the advertising environment, nor can they meet the multi-objective, multi-constrained long-term delivery needs proposed by advertisers. Consequently, it is difficult to achieve a stable, flexible, and optimally effective advertising bidding strategy overall. Summary of the Invention

[0006] This application provides an automatic advertising bidding method and system based on predictive delivery planning, which can meet the multi-objective, multi-constrained long-term delivery needs of advertisers, thereby making it difficult to achieve a stable, flexible and optimal advertising bidding strategy overall. This application provides the following technical solutions: In a first aspect, the present application provides an automatic advertising bidding method based on predictive delivery planning, the method comprising: Build a strategic delivery landscape consisting of multi-dimensional advertising campaign measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery landscapes; Encode advertisers' placement requirements into standardized commercial intent codes; The extracted high-quality strategic delivery images and the advertiser's commercial intent encoding are used to construct a forward perturbation and conditional reverse recovery process to train a preset conditional image synthesizer. During the actual bidding phase, starting from a random initial strategic delivery image, the conditional image synthesizer gradually recovers and generates an optimal strategic delivery image that meets delivery requirements; A non-retroactive bidding command derivation device is constructed and trained based on the optimal strategic delivery landscape. The non-retroactive bidding command derivation device generates bidding commands in real time by combining historical trends with target states.

[0007] In a specific implementation scheme, constructing a strategic delivery picture composed of multi-dimensional advertising campaign metric vectors based on historical advertising delivery data includes: Collect historical advertising data, and discretize each historical advertising data into a complete advertising cycle as the time unit on the time axis. continuous observation nodes, at each observation node , with a multi-dimensional advertising campaign metric vector To represent the running status of the advertisement at that moment; By collecting The advertising activity measurement vectors on the observation nodes are used to construct a data matrix to characterize the dynamic evolution process of the entire cycle, which is recorded as the strategic delivery landscape. : ; in, It represents the strategic deployment picture of a complete advertising cycle. Indicates in The advertising activity measurement vector recorded on each observation node; It represents the total number of observation nodes obtained after discretizing the advertising delivery cycle. Indicates the number of feature dimensions contained in each campaign measurement vector, Indicates a OK A real matrix with columns.

[0008] In a specific implementation scheme, encoding the advertiser's delivery requirements into a standardized commercial intent code includes: Advertisers’ delivery requirements are uniformly coded into standardized commercial intent codes, recorded as ,in Indicates the number of encoding dimensions, represents the set of real numbers; Commercial Intent Encoding It includes core optimization intent, boundary constraints and process form preferences; the core optimization intent is used to express the advertiser's main goal, the boundary constraints are used to describe the constraints that must be met, and the process form preferences are used to reflect the advertiser's preference for the delivery process form.

[0009] In a specific embodiment, the method of using the extracted high-quality strategic placement image and the advertiser's commercial intent encoding to construct a forward perturbation and conditional reverse recovery process to train a preset conditional image synthesizer includes: During the forward perturbation process, the extracted high-quality strategy is presented with a picture Gaussian perturbations of preset strength are gradually injected into the image to obtain different perturbation levels. The perturbation is performed based on the set maximum perturbation step number. , at each step The disturbance picture at the current moment The perturbation picture at the previous moment Generate, through the closed function can calculate the disturbance picture at any time in one step, its mathematical form is: ; in, is a standard Gaussian random matrix, It is preset, Changing the signal-to-noise ratio control parameter, is the perturbation function; Building a deep neural network during conditioned reverse recovery , neural network The input includes the picture under the current disturbance level , the current perturbation time step and business intent encoding , whose task is to predict the original perturbation injected in this step given the triple input , neural network The optimization goal of training is to minimize the following expected error : ; in, represents the expected value of all samples in the training data, is the real Gaussian noise injected during the forward perturbation process, represents the model’s predicted value for the noise, Represents the Frobenius norm of the matrix.

[0010] In a specific implementation scheme, starting from a random initial strategic delivery landscape in the actual bidding phase, gradually recovering and generating an optimal strategic delivery landscape that meets delivery requirements through the conditional landscape synthesizer includes: In the actual bidding stage, from a completely random initial strategic delivery landscape Starting from the commercial intent encoding of a given advertiser Under the guidance of ,by As input, execute Wheel reverse recovery operation, that is, at each time step , deep neural networks Receive the current disturbance landscape , the current perturbation time step and business intent encoding , by reverse reasoning, estimate the original perturbation in the current image , and calculate the previous step picture based on the estimated results The process is iterated continuously, making the image gradually clearer and more stable from the initial Gaussian noise. Generate the optimal strategic deployment picture .

[0011] In a specific possible implementation scheme, the non-retroactive bidding command derivation device is constructed and trained based on the optimal strategic delivery landscape, wherein the non-retroactive bidding command derivation device generates bidding commands in real time by combining historical trends and target states, including: Construct a non-retrospective bidding command deriver, whose task is to combine the current and recent historical observation states with the next moment target state during the system operation, and derive the optimal bidding command to drive the system towards the target state in real time. ; The design goal of this derivation is to answer the following questions: In order to make the next state of the system as close as possible to the planned target state , what bidding command should be executed ; Model the derivation as a function , its mathematical form is as follows: ; in, For the current moment The bid command that should be executed, Indicates that the current time is the end point and the length is The historical measurement observation window of Indicates that the system is at time The operational measurement status of Indicates the current image The target state of each observation node; The non-retrospective bid command deriver is called in real time on each observation node to collect the actual running metrics of the past H time steps , and combined with the current picture The target state of the observation node , as input and input into the derivation, the derivation performs feature extraction and deep modeling on the input, and outputs the bidding command that should be executed at the current moment .

[0012] In a specific implementation scheme, a standard supervised learning method is used to train the non-retrospective bidding command derivation engine. The specific process is as follows: a large-scale training sample pair is constructed from historical advertising data. , where the input of each sample is It is a combination of the historical measurement window and the next measurement vector actually achieved, that is, the historical observation sequence at a certain moment and the next target state preset in the picture, label The bid command actually executed in this historical window ; The training goal of the non-retrospective bidding command deriver is to minimize the distance between the predicted command and the true command, using the L2 loss function as follows: ; in, represents the expectation on all training sample pairs, Indicates the bid commands actually executed in history. Represents the model based on the input The predicted bid command.

[0013] In a second aspect, the present application provides an automatic advertising bidding system based on predictive delivery planning, which adopts the following technical solutions: An automatic advertising bidding system based on predictive delivery planning, comprising: A high-quality picture extraction module is used to construct a strategic delivery picture composed of multi-dimensional advertising activity measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery pictures; The delivery demand encoding module is used to encode the advertiser's delivery demand into a standardized commercial intent code; An optimal scenario generation module is used to use the extracted high-quality strategic delivery scenarios and the advertiser's commercial intent encoding to construct a preset conditional scenario synthesizer for training through a forward perturbation and conditional reverse recovery process. During the actual bidding phase, starting from a random initial strategic delivery scenario, the conditional scenario synthesizer gradually recovers and generates an optimal strategic delivery scenario that meets delivery requirements. A bid command generation module is used to build and train a non-retroactive bid command deriver based on the optimal strategic delivery landscape. The non-retroactive bid command deriver generates bid commands in real time by combining historical trends and target states.

[0014] In a third aspect, the present application provides an electronic device comprising a processor and a memory; the memory stores a program, which is loaded and executed by the processor to implement an automatic advertising bidding method based on predictive delivery planning as described in the first aspect.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the storage medium stores a program, and when the program is executed by a processor, it is used to implement an automatic advertising bidding method based on predictive delivery planning as described in the first aspect.

[0016] In summary, the beneficial effects of this application include at least: (1) By constructing an optimal strategic delivery landscape covering the entire delivery cycle, a global perspective and long-term planning of the advertising bidding strategy is achieved. Unlike traditional instant response models that only make local adjustments based on the current status, the strategic delivery landscape generated by this application not only reflects the structural characteristics of historical high-quality delivery strategies, but also integrates the advertiser's current business intentions, achieving comprehensive planning and guidance for future delivery paths. This global planning capability enables the system to proactively plan future delivery nodes, effectively avoiding frequent strategy adjustments caused by short-term fluctuations, thereby improving the scientific nature and predictability of the overall delivery strategy and ensuring the steady achievement of advertising delivery goals.

[0017] (2) By generating the global optimal delivery landscape in one go, this application significantly reduces the error accumulation problem caused by step-by-step decision-making. Combined with a non-retrospective bid command deriver, the system can derive reasonable bid adjustment commands in real time and continuously based on the current and historical observation states, achieving dynamic approximation to the global planning goal. This mechanism effectively suppresses the interference of random fluctuations in the advertising environment on the bidding strategy, improves the stability and robustness of the model, makes advertising bidding decisions more reliable, avoids frequent and drastic fluctuations, and ensures the continuous optimization of delivery effects.

[0018] (3) By introducing commercial intent encoding, advertisers’ diverse delivery objectives (such as return maximization, cost control, and delivery rhythm management) are uniformly expressed as standardized instruction inputs, enabling effective integration and management of multiple objectives and multiple constraints. Based on this encoding, the conditional landscape synthesizer flexibly generates strategic delivery landscapes that meet specific delivery requirements, supporting adaptation and response to different business needs and constraints. This unified framework not only improves the model’s scope of application and scalability, but also ensures the coordination and balance of the generated delivery strategies under multiple objectives and constraints, meeting the complex and changing actual needs of advertisers.

[0019] By extracting historical high-quality strategic delivery landscapes and combining them with the advertiser's personalized commercial intent encoding, a conditional landscape synthesizer is constructed and trained to generate the optimal strategic delivery landscape covering the entire future delivery cycle as the system's global operation plan. This landscape takes into account the structural characteristics of historical high-performance strategies and the current delivery goals, achieving long-term planning and target guidance for the delivery path. Subsequently, a non-retrospective bidding command deriver is designed to combine the current and historical observation states with the target state in the strategic delivery landscape to derive in real time the optimal bidding command that drives the system to evolve toward the global planning goal, ensuring that the bidding strategy has both dynamic adaptability and can stably approach the long-term planning trajectory. Through this method, the present application effectively overcomes the problems of short-sightedness and volatility sensitivity of the immediate response model, achieves stability, flexibility and effect optimization of the advertising bidding strategy, meets the complex delivery requirements of multiple objectives and multiple constraints, and improves the overall advertising delivery effect and resource utilization efficiency.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application and to implement it in accordance with the contents of the specification, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flowchart of an automatic advertising bidding method based on predictive delivery planning in an embodiment of the present application.

[0022] Figure 2 This is a structural block diagram of an automatic advertising bidding system based on predictive delivery planning in an embodiment of the present application.

[0023] Figure 3 It is a block diagram of an electronic device for automatic advertisement bidding based on predictive delivery planning in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0025] Optionally, the present application uses the automatic advertising bidding method based on predictive delivery planning provided in each embodiment as an example for use in an electronic device, where the electronic device is a terminal or a server. The terminal can be a computer, a tablet computer, etc. This embodiment does not limit the type of electronic device.

[0026] First, referring to Table 1, the terms in the field used in this application and their English expressions and abbreviations are introduced.

[0027]

[0028] Reference Figure 1 , is a flowchart of an automatic advertising bidding method based on predictive delivery planning provided by an embodiment of the present application, which includes at least the following steps: Step S101: construct a strategic delivery landscape consisting of multi-dimensional advertising activity measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery landscapes.

[0029] In step S101, historical advertising delivery data is first collected. The historical advertising delivery data includes advertising activity-related data recorded at different time nodes in multiple completed advertising delivery cycles. The data covers the time process information, budget usage, resource consumption speed, conversion revenue performance, unit cost fluctuations and other business indicators closely related to the health of the delivery process.

[0030] Specifically, after collecting historical advertising data, each historical advertising data is discretized on the time axis using a complete advertising cycle (for example, one day) as the time unit. continuous observation nodes. At each observation node , with a multi-dimensional advertising campaign metric vector This measure vector may include the following typical dimensions: a progress indicator representing the remaining proportion of the cycle (e.g. ), budget resource indicators (such as remaining budget and budget consumption rate), performance indicators (such as current cumulative return on investment and unit cost performance), and other characteristic indicators used to characterize the stability and continuity of advertising delivery or other business attributes. These dimensions can be flexibly configured and expanded according to actual business needs.

[0031] By collecting The advertising activity measurement vectors on the observation nodes are used to construct a data matrix to characterize the dynamic evolution process of the entire cycle, which is recorded as the strategic delivery landscape. : ; in, The strategic deployment picture of a complete advertising cycle is essentially a two-dimensional data matrix composed of multiple advertising campaign measurement vectors in chronological order. Indicates in The advertising activity measurement vector recorded at each observation node is used to describe the delivery status at that point in time; It represents the total number of observation nodes obtained after discretizing the advertising delivery cycle. Indicates the number of characteristic dimensions contained in each campaign measurement vector, that is, the number of measurement indicators, such as cycle progress, budget status, performance, and other dimensions; Indicates a OK A real matrix with columns.

[0032] In implementation, the strategic deployment landscape The matrix fully expresses the joint evolution trajectory of advertising campaigns in the time dimension and feature space. Based on a large number of historical strategic delivery scenarios, the quality of each scenario is evaluated by setting evaluation criteria, such as measuring its comprehensive performance in multiple dimensions such as resource utilization efficiency, delivery stability, and target achievement rate. Then, the scenario samples with excellent performance are selected as high-quality strategic delivery scenarios, which are recorded as .

[0033] Optionally, this application does not restrict the specific screening requirements for high-quality strategic deployment scenarios, which are determined by the choices made during specific implementation.

[0034] Step S102: Encode the advertiser's delivery requirements into a standardized commercial intent code.

[0035] In step S102, the advertiser's diverse and subjective delivery requirements are uniformly encoded into a standardized, machine-understandable commercial intent code, denoted as ,in Indicates the number of encoding dimensions, Represents a set of real numbers. The commercial intent encoding is designed to clearly quantify complex advertising delivery objectives and constraints, serving as guiding constraints for subsequently generating an optimal strategic delivery landscape.

[0036] In a preferred embodiment, the commercial intent code Includes but is not limited to core optimization intent, boundary constraints, and process shape preferences; the core optimization intent is used to express the advertiser's main goal, such as maximizing return on investment. The intent is normalized into a scalar value, such as Boundary constraints are used to describe constraints that must be met, such as the average cost per click (CPC) must be lower than a preset threshold. Such constraints are represented by binary indicator variables. Indicates that the constraint must be met, and 0 indicates no such restriction. The process form preference is used to reflect the advertiser's preference for the form of the delivery process, for example, budget consumption should be smoother. This preference is obtained by first calculating the corresponding quantitative indicators and then encoding whether the indicators meet the requirements into a binary variable, for example Indicates that the smoothing requirement is met.

[0037] The above commercial intent encoding effectively converts advertisers' diverse delivery requirements into structured numerical vectors, facilitating input and processing by subsequent machine learning models. Alternatively, in actual implementation, advertisers' natural language requirements or option inputs can be converted into structured commercial intent encodings through rule-based methods, questionnaire mapping, or feature templates using predefined labels. This application does not limit the specific encoding method.

[0038] Step S103: Utilize the extracted high-quality strategic delivery landscape and the advertiser's commercial intent encoding to construct a forward perturbation and conditional reverse recovery process to train a preset conditional landscape synthesizer. In the actual bidding stage, starting from a random initial strategic delivery landscape, the conditional landscape synthesizer gradually recovers and generates an optimal strategic delivery landscape that meets the delivery requirements.

[0039] In step S103, after extracting high-quality strategic delivery scenarios and encoding the advertiser's personalized and diverse delivery requirements into standardized commercial intent codes, a conditionalized scenario synthesizer is used to generate a clearly structured, high-quality optimal strategic delivery scenario that aligns with current commercial intent, covers the entire future delivery cycle, and is equivalent to a detailed, life-long operational plan that defines the ideal state the system should achieve at each observation node. The conditionalized scenario synthesizer learns the complex data distribution of historical high-quality strategic delivery scenarios and, guided by the advertiser's commercial intent, performs conditional generation, ensuring that the generated results both conform to the structural characteristics of historical high-performing strategies and meet the specific requirements of current delivery goals. Its working mechanism is based on the joint modeling and learning of a controllable random perturbation process and its inverse recovery process, and is divided into two phases: a forward perturbation process and a conditionalized inverse recovery process.

[0040] Specifically, in the forward perturbation process, the extracted high-quality strategy is first presented with a picture The Gaussian perturbation of preset intensity is gradually injected into the image to obtain image representations of different perturbation levels. The fundamental purpose of this process is to construct a continuous perturbation trajectory sequence to train the subsequent model to perform reverse recovery of the image state under any perturbation degree, so as to grasp the deep structural characteristics of the image distribution. The execution of the perturbation is based on the set maximum perturbation step number. The disturbance process can be understood as the deployment of high-quality strategies This clear and complete original image is gradually injected with noise, making the image gradually blurred or degraded at each step, and eventually evolving into an image completely dominated by noise. During the training process, it is from these degraded images of varying degrees that we learn how to restore them back to high-quality images under the guidance of business intentions. The disturbance picture at the current moment The perturbation picture at the previous moment Generate, through the closed function can calculate the disturbance picture at any time in one step, its mathematical form is: ; in, is a standard Gaussian random matrix, It is preset, Changing the signal-to-noise ratio control parameter, is the perturbation function. Through the above controllable noise injection, a full set of training samples from slightly degraded to severely degraded can be obtained. The training samples not only contain the original high-quality strategic deployment picture , and also includes strategic deployment scenarios with different degrees of degradation.

[0041] In the process of conditional reverse recovery, the goal of the conditional image synthesizer is to learn how to gradually recover the original high-quality strategic deployment images from these strategic deployment images that have been disturbed to varying degrees. The core of this process is to build a deep neural network with strong expressive power. , in this application, It is a transformation network suitable for matrix data, which can effectively model the degradation law of the image under noise disturbance and perform reverse reasoning based on the triple conditions of disturbance image, disturbance degree and business intention. The input includes the picture under the current disturbance level , the current perturbation time step and business intent encoding , whose task is to accurately predict the original perturbation injected in this step given the above triple input , that is, the network must learn to restore the noise component corresponding to each step of the degraded image. In this way, the model essentially learns the path for the image to return to a clear state from the noise, thus having the ability to generate high-quality images. Neural Network The optimization goal of training is to minimize the following expected error : ; in, represents the expected value of all samples in the training data, is the real Gaussian noise injected during the forward perturbation process, represents the model’s predicted value for the noise, Represents the Frobenius norm of the matrix. Through optimization, the neural network It can gradually improve its ability to recover from disturbance levels, and eventually have the ability to recover a strategic deployment picture with clear structure and consistent semantics from any degraded picture under the guidance of business intentions.

[0042] In the actual bidding stage, from a completely random initial strategic delivery landscape Starting from the commercial intent encoding of a given advertiser Under the guidance of ,by As input, execute Wheel reverse recovery operation, that is, at each time step , deep neural networks Receive the current disturbance landscape , the current perturbation time step and business intent encoding , by reverse reasoning, estimate the original perturbation in the current image , and calculate a clearer picture of the previous step based on the estimated results The process is iterated continuously, making the image gradually clearer and more stable from the initial Gaussian noise, and finally When generating a clear, theoretically optimal strategic deployment picture tailored for the current advertiser .

[0043] Optionally, the conditional scene synthesizer is essentially a generative model with conditional generation capabilities. This application does not limit the specific type or network structure of the conditional scene synthesizer used. As long as the model has the ability to learn distribution features from historical scenes and realize scene generation under the guidance of business intentions, it can be used as an implementation method of this application.

[0044] Step S104: construct and train a non-retroactive bidding command derivation device based on the optimal strategic delivery landscape. The non-retroactive bidding command derivation device generates bidding commands in real time based on historical trends and target states.

[0045] In step S104, generate the optimal strategic deployment scenario ,in Indicates that the system The desired measurement state at each observation node is used to describe the target operation trajectory of the system throughout the entire cycle. In order to push the actual operation process of the system to be as close as possible to the trajectory depicted by the picture, an efficient, stable and real-time responsive mechanism is needed to convert the target state of each node in the picture into executable bidding operation instructions. To this end, this step constructs a non-retrospective bidding command deriver, whose task is to combine the current and recent historical observation states with the target state at the next moment during the system operation, and derive the optimal bidding command that drives the system to evolve towards the target state in real time. .

[0046] The design goal of this derivation is to answer the following questions: In order to make the next state of the system as close as possible to the planned target state , what bidding command should be executed .

[0047] To this end, it is proposed to model the deriver as a function , its mathematical form is as follows: ; in, For the current moment The bidding command to be executed includes the adjustment amount of each adjustable parameter in the bidding system; Indicates that the current time is the end point and the length is The historical measurement observation window of Indicates that the system is at time The operational measurement status of Indicates the current image The target state of each observation node. The non-backtracking nature of this function means that its input depends only on the current node and its previous observation states, without involving future measurement information. This makes it suitable for online real-time reasoning. In actual deployments, by introducing a history window of length H rather than relying solely on the current measurement state, the inference function can more fully capture short-term dynamic trends in system operation, improving the robustness of command generation and avoiding overreaction to transient fluctuations.

[0048] In actual operation, the non-retrospective bidding command derivation is called in real time at each observation node. The system first collects the actual operation metrics of the past H time steps. , and combined with the current picture The target state of the observation node , as input and input into the derivation, the derivation performs feature extraction and deep modeling on the input, and outputs the bidding command that should be executed at the current moment .

[0049] In addition, as an example, a standard supervised learning method is used to train the non-retroactive bidding command derivation device. The specific process is as follows: a large-scale training sample pair is constructed from historical advertising data. , where the input of each sample is It is a combination of the historical measurement window and the next measurement vector actually achieved, that is, the historical observation sequence at a certain moment and the next target state preset in the picture, which together reflect the input context at that time, and the label This is the bid command actually executed in this historical window. , which represents the bidding operation made by the system to achieve business goals in the current delivery environment.

[0050] The training goal of this derivation is to minimize the distance between the predicted command and the true command, for example, using the L2 loss function as follows: ; in, represents the expectation on all training sample pairs, Indicates the bid commands actually executed in history. Represents the model based on the input Predicted bid commands. Through the above training process, the non-retrospective bid command deriver can learn the decision-making patterns summarized from historical operation trajectories and image target states, thereby achieving adaptive generation of bidding behaviors in complex delivery environments, ensuring that the automatic advertising bidding method has good robustness and convergence while meeting real-time requirements.

[0051] In summary, by extracting historical high-quality strategic delivery landscapes and combining them with the advertiser's personalized commercial intent encoding, a conditional landscape synthesizer is constructed and trained to generate the optimal strategic delivery landscape covering the entire future delivery cycle as the system's global operation plan. This landscape takes into account the structural characteristics of historical high-performance strategies and the current delivery goals, achieving long-term planning and target guidance for the delivery path. Subsequently, a non-retrospective bidding command deriver is designed to combine the current and historical observation states with the target state in the strategic delivery landscape to derive in real time the optimal bidding command that drives the system to evolve toward the global planning goal, ensuring that the bidding strategy has both dynamic adaptability and can stably approach the long-term planning trajectory. Through this method, the present application effectively overcomes the problems of short-sightedness and volatility sensitivity of the immediate response model, achieves stability, flexibility and effect optimization of the advertising bidding strategy, meets the complex delivery requirements of multiple objectives and multiple constraints, and improves the overall advertising delivery effect and resource utilization efficiency.

[0052] Figure 2This is a structural block diagram of an automatic advertising bidding system based on predictive delivery planning provided by an embodiment of the present application. The system includes at least the following modules: A high-quality picture extraction module is used to construct a strategic delivery picture composed of multi-dimensional advertising activity measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery pictures; The delivery demand encoding module is used to encode the advertiser's delivery demand into a standardized commercial intent code; The optimal image generation module is used to use the extracted high-quality strategic delivery images and the advertiser's commercial intent encoding to construct a forward perturbation and conditional reverse recovery process to train a preset conditional image synthesizer. During the actual bidding phase, starting from a random initial strategic delivery image, the conditional image synthesizer gradually recovers and generates the optimal strategic delivery image that meets the delivery requirements; The bid command generation module is used to build and train a non-retroactive bid command derivation device based on the optimal strategic delivery landscape. The non-retroactive bid command derivation device generates bid commands in real time by combining historical trends and target states.

[0053] For relevant details, please refer to the above method embodiment.

[0054] Figure 3 4 is a block diagram of an electronic device provided in one embodiment of the present application. The device includes at least a processor 401 and a memory 402.

[0055] Processor 401 may include one or more processing cores, such as a quad-core processor or an octa-core processor. Processor 401 may be implemented in hardware using at least one of the following: a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), or a PLA (Programmable Logic Array). Processor 401 may also include a main processor and a coprocessor. The main processor is used to process data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 401 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing content displayed on the display screen. In some embodiments, processor 401 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0056] Memory 402 may include one or more computer-readable storage media, which may be non-transitory. Memory 402 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in memory 402 is used to store at least one instruction, which is executed by processor 401 to implement the automatic advertising bidding method based on predictive delivery planning provided in the method embodiment of the present application.

[0057] In some embodiments, the electronic device may optionally include a peripheral device interface and at least one peripheral device. The processor 401, memory 402, and peripheral device interface may be connected via a bus or signal lines. Each peripheral device may be connected to the peripheral device interface via a bus, signal lines, or circuit boards. Illustratively, the peripheral devices include, but are not limited to, radio frequency circuitry, a touchscreen display, audio circuitry, and a power supply.

[0058] Of course, the electronic device may also include fewer or more components, which is not limited in this embodiment.

[0059] Optionally, the present application also provides a computer-readable storage medium, in which a program is stored. The program is loaded and executed by a processor to implement the automatic advertising bidding method based on predictive delivery planning of the above method embodiment.

[0060] Optionally, the present application also provides a computer product, which includes a computer-readable storage medium, in which a program is stored. The program is loaded and executed by a processor to implement the automatic advertising bidding method based on predictive delivery planning of the above method embodiment.

[0061] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. An automatic advertising bidding method based on predictive delivery planning, characterized in that: The method comprises: Build a strategic delivery landscape consisting of multi-dimensional advertising campaign measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery landscapes; Encode advertisers' placement requirements into standardized commercial intent codes; The extracted high-quality strategic delivery images and the advertiser's commercial intent encoding are used to construct a forward perturbation and conditional reverse recovery process to train a preset conditional image synthesizer. During the actual bidding phase, starting from a random initial strategic delivery image, the conditional image synthesizer gradually recovers and generates an optimal strategic delivery image that meets delivery requirements; A non-retroactive bidding command derivation device is constructed and trained based on the optimal strategic delivery landscape. The non-retroactive bidding command derivation device generates bidding commands in real time by combining historical trends with target states.

2. The automatic advertising bidding method based on predictive delivery planning according to claim 1, characterized in that: The construction of a strategic delivery picture composed of multi-dimensional advertising activity measurement vectors based on historical advertising delivery data includes: Collect historical advertising data, and discretize each historical advertising data into a complete advertising cycle as the time unit on the time axis. continuous observation nodes, at each observation node , with a multi-dimensional advertising campaign metric vector To represent the running status of the advertisement at that moment; By collecting The advertising activity measurement vectors on the observation nodes are used to construct a data matrix to characterize the dynamic evolution process of the entire cycle, which is recorded as the strategic delivery landscape. : ； in, It represents the strategic deployment picture of a complete advertising cycle. Indicates in The advertising activity measurement vector recorded on each observation node; It represents the total number of observation nodes obtained after discretizing the advertising delivery cycle. Indicates the number of feature dimensions contained in each campaign measurement vector, Indicates a OK A real matrix with columns.

3. The automatic advertising bidding method based on predictive delivery planning according to claim 1, characterized in that: The encoding of the advertiser's delivery requirements into a standardized commercial intent code includes: Advertisers’ delivery requirements are uniformly coded into standardized commercial intent codes, recorded as ,in Indicates the number of encoding dimensions, represents the set of real numbers; Commercial Intent Encoding It includes core optimization intent, boundary constraints and process form preferences; the core optimization intent is used to express the advertiser's main goal, the boundary constraints are used to describe the constraints that must be met, and the process form preferences are used to reflect the advertiser's preference for the delivery process form.

4. The automatic advertising bidding method based on predictive delivery planning according to claim 1, characterized in that: The method of using the extracted high-quality strategic placement image and the advertiser's commercial intent encoding to construct a forward perturbation and conditional reverse recovery process to train a preset conditional image synthesizer includes: During the forward perturbation process, the extracted high-quality strategy is presented with a picture Gaussian perturbations of preset strength are gradually injected into the image to obtain different perturbation levels. The perturbation is performed based on the set maximum perturbation step number. , at each step The disturbance picture at the current moment The perturbation picture at the previous moment Generate, through the closed function can calculate the disturbance picture at any time in one step, its mathematical form is: ； in, is a standard Gaussian random matrix, It is preset, Changing the signal-to-noise ratio control parameter, is the perturbation function; Building a deep neural network during conditioned reverse recovery , neural network The input includes the picture under the current disturbance level , the current perturbation time step and business intent encoding , whose task is to predict the original perturbation injected in this step given the triple input , neural network The optimization goal of training is to minimize the following expected error : ； in, represents the expected value of all samples in the training data, is the real Gaussian noise injected during the forward perturbation process, represents the model’s predicted value for the noise, Represents the Frobenius norm of the matrix.

5. The automatic advertising bidding method based on predictive delivery planning according to claim 4 is characterized in that: The process of starting from a random initial strategic delivery scenario in the actual bidding phase and gradually recovering and generating an optimal strategic delivery scenario that meets delivery requirements through the conditional scenario synthesizer includes: In the actual bidding stage, from a completely random initial strategic delivery landscape Starting from the commercial intent encoding of a given advertiser Under the guidance of ,by As input, execute Wheel reverse recovery operation, that is, at each time step , deep neural networks Receive the current disturbance landscape , the current perturbation time step and business intent encoding , by reverse reasoning, estimate the original perturbation in the current image , and calculate the previous step picture based on the estimated results The process is iterated continuously, making the image gradually clearer and more stable from the initial Gaussian noise. Generate the optimal strategic deployment picture .

6. The automatic advertising bidding method based on predictive delivery planning according to claim 2, characterized in that: The non-retroactive bidding command derivation device is constructed and trained based on the optimal strategic delivery scenario, wherein the non-retroactive bidding command derivation device generates bidding commands in real time by combining historical trends and target states, including: Construct a non-retrospective bidding command deriver, whose task is to combine the current and recent historical observation states with the next moment target state during the system operation, and derive the optimal bidding command to drive the system towards the target state in real time. ; The design goal of this derivation is to answer the following questions: In order to make the next state of the system as close as possible to the planned target state , what bidding command should be executed ; Model the derivation as a function , its mathematical form is as follows: ； in, For the current moment The bid command that should be executed, Indicates that the current time is the end point and the length is The historical measurement observation window of Indicates that the system is at time The operational measurement status of Indicates the current image The target state of each observation node; The non-retrospective bid command deriver is called in real time on each observation node to collect the actual running metrics of the past H time steps , and combined with the current picture The target state of the observation node , as input and input into the derivation, the derivation performs feature extraction and deep modeling on the input, and outputs the bidding command that should be executed at the current moment .

7. The automatic advertising bidding method based on predictive delivery planning according to claim 6, characterized in that: The non-retrospective bidding command derivation is trained using a standard supervised learning method. The specific process is as follows: a large-scale training sample pair is constructed from historical advertising data. , where the input of each sample is It is a combination of the historical measurement window and the next measurement vector actually achieved, that is, the historical observation sequence at a certain moment and the next target state preset in the picture, label The bid command actually executed in this historical window ; The training goal of the non-retrospective bidding command deriver is to minimize the distance between the predicted command and the true command, using the L2 loss function as follows: ； in, represents the expectation on all training sample pairs, Indicates the bid commands actually executed in history. Represents the model based on the input The predicted bid command.

8. An automatic advertising bidding system based on predictive delivery planning, characterized in that: include: A high-quality picture extraction module is used to construct a strategic delivery picture composed of multi-dimensional advertising activity measurement vectors based on historical advertising delivery data, and extract high-quality strategic delivery pictures; The delivery demand encoding module is used to encode the advertiser's delivery demand into a standardized commercial intent code; An optimal scenario generation module is used to use the extracted high-quality strategic delivery scenarios and the advertiser's commercial intent encoding to construct a preset conditional scenario synthesizer for training through a forward perturbation and conditional reverse recovery process. During the actual bidding phase, starting from a random initial strategic delivery scenario, the conditional scenario synthesizer gradually recovers and generates an optimal strategic delivery scenario that meets delivery requirements. A bid command generation module is used to build and train a non-retroactive bid command deriver based on the optimal strategic delivery landscape. The non-retroactive bid command deriver generates bid commands in real time by combining historical trends and target states.

9. An electronic device, characterized in that: The device includes a processor and a memory; a program is stored in the memory, and the program is loaded and executed by the processor to implement an automatic advertising bidding method based on predictive delivery planning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The storage medium stores a program, which, when executed by a processor, is used to implement an automatic advertising bidding method based on predictive delivery planning as described in any one of claims 1 to 7.

Images