An advertisement exposure estimation method, device, equipment and storage medium

Abstract

Embodiments of the application disclose an advertisement exposure estimation method, device and equipment in the field of artificial intelligence, and a storage medium. The method comprises the following steps: determining candidate exposure inventory according to a targeting condition of a target advertisement; determining a winning rate of the target advertisement in a candidate exposure opportunity according to bidding information and advertisement characteristics of the target advertisement and opportunity characteristics of the candidate exposure opportunity by using an exposure winning rate estimation model for the candidate exposure opportunity in the candidate exposure inventory; the exposure winning rate estimation model is trained based on bidding information and advertisement characteristics of a plurality of candidate advertisements in a historical exposure opportunity, opportunity characteristics of the historical exposure opportunity, and exposure results of the historical exposure opportunity; and determining an exposure estimation result corresponding to the target advertisement according to the winning rate of the target advertisement in the candidate exposure opportunity. The method can comprehensively consider competition among advertisements in the process of estimating advertisement exposure, and improve the accuracy of the advertisement exposure estimation result.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence (AI) technology, and in particular to an advertising exposure estimation method, apparatus, device, and storage medium. Background Art

[0002] Bidding advertising is a new form of online advertising that is independently placed and managed by users, with payment based on advertising performance. To help advertisers better manage their ads, many advertising platforms estimate the exposure of their ads over a period of time and provide relevant reference information based on these estimates.

[0003] In related technologies, advertising delivery platforms mainly use the following methods to achieve advertising exposure estimation: for advertisements whose exposure is to be estimated, the characteristics of the advertisement itself (such as the size of the advertising material, the attractiveness of the advertising material, etc.), the targeting characteristics of the advertisement (such as the characteristics of the target audience of the advertisement, etc.) and the bidding information are input into a pre-trained regression model, and the regression model is used to estimate the exposure of the advertisement in a certain period of time in the future.

[0004] In actual advertising scenarios, tens of millions of ads usually compete for each exposure opportunity. However, the above-mentioned advertising exposure estimation method only considers the situation of the ad itself when estimating the exposure of the ad, and ignores the competition between ads in the actual scenario. As a result, the advertising exposure estimation results obtained by the above-mentioned advertising exposure estimation method are less accurate. Summary of the Invention

[0005] The embodiments of the present application provide an advertising exposure estimation method, apparatus, device, and storage medium, which can comprehensively consider the competition between advertisements in the process of estimating advertising exposure, thereby improving the accuracy of advertising exposure estimation results.

[0006] In view of this, a first aspect of the present application provides an advertisement exposure estimation method, the method comprising:

[0007] Determining a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions;

[0008] For the candidate exposure opportunities in the candidate exposure inventory, an exposure win rate prediction model is used to determine the target ad's win rate in the candidate exposure opportunities based on the bid information and ad features of the target ad, as well as the opportunity features of the candidate exposure opportunities. The exposure win rate prediction model is trained based on the bid information and ad features of multiple candidate ads in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target ad being exposed through the candidate exposure opportunities.

[0009] An exposure estimation result corresponding to the target advertisement is determined according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0010] A second aspect of the present application provides an advertising exposure estimation device, comprising:

[0011] An exposure inventory determination module is configured to determine a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions;

[0012] A win rate estimation module is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, the win rate of the target advertisement in the candidate exposure opportunities using an exposure win rate estimation model, based on the bid information and ad features of the target advertisement, and the opportunity features of the candidate exposure opportunities. The exposure win rate estimation model is trained based on the bid information and ad features of each of multiple candidate advertisements in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target advertisement being exposed through the candidate exposure opportunities.

[0013] The exposure result determination module is configured to determine an exposure estimation result corresponding to the target advertisement according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0014] A third aspect of the present application provides a device, comprising a processor and a memory:

[0015] The memory is used to store computer programs;

[0016] The processor is configured to execute the steps of the advertising exposure estimation method as described in the first aspect according to the computer program.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the steps of the advertising exposure estimation method described in the first aspect.

[0018] In a fifth aspect, the present application provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the advertising exposure estimation method described in the first aspect.

[0019] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0020] The embodiment of the present application provides an advertising exposure estimation method, which innovatively starts from the dimension of exposure opportunities and estimates the exposure of the target advertisement in the future period based on the winning rate of the target advertisement in the candidate exposure opportunities determined by the exposure winning rate estimation model. Specifically, in the advertising exposure estimation method provided in the embodiment of the present application, first, based on the targeting conditions of the target advertisement, a candidate exposure inventory consisting of candidate exposure opportunities matching the targeting conditions is determined; then, for the candidate exposure opportunities in the candidate exposure inventory, the exposure winning rate estimation model is used to determine the winning rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertising features of the target advertisement and the opportunity characteristics of the candidate exposure opportunities. The winning rate can characterize the probability of the target advertisement being exposed through the candidate exposure opportunity; and then, based on the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory, the exposure estimation result corresponding to the target advertisement is determined. Since the exposure winning rate prediction model is trained based on the bidding information and advertising features of multiple candidate advertisements in historical exposure opportunities, the opportunity features of historical exposure opportunities, and the exposure results of historical exposure opportunities, the exposure winning rate prediction model can learn which advertisement is more likely to win in the competition for exposure opportunities based on training samples during the training process; accordingly, when determining the winning rate of the target advertisement in the candidate exposure opportunities, the exposure winning rate prediction model can comprehensively consider the competition between the target advertisement and other advertisements, and then determine the probability of the target advertisement winning among the many candidate advertisements corresponding to the candidate exposure opportunities; determining the exposure estimation result corresponding to the target advertisement based on the winning rate estimated by the exposure winning rate prediction model can ensure the accuracy of the determined exposure estimation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the implementation principle of the advertising exposure process;

[0022] Figure 2 An application scenario of the advertisement exposure estimation method provided by the embodiments of the present application is shown in the figure;

[0023] Figure 3 A flowchart of the advertisement exposure estimation method provided by the embodiments of the present application is shown in the figure;

[0024] Figure 4 An interface for modifying bidding information provided by the embodiments of the present application is shown in the figure;

[0025] Figure 5 An interface for displaying advertisement delivery reference information provided by the embodiments of the present application is shown in the figure;

[0026] Figure 6 An interface for displaying exposure trend provided by the embodiments of the present application is shown in the figure;

[0027] Figure 7 A flowchart of the exposure winning rate estimation model training method provided by the embodiments of the present application is shown in the figure;

[0028] Figure 8 An implementation principle of the advertisement exposure estimation method provided by the embodiments of the present application is shown in the figure;

[0029] Figure 9 An implementation principle of the estimated advertisement exposure result provided by the embodiments of the present application is shown in the figure;

[0030] Figure 10 A structure diagram of a first advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0031] Figure 11 A structure diagram of a second advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0032] Figure 12 A structure diagram of a third advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0033] Figure 13 A structure diagram of a fourth advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0034] Figure 14 A structure diagram of a fifth advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0035] Figure 15 A structure diagram of a sixth advertisement exposure estimation device provided by the embodiments of the present application is shown in the figure;

[0036] Figure 16 A structure diagram of a terminal device provided by the embodiments of the present application is shown in the figure;

[0037] Figure 17A schematic diagram of the structure of the server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0039] The terms "first," "second," "third," "fourth," etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential sequence. It should be understood that the numbers used in this way are interchangeable where appropriate so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or apparatus.

[0040] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0041] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0042] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0043] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0044] The solutions provided in the embodiments of this application involve artificial intelligence machine learning technology, which is specifically illustrated by the following embodiments:

[0045] In order to facilitate the understanding of the technical solutions provided by the embodiments of the present application, Figure 1 The schematic diagram of the implementation principle of the advertising exposure process shown in the figure introduces the exposure process of bidding advertising. Figure 1 As shown in the figure, for one exposure (Pageview, PV), it is necessary to first use recall processing (Matching) to screen out advertisements whose targeting conditions match this exposure from tens of millions of advertisements. After the recall processing, tens of thousands of advertisements are usually still retained; then, it is necessary to perform rough ranking (Ranking) processing on the advertisements retained after the recall processing based on the personal historical browsing habits and / or personal interests of the users facing this exposure, so as to retain thousands of advertisements; furthermore, it is necessary to perform fine ranking (Re-ranking) processing on the advertisements retained after the rough ranking processing based on their respective corresponding click-through rates (CTR) and conversion rates (CVR) to retain hundreds of advertisements; finally, one advertisement that is played through this exposure is selected from these hundreds of advertisements.

[0046] In related technologies, advertising platforms typically estimate ad exposure based on the dimensions of the ad itself. This involves using regression models to estimate the ad's exposure over the next period of time based on the ad's own characteristics (such as the size of the ad creative, the attractiveness of the ad creative, etc.), targeting characteristics (such as the characteristics of the target audience), and bidding information. As mentioned above, in actual ad exposure scenarios, tens of millions of ads compete for each exposure opportunity. However, when estimating ad exposure using the aforementioned ad exposure estimation method, only the relevant characteristics of the ad itself are considered, without considering competition between ads. Consequently, the accuracy of the ad exposure estimation results obtained using this method is low.

[0047] In response to the problems existing in the above-mentioned related technologies, an embodiment of the present application provides an advertising exposure estimation method, which can comprehensively consider the competition between advertisements when estimating the exposure of advertisements, thereby improving the accuracy of the advertising exposure estimation results.

[0048] Specifically, in the advertising exposure estimation method provided in the embodiment of the present application, a candidate exposure inventory is first determined based on the targeting conditions of the target advertisement, and the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions; then, for the candidate exposure opportunities in the candidate exposure inventory, an exposure winning rate estimation model is used to determine the winning rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertising characteristics of the target advertisement and the opportunity characteristics of the candidate exposure opportunities, and the winning rate is used to characterize the probability of the target advertisement being exposed through the candidate exposure opportunities; the exposure winning rate estimation model is trained based on the bid information and advertising characteristics of multiple candidate advertisements in historical exposure opportunities, the opportunity characteristics of the historical exposure opportunities, and the exposure results of the historical exposure opportunities; and further, based on the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory, the exposure estimation result corresponding to the target advertisement is determined.

[0049] Compared to the implementation method in the related art that starts from the dimension of advertisement and estimates advertisement exposure only based on the relevant features of the advertisement itself, the advertisement exposure estimation method provided in the embodiment of the present application innovatively starts from the dimension of exposure opportunity and determines the exposure estimation result corresponding to the target advertisement based on the winning rate of the target advertisement in the candidate exposure opportunities estimated by the exposure winning rate estimation model. Since the exposure winning rate estimation model is trained based on the bidding information and advertising features of multiple candidate advertisements in historical exposure opportunities, the opportunity features of historical exposure opportunities, and the exposure results of historical exposure opportunities, the exposure winning rate estimation model can learn which advertisement is more likely to win in the competition for exposure opportunities based on the training samples during the training process; accordingly, when estimating the winning rate of the target advertisement in the candidate exposure opportunities, the exposure winning rate estimation model will also comprehensively consider the competition between the target advertisement and other advertisements; furthermore, the exposure estimation result corresponding to the target advertisement is determined based on the winning rate estimated by the exposure winning rate estimation model, which can ensure the accuracy of the determined exposure estimation result.

[0050] It should be understood that the advertising exposure estimation method provided in the embodiments of the present application can be executed by a device for implementing the advertising exposure estimation task, which can be a terminal device or a server. The terminal device can specifically be a computer, a smartphone, a tablet computer, a personal digital assistant (PDA), etc.; the server can specifically be an application server or a web server. In actual deployment, it can be a standalone server, a cluster server, or a cloud server.

[0051] To facilitate understanding of the advertising exposure estimation method provided in the embodiments of the present application, the following uses a server as an example to illustrate an application scenario of the advertising exposure estimation method.

[0052] See also Figure 2 , Figure 2 Schematic diagram of the application scenario of the advertising exposure estimation method provided in the embodiment of the present application. Figure 2 As shown, the application scenario includes a server 210 and a database 220. The server 210 can access the database 220 through a network, or the database 220 can be integrated into the server 210. The server 210 is used to execute the advertising exposure estimation method provided in the embodiment of the present application, and estimate the exposure of the target advertisement for which exposure is to be estimated within a period of time in the future; the database 220 is used to store advertisements delivered on the target advertising delivery platform and their corresponding related information (such as targeting conditions, bidding information, advertising features, etc.).

[0053] In actual applications, when server 210 needs to estimate the exposure of a target advertisement within a certain period of time in the future (e.g., the next hour), server 210 can retrieve the targeting conditions corresponding to the target advertisement from database 220. The targeting conditions are used to indicate the exposure conditions corresponding to the target advertisement, such as the exposure region corresponding to the target advertisement and the characteristics of the users to whom the target advertisement is exposed. Furthermore, server 210 can determine a candidate exposure inventory based on the targeting conditions of the target advertisement. The candidate exposure inventory includes all candidate exposure opportunities that match the targeting conditions within the next period of time.

[0054] Then, server 210 can use a pre-trained exposure win rate prediction model to determine the target ad's win rate for the candidate exposure opportunities in the candidate exposure inventory based on the target ad's bid information and ad features, as well as the opportunity features of the candidate exposure opportunities. This win rate can represent the probability of the target ad being exposed through the candidate exposure opportunity. The target ad's bid information and ad features can be retrieved by server 210 from database 220. The opportunity features of the candidate exposure opportunities can include: viewing target features corresponding to the candidate exposure opportunity (such as the gender, age, and region of the viewing user corresponding to the candidate exposure opportunity), and exposure condition features corresponding to the candidate exposure opportunity (such as the exposure website, exposure placement, and exposure time corresponding to the candidate exposure opportunity). The above-mentioned exposure win rate prediction model is trained based on the bid information and ad features of multiple candidate ads in historical exposure opportunities, the opportunity features of historical exposure opportunities, and the exposure results of historical exposure opportunities. During the training process, the exposure win rate prediction model can learn from the training sample data which ads are more likely to win in the competition for exposure opportunities.

[0055] It should be understood that, in actual applications, server 210 may perform the above processing on only a portion of the candidate exposure opportunities in the candidate exposure inventory. That is, server 210 may first screen out a portion of the candidate exposure opportunities that are more likely to expose the target advertisement from the candidate exposure inventory, and then perform the above processing on only these candidate exposure opportunities. Of course, server 210 may also perform the above processing on all candidate exposure opportunities in the candidate exposure inventory.

[0056] Furthermore, server 210 determines an estimated exposure result for the target ad based on the target ad's win rate among the candidate exposure opportunities included in the candidate exposure inventory. For example, server 210 may calculate the sum of the target ad's win rates across multiple candidate exposure opportunities as the estimated exposure result for the target ad.

[0057] It should be understood that Figure 1The application scenarios shown are merely examples. In actual applications, server 210 may also interact with terminal devices to implement the advertising exposure estimation method provided in the embodiments of this application. For example, the terminal device may send the targeting conditions of a target advertisement, as well as the bid information and advertisement features of the target advertisement to server 210. Server 210 then determines an exposure estimation result corresponding to the target advertisement based on the information sent by the terminal device. The application scenarios of the advertising exposure estimation method provided in the embodiments of this application are not limited herein.

[0058] The following is a detailed introduction to the advertising exposure estimation method provided by this application through a method embodiment.

[0059] See also Figure 3 , Figure 3 This is a flow chart of the advertising exposure estimation method provided in the embodiment of the present application. For the sake of convenience, the following embodiment is still introduced by taking the execution subject of the advertising exposure estimation method as the server as an example. Figure 3 As shown, the advertising exposure estimation method includes the following steps:

[0060] Step 301: Determine a candidate exposure inventory based on the targeting conditions of a target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions.

[0061] When the server needs to estimate the exposure of a target advertisement in the future, the server may first determine a candidate exposure inventory based on the targeting conditions of the target advertisement. The candidate exposure inventory includes all candidate exposure opportunities that match the targeting conditions of the target advertisement in the future.

[0062] It should be noted that the targeting conditions of target ads are essentially the exposure conditions corresponding to the target ads, which may include at least one of the following elements: the requirements of the target ads for the users they are exposed to (such as user gender, user age, user address, etc.), the requirements of the target ads for traffic attributes (such as the exposure website of the target ads, the placement of the target ads, the exposure time of the target ads, etc.). Of course, in actual applications, other types of elements can be added to the targeting conditions according to actual needs. This application does not impose any restrictions on the targeting conditions of target ads.

[0063] It should be noted that the candidate exposure opportunities included in the candidate exposure inventory are exposure opportunities that match the targeting conditions of the target advertisement within a period of time in the future, which can be understood as exposure opportunities for the target advertisement to compete for; the period of time in the future here can be determined according to actual needs, for example, it can be one hour after the current time point. This application does not impose any restrictions on the time period to which the candidate exposure opportunity belongs.

[0064] In one possible implementation, the server may predict candidate exposure inventory based on the targeting conditions of the target advertisement through a deep spatiotemporal tensor decomposition model. The deep spatiotemporal tensor decomposition model is used to predict the exposure inventory of the target advertisement within a preset time period through tensor decomposition based on the attribute combination characteristics of the targeting conditions and the temporal characteristics of historical advertisement delivery data on the target advertisement delivery platform. Here, the target advertisement delivery platform is the advertisement delivery platform used to deliver the target advertisement.

[0065] The following is an introduction to the deep spatiotemporal tensor decomposition model: The deep spatiotemporal tensor decomposition model mainly consists of three parts: the spatial model, the temporal feature modeling structure, and the multi-task model.

[0066] The spatial model is used to extract the attribute combination features of the directional conditions. For the input attribute combination of the directional conditions, the spatial model can first use the feature mapping layer (Attention Embedding) based on the attention mechanism to extract the relationship between different values ​​of the same category attributes. The existence of the attention mechanism can make the spatial model pay more attention to the key input parts. After obtaining the encoding of all attributes, the spatial model can splice the encoding of all attributes as the input of the DeepCross network, which is used to extract the association relationship between attributes of different categories; for the DeepCross network, its initialization input is usually denoted as u0, and the g-th layer can be recursively represented as Among them, w g and b g are the parameters of the neural network.

[0067] The temporal feature modeling structure combines Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) to extract the temporal features of historical advertising data on the target advertising platform. First, CNN is used to extract shallow features, and then the extracted features are used as the input of RNN to obtain temporal features. *,M ={u lyt |1≤l≤L,1≤y≤Y,1≤t≤M} represents the historical data of M days in all known years. This part is used to extract time series features and period features. *-1,M ={u lyt |1≤l≤L,1≤y≤Y-1,M≤t≤T} represents historical data for the same period and is used to extract long-term features.

[0068] The composite task model has two main tasks: one is to reconstruct historical data using autoencoders, and the other is to estimate the advertising exposure inventory using tensor decomposition. When training the deep spatiotemporal tensor decomposition model, the loss function of the above two tasks can be minimized by formula (1):

[0069] Loss total =ηLoss AE +(1-η)Loss pred (1)

[0070] Typically, η can be set to less than 0.5, as the primary task of the deep spatiotemporal tensor decomposition model is inventory prediction. The design of a composite task model has the following advantages: minimizing information loss; sharing CNN and RNN layers to reduce model complexity; and improving the accuracy of exposure inventory estimates.

[0071] It should be understood that in actual applications, in addition to using the above-mentioned deep spatiotemporal tensor decomposition model to determine the candidate exposure inventory based on the targeting conditions of the target advertisement, the server can also determine the candidate exposure inventory through other methods. This application does not impose any limitations on the method for determining the candidate exposure inventory.

[0072] Step 302: For the candidate exposure opportunities in the candidate exposure inventory, determine the winning rate of the target advertisement in the candidate exposure opportunities through an exposure winning rate prediction model according to the bid information and advertisement features of the target advertisement and the opportunity features of the candidate exposure opportunities.

[0073] After the server determines the candidate exposure inventory based on the target ad's targeting criteria, it can determine the probability of the target ad being exposed through the candidate exposure opportunities within the candidate exposure inventory, i.e., determine the target ad's win rate for the candidate exposure opportunities. Specifically, the server can input the target ad's bid information and ad features, as well as the opportunity features of the candidate exposure opportunities, into a pre-trained exposure win rate prediction model. The exposure win rate prediction model analyzes and processes the input content and accordingly outputs the target ad's win rate for the candidate exposure opportunities.

[0074] It should be noted that the exposure win rate prediction model is trained based on the bid information and advertising features of multiple candidate ads in historical exposure opportunities, the opportunity features of historical exposure opportunities, and the exposure results of historical exposure opportunities; that is, the bid information and advertising features corresponding to the candidate ads in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the candidate ads in the historical exposure opportunities can be used to form training samples, and the exposure win rate prediction model is trained using a training sample set composed of such training samples. In the process of training the exposure win rate prediction model, the exposure win rate prediction model can learn from the training samples which ads are more likely to win in the competition for exposure opportunities, that is, the exposure win rate prediction model can learn the ability to comprehensively consider the competition between ads during the training process. The following will introduce in detail the training method of the exposure win rate prediction model through another method embodiment.

[0075] It should be noted that the bid information of the target advertisement is the bid amount set by the user who places the advertisement. For example, the bid information of the target advertisement can be the amount paid by the user who places the advertisement each time the target advertisement is clicked by a user. The advertising features of the target advertisement can be the features of the target advertisement itself, such as the size of the target advertisement's material, the attractiveness of the material, etc. This application does not make any specific limitations on the advertising features. The opportunity features of the candidate exposure opportunity may include the viewing object features corresponding to the candidate exposure opportunity, and the exposure condition features corresponding to the candidate exposure opportunity, wherein the viewing object features may include the gender, age, and region of the viewing object, etc., and the exposure condition features may include the exposure website, exposure position, exposure time, etc. Of course, the opportunity features of the candidate exposure opportunity may also include other types of features related to the candidate exposure opportunity, and this application does not make any specific limitations on the opportunity features.

[0076] In specific implementation, the server can determine the winning rate of the target advertisement in the candidate exposure opportunity through the exposure winning rate prediction model according to the bid information and advertisement features of the target advertisement, the viewing object features corresponding to the candidate exposure opportunity, and the exposure condition features corresponding to the candidate exposure opportunity. For example, the server can input the bid information of the target advertisement, the features of the target advertisement itself (such as the size of the advertisement material, the attractiveness of the advertisement material, etc.), the viewing object features corresponding to the candidate exposure opportunity (such as the gender, age, address, etc. of the viewing object), and the exposure condition features corresponding to the candidate exposure opportunity (such as the exposure website, exposure position, exposure time, etc. of the candidate exposure opportunity) into the exposure winning rate prediction model. The exposure winning rate prediction model analyzes and processes the input content and accordingly outputs the winning rate of the target advertisement in the candidate exposure opportunity.

[0077] As an example, the above-mentioned exposure win rate prediction model may include a feature mapping (Embedding) layer, a factorization machine (FM) layer, and a deep neural network (DNN) layer. When determining the win rate of a target advertisement in a candidate exposure opportunity through the exposure win rate prediction model, the dense feature vectors corresponding to the bid information and advertising features, viewing object features, and exposure condition features of the target advertisement can be determined respectively through the Embedding layer; then, the dense feature vectors corresponding to the bid information and advertising features, viewing object features, and exposure condition features of the target advertisement are processed respectively through the FM layer and the DNN layer to obtain the feature scalars corresponding to the bid information and advertising features, viewing object features, and exposure condition features of the target advertisement; then, the feature scalars corresponding to the bid information and advertising features, viewing object features, and exposure condition features of the target advertisement are added together to obtain a target scalar; finally, the win rate of the target advertisement in the candidate exposure opportunity is determined based on the target scalar through the Sigmoid function.

[0078] Specifically, when the exposure win rate prediction model determines the win rate of the target ad among the candidate exposure opportunities, it can first map the bid information and ad features of the target ad into corresponding dense feature vectors through the first feature mapping module in the Embedding layer, map the viewing object features corresponding to the candidate exposure opportunities into corresponding dense feature vectors through the second feature mapping module in the Embedding layer, and map the exposure condition features corresponding to the candidate exposure opportunities into corresponding dense feature vectors through the third feature mapping module in the Embedding layer. Then, the dense feature vectors corresponding to the bid information and ad features of the target ad are processed through the FM layer and the DNN layer to obtain feature scalars corresponding to the bid information and ad features of the target ad; the dense feature vectors corresponding to the viewing object features are processed through the FM layer and the DNN layer to obtain feature scalars corresponding to the viewing object features; and the dense feature vectors corresponding to the exposure condition features are processed through the FM layer and the DNN layer to obtain feature scalars corresponding to the exposure condition features. Then, the target ad's bid information is added to the feature scalars corresponding to the ad features, the feature scalars corresponding to the viewing object features, and the dense feature vector corresponding to the exposure condition features to obtain the target scalar. Finally, this target scalar is mapped to the interval [0, 1] using the Sigmoid function to obtain the target ad's winning rate among the candidate exposure opportunities.

[0079] It should be understood that in actual applications, other model structures can be set as exposure win rate prediction models according to actual needs. This application does not impose any restrictions on the model structure of the exposure win rate prediction model, nor does it impose any restrictions on the specific working principle of the exposure win rate prediction model.

[0080] In one possible implementation, the server may determine the target ad's winning rate among each candidate exposure opportunity included in the candidate exposure inventory. Specifically, the server may use an exposure winning rate prediction model to determine the target ad's winning rate for each candidate exposure opportunity in the candidate exposure inventory based on the target ad's bid information and ad features, as well as the opportunity features of the candidate exposure opportunity. In this way, the server obtains the winning rate of the target ad among all candidate exposure opportunities included in the candidate exposure inventory.

[0081] In another possible implementation, in order to reduce the required processing resources and further improve the accuracy of the exposure estimation results, the method provided in the embodiment of the present application can first screen out candidate exposure opportunities that are more likely to expose the target advertisement from the candidate exposure inventory as target exposure opportunities, and then determine the winning rate of the target advertisement in the target exposure opportunities through the exposure winning rate estimation model.

[0082] As an example, the server can filter out target exposure opportunities from the candidate exposure inventory in the following manner: for each candidate exposure opportunity in the candidate exposure inventory, determine the historical exposure opportunity corresponding to the candidate exposure opportunity, and obtain the refined advertising queue corresponding to the historical exposure opportunity as the refined advertising queue corresponding to the candidate exposure opportunity; then, for each candidate exposure opportunity in the candidate exposure inventory, determine whether the target advertisement meets the reference revenue condition of the refined advertising queue corresponding to the candidate exposure opportunity. If so, the candidate exposure opportunity can be determined as the target exposure opportunity.

[0083] Specifically, the server can determine the historical exposure opportunity corresponding to each candidate exposure opportunity in the candidate exposure inventory. For example, assuming that the candidate exposure opportunity is the exposure through the opening screen interface of website A at 16:30 today, the historical exposure opportunity corresponding to the candidate exposure opportunity can be the exposure through the opening screen interface of website A at 16:30 yesterday. Since the database used to store advertising exposure records will store the refined advertising queue corresponding to each historical exposure opportunity, the server can retrieve the refined advertising queue of the historical exposure opportunity corresponding to the candidate exposure opportunity from the database, and use the refined advertising queue as the refined advertising queue corresponding to the candidate exposure opportunity; it should be understood that the refined advertising queue refers to a queue composed of candidate advertisements retained after refined processing, which usually includes a specific number of candidate advertisements, for example, it can include 200 candidate advertisements.

[0084] Furthermore, the server can determine, for each candidate exposure opportunity in the candidate exposure inventory, whether the target advertisement meets the reference revenue condition of the precision advertising queue corresponding to the candidate exposure opportunity; if the target advertisement meets the reference revenue condition of the precision advertising queue, it is considered that the target advertisement is more likely to be exposed through the candidate exposure opportunity, and the candidate exposure opportunity is determined as the target exposure opportunity; conversely, if the target advertisement does not meet the reference revenue condition of the precision advertising queue, it is considered that the target advertisement is unlikely to be exposed through the candidate exposure opportunity, and in order to save processing resources, no subsequent processing may be performed on the candidate exposure opportunity.

[0085] As an example, the server can determine whether the target advertisement meets the reference revenue condition of the precision advertisement queue corresponding to the candidate exposure opportunity in the following manner: when the target advertisement is included in the precision advertisement queue corresponding to the candidate exposure opportunity, determine that the target advertisement meets the reference revenue condition of the precision advertisement queue, and determine the candidate exposure opportunity as the target exposure opportunity; when the precision advertisement queue corresponding to the candidate exposure opportunity does not include the target advertisement, determine the estimated revenue per thousand impressions (effective cost per mile, eCPM) of the target advertisement based on the bid information, estimated click-through rate and estimated conversion rate of the target advertisement, and determine whether the estimated eCPM of the target advertisement is greater than the minimum eCPM of the precision advertisement queue corresponding to the candidate exposure opportunity. If so, determine that the target advertisement meets the reference revenue condition of the precision advertisement queue, and determine the candidate exposure opportunity as the target exposure opportunity.

[0086] Specifically, if the targeted advertisement is included in the refined advertisement queue corresponding to the candidate exposure opportunity, it can be directly considered that the targeted advertisement meets the reference revenue condition of the refined advertisement queue, and the candidate exposure opportunity can be directly used as the target exposure opportunity.

[0087] If the target ad is not included in the precise ranking ad queue corresponding to the candidate exposure opportunity, the lowest eCPM of the precise ranking ad queue can be obtained and recorded as eCPM_0. The lowest eCPM corresponding to the precise ranking ad queue is actually the eCPM corresponding to the candidate ad with the lowest eCPM in the precise ranking ad queue. Furthermore, the estimated eCPM_ad of the target ad is calculated using formula (2) based on the bid information, estimated click-through rate (pCTR), and estimated conversion rate (pCVR) of the target ad:

[0088] eCPM_ad=1000*bid*pCTR*pCVR (2)

[0089] If eCPM_ad is greater than eCPM_0, it can be considered that the target advertisement meets the reference revenue conditions of the precise advertising queue, and the candidate exposure opportunity can be used as the target exposure opportunity; conversely, if eCPM_ad is less than or equal to eCPM_0, it can be considered that the target advertisement does not meet the reference revenue conditions of the precise advertising queue, and the candidate exposure opportunity can be omitted from subsequent processing.

[0090] After the server has screened target impressions from the candidate exposure inventory using the aforementioned method, it can determine the target ad's chance of winning for each selected target impression using the exposure win rate prediction model based on the target ad's bid information and ad features, as well as the opportunity characteristics of the target impression opportunity. The server does not need to perform the aforementioned processing for other candidate exposure opportunities in the candidate exposure inventory, thereby reducing processing resources while ensuring the accuracy of the exposure estimation results.

[0091] Step 303: Determine an exposure estimation result corresponding to the target advertisement based on the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0092] After determining the winning rate of the target advertisement among the candidate exposure opportunities included in the candidate exposure inventory, the server may determine the exposure estimation result of the target advertisement in the future period based on the winning rate of the target advertisement among the candidate exposure opportunities included in the candidate exposure inventory.

[0093] In one possible implementation, if the server determines the winning rate of the target advertisement in each candidate exposure opportunity included in the candidate exposure inventory through step 302, then when the server determines the exposure estimation result corresponding to the target advertisement, it can calculate the sum of the winning rates of the target advertisement in each candidate exposure opportunity, and then use the sum as the exposure estimation result corresponding to the target advertisement.

[0094] In another possible implementation, if the server selects target exposure opportunities from the candidate exposure inventory and determines the winning rate of the target advertisement in each target exposure opportunity through 302, then when the server determines the exposure estimation result corresponding to the target advertisement, it can calculate the sum of the winning rates of the target advertisement in each target exposure opportunity, and then use the sum as the exposure estimation result corresponding to the target advertisement.

[0095] It should be understood that in actual applications, in addition to calculating the sum of the winning rates of the target advertisement in multiple candidate exposure opportunities as the exposure estimation result, the server can also adopt other methods to determine the exposure estimation result corresponding to the target advertisement based on the winning rate of the target advertisement in multiple candidate exposure opportunities. For example, the average value of the winning rate of the target advertisement in multiple candidate exposure opportunities can be calculated as the exposure estimation result. This application does not impose any limitation on the method of determining the exposure estimation result.

[0096] Optionally, in one possible application scenario, the server may receive a bid modification request initiated by an advertising delivery client for the target ad, the bid modification request including the target ad's updated bid information. The server may then determine, for candidate exposure opportunities in the candidate exposure inventory, the target ad's updated bid information and ad features, as well as the opportunity features of the candidate exposure opportunities, using an exposure win rate prediction model. Furthermore, based on the target ad's updated bid information and ad features in the candidate exposure opportunities included in the candidate exposure inventory, the server may determine an updated exposure prediction result corresponding to the target ad. The server may then send the updated exposure prediction result to the advertising delivery client.

[0097] Figure 4 This is a schematic diagram of an exemplary bid information modification interface on an advertising client. Figure 4 As shown, a user can enter the updated bid information for a target ad in the bid information input field on the bid information modification interface. Upon detecting that the user has triggered and clicked the OK control, the ad delivery client generates a bid modification request and sends it to the server. This bid modification request includes the updated bid information for the target ad entered by the user. Upon receiving this bid modification request, the server executes steps 302 and 303 based on the updated bid information of the target ad to determine an updated exposure estimate for the target ad, such as an updated exposure estimate for the target ad within the next hour. The server can then feed this updated exposure estimate back to the ad delivery client, which then displays it.

[0098] It should be understood that in actual applications, the server may also determine other relevant reference information based on the updated bid information and / or the updated exposure estimate results, such as the bid adjustment ratio, the exposure estimate result adjustment ratio, and bid suggestion information, and feedback the determined reference information to the advertising client. This application does not impose any limitations on the reference information determined by the server.

[0099] Optionally, in a possible application scenario, the server may send advertising delivery reference information to the advertising delivery client, where the advertising delivery reference information includes prompt information corresponding to the potential advertisement, where the prompt information is used to indicate a bid adjustment reference strategy corresponding to the potential advertisement.

[0100] Specifically, the server can determine whether each advertisement delivered through a certain advertising delivery client is a potential advertisement. Potential advertisements refer to advertisements that can significantly improve exposure by adjusting the bid information to a certain extent. For example, an advertisement that can improve exposure by more than 20% by increasing the bid information by 10% is considered a potential advertisement. The server determines the corresponding advertising delivery reference information for each advertisement delivered by a certain advertising delivery client. For example, the advertising delivery reference information may include the current status of the advertisement, bid information, daily budget, actual exposure, and actual click volume. In addition, the advertising delivery reference information may also include prompt information corresponding to the potential advertisement, that is, the server can notify the advertising delivery client of the potential advertisements included in the advertisements it delivers through the advertising delivery reference information, and configure corresponding prompt information for the potential advertisements. The prompt information is used to indicate the bid adjustment reference strategy corresponding to the potential advertisement. For example, increasing the bid for a potential advertisement by 11% can increase exposure by 26%.

[0101] Figure 5 This is a schematic diagram of an exemplary advertising placement reference information display interface on an advertising placement client. Figure 5 As shown, after the advertising delivery client receives the advertising delivery reference information sent by the server, it can display the corresponding advertising delivery reference information for each advertisement it delivers; for potential advertisements, the advertising delivery client can mark a potential identifier next to its bid information to prompt the user that the advertisement is a potential advertisement. When the user triggers a click operation on the potential identifier or moves the cursor to the potential identifier, the bid adjustment strategy corresponding to the potential advertisement can be displayed accordingly.

[0102] Optionally, in a possible application scenario, the server may set a plurality of continuous candidate bid information corresponding to the target advertisement. For each candidate bid information, the server determines the winning rate of the target advertisement in the candidate exposure opportunity when using the candidate bid information based on the candidate bid information and the advertisement characteristics, as well as the opportunity characteristics of the candidate exposure opportunity through the exposure winning rate prediction model; and, for each candidate bid information, determines the exposure estimation result of the target advertisement when using the candidate bid information based on the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory when using the candidate bid information. Accordingly, the server can generate an exposure change trend based on the exposure estimation results when the target advertisement uses various candidate bid information. The exposure change trend can characterize the changing trend of the exposure estimation results corresponding to the target advertisement under multiple candidate bid information; and then, send the exposure change trend to the advertising delivery client corresponding to the target advertisement.

[0103] Specifically, to provide better bidding recommendations to users placing target ads, the server may set multiple consecutive candidate bid information based on the target ad's current bid information, and execute steps 302 and 303 based on each set of candidate bid information to determine the exposure estimate results for the target ad when using each candidate bid information. Furthermore, based on the exposure estimate results for the target ad when using each candidate bid information, the server may draw a curve chart reflecting the relationship between the bid information and the exposure estimate results as an exposure change trend, and send the exposure change trend to the ad delivery client that placed the target ad, so that the user placing the target ad can better determine the adjustment strategy for the target ad's bid information based on the exposure change trend. Figure 6 This is a schematic diagram of an exemplary interface displaying exposure change trends on an advertising delivery client.

[0104] It should be understood that in actual applications, the server can also determine the estimated click-through rate and estimated conversion rate under each candidate bidding information based on the candidate bidding information; and, for each candidate bidding information, the corresponding bid change range, estimated exposure change range, estimated click-through rate change range and estimated conversion rate change range can be determined respectively, so as to facilitate users to compare with the current situation of the target advertisement.

[0105] The above-mentioned advertising exposure estimation method innovatively starts from the dimension of exposure opportunities and determines the exposure estimation result corresponding to the target advertisement based on the winning rate of the target advertisement in the candidate exposure opportunities estimated by the exposure winning rate estimation model. Since the exposure winning rate estimation model is trained based on the bidding information and advertising characteristics of multiple candidate advertisements in historical exposure opportunities, the opportunity characteristics of historical exposure opportunities, and the exposure results of historical exposure opportunities, the exposure winning rate estimation model can learn which advertisements are more likely to win in the competition for exposure opportunities based on the training sample data during the training process; accordingly, when estimating the winning rate of the target advertisement in the candidate exposure opportunities, the exposure winning rate estimation model will also comprehensively consider the competition between the target advertisement and other advertisements; furthermore, the exposure estimation result corresponding to the target advertisement is determined based on the winning rate estimated by the exposure winning rate estimation model, which can ensure the accuracy of the determined exposure estimation result.

[0106] Whether the advertising exposure estimation method provided in the embodiment of the present application can accurately estimate advertising exposure depends to a large extent on the model performance of the above-mentioned exposure win rate estimation model. The training method of the exposure win rate estimation model is introduced in detail below through the method embodiment.

[0107] See also Figure 7 , Figure 7 A flow chart of the training method of the exposure win rate prediction model provided in the embodiment of the present application. For the sake of convenience, the following embodiment is still introduced by taking the execution subject of the training method of the exposure win rate prediction model as an example. Figure 7 As shown, the training method of the exposure win rate prediction model includes the following steps:

[0108] Step 701: Obtain a training sample set; the training sample set includes multiple training samples; the training samples include bidding information and advertising features of candidate advertisements in the historical exposure opportunities, opportunity features of the historical exposure opportunities, and exposure results of the candidate advertisements in the historical exposure opportunities.

[0109] Before the server trains the exposure win rate prediction model, it needs to first obtain a training sample set, which usually includes a large number of training samples, where each training sample includes the bidding information and advertising features of the candidate advertisement in the historical exposure opportunity, the opportunity features of the historical exposure opportunity, and the exposure result of the candidate advertisement in the historical exposure opportunity.

[0110] It should be noted that a historical exposure opportunity may be an exposure opportunity that occurred within a specific historical time period, for example, an exposure opportunity that occurred within an hour before the current time point. A candidate ad in a historical exposure opportunity may specifically be an ad in the optimized ad queue corresponding to the historical exposure opportunity. The bid information of a candidate ad is the bid information used by the candidate ad in the historical exposure opportunity. The ad features of a candidate ad may be features of the candidate ad itself, such as the size of the ad creative and the attractiveness of the ad creative. This application does not specifically define these ad features. The opportunity features of a historical exposure opportunity may include the corresponding viewing audience features of the historical exposure opportunity and the corresponding exposure condition features of the historical exposure opportunity. Viewer features may include the gender, age, and region of the viewing audience. Exposure condition features may include the exposure website, exposure placement, and exposure time. This application does not specifically define these opportunity features. A candidate ad's exposure results in a historical exposure opportunity include two types: exposure in the historical exposure opportunity, in which case the exposure result may be recorded as 1; and no exposure in the historical exposure opportunity, in which case the exposure result may be recorded as 0.

[0111] In one possible implementation, the server may first determine a period of preset duration before the model training moment as the target historical period; then, for each historical exposure opportunity occurring within the target historical period, obtain the refined advertising queue corresponding to the historical exposure opportunity; and, for each candidate advertisement in the refined advertising queue corresponding to each historical exposure opportunity, utilize the bid information and advertising features of the candidate advertisement, the opportunity features of the historical exposure opportunity, and the exposure results of the candidate advertisement in the historical exposure opportunity to form a training sample corresponding to the candidate advertisement in the historical exposure opportunity; and further, utilize the training samples corresponding to each candidate advertisement in each historical exposure opportunity within the target historical period to form a training sample set.

[0112] For example, the server can use the hour before the model training moment as the target historical period, and construct a training sample set based on all historical exposure opportunities that occurred during the target historical period. Assuming that the fine-ranked advertising queue corresponding to each historical exposure opportunity includes 200 candidate ads, the server can construct 200 training samples based on the historical exposure opportunity, of which 199 training samples include a historical exposure result of 0 (corresponding to candidate ads that were not exposed through the historical exposure opportunity), and 1 training sample includes a historical exposure result of 1 (corresponding to candidate ads that were exposed through the historical exposure opportunity); if N exposures occurred during the target historical period, the constructed training sample set should include N*200 training samples.

[0113] It should be understood that in actual applications, the server can set the above-mentioned target historical period according to actual needs, and this application does not impose any limitation on the target historical period.

[0114] Step 702: Training the exposure win rate prediction model based on the training samples in the training sample set.

[0115] After the server obtains the training sample set, it can use the training samples in the training sample set to train the exposure win rate prediction model. Specifically, the server can input the pricing information and advertising features of the candidate advertisements included in the training samples, as well as the opportunity features of historical exposure opportunities, into the exposure win rate prediction model to be trained. After the exposure win rate prediction model analyzes and processes the input content, it will accordingly output the predicted win rate of the candidate advertisement in the historical exposure opportunity. Furthermore, the server can construct a loss function based on the historical exposure results included in the training samples and the predicted win rate, and adjust the model parameters of the exposure win rate prediction model based on the loss function through the gradient descent algorithm until the exposure win rate prediction model reaches the training end condition.

[0116] It should be understood that in actual applications, the server may consider that the exposure win rate prediction model has met the training end condition when the model performance of the trained exposure win rate prediction model meets the preset conditions, and may confirm that the training of the exposure win rate prediction model is complete. Alternatively, the server may consider that the exposure win rate prediction model has met the training end condition when the number of iterative training of the exposure win rate prediction model reaches a preset number, and may confirm that the training of the exposure win rate prediction model is complete.

[0117] It should be noted that in actual applications, to ensure that the exposure win rate prediction model can continuously and accurately estimate the ad's win rate among candidate exposure opportunities, the server can periodically train the exposure win rate prediction model. For example, the model can be trained every other day. When periodically training the exposure win rate prediction model, a training sample set can be constructed based on historical exposure opportunities that occurred in the previous cycle.

[0118] The exposure win rate prediction model training method described above utilizes the bidding information and advertising features of multiple candidate advertisements in historical exposure opportunities, the opportunity features of historical exposure opportunities, and the exposure results of historical exposure opportunities to train the exposure win rate prediction model. This allows the exposure win rate prediction model to learn, based on training samples, which advertisement is more likely to win in the competition for exposure opportunities during the training process. Accordingly, when the exposure win rate prediction model is subsequently used to estimate the winning rate of advertisements in exposure opportunities, the competition between advertisements can be comprehensively considered, thereby ensuring the accuracy of the determined exposure prediction results.

[0119] In order to further understand the advertising exposure estimation method provided in the embodiment of the present application, Figure 8 The schematic diagram of the implementation principle of the advertising exposure estimation method shown in the figure provides an overall exemplary introduction to the advertising exposure estimation method provided in the embodiment of the present application.

[0120] like Figure 8 As shown, the implementation process of the advertising exposure estimation method provided in the embodiment of the present application is mainly divided into two parts, namely determining the target exposure opportunity and estimating the advertising exposure result.

[0121] When determining target exposure opportunities, we can first use the deep spatiotemporal tensor decomposition model to estimate the candidate exposure inventory suitable for the target ad based on the target ad's targeting conditions (such as user attributes: gender, age, address, etc., traffic attributes: website, position, time, etc.). The candidate exposure inventory can be understood as a set of candidate exposure opportunities that the target ad can participate in. The operation here is similar to Figure 1 The recall processing stage of the ad exposure process is shown. Here, each candidate exposure opportunity in the candidate exposure inventory corresponds to a refined ad queue, which is actually a refined ad queue of the historical mapped exposure opportunities corresponding to the candidate exposure opportunity.

[0122] For each candidate exposure opportunity in the candidate exposure inventory, if the target ad is included in the refined ad queue corresponding to the candidate exposure opportunity, the candidate exposure opportunity can be directly used as the target exposure opportunity. If the target ad is not included in the refined ad queue corresponding to the candidate exposure opportunity, the server can obtain the minimum eCPM_0 of the ads in the refined ad queue corresponding to the candidate exposure opportunity; at this time, the server can calculate the estimated eCPM_ad of the target ad = 1000*bid*pCTR*pCVR based on the bid, estimated click-through rate, pCTR, and estimated conversion rate, pCVR of the target ad; if eCPM_ad is greater than eCPM_0, it can be considered that the target ad is likely to enter the refined ad queue corresponding to the candidate exposure opportunity, and the candidate exposure opportunity can be used as the target exposure opportunity; conversely, if eCPM_ad is not greater than eCPM_0, it can be considered that the target ad is unlikely to enter the refined ad queue corresponding to the candidate exposure opportunity, and the candidate exposure opportunity will no longer be processed. The operation here is similar to Figure 1 The coarse processing stage of the ad exposure process is shown.

[0123] After the above processing, if the target ad is included in the refined ad queue corresponding to the target exposure opportunity, the target exposure opportunity is directly input into the next level of processing. If the target ad is not included in the refined ad queue corresponding to the target exposure opportunity, the target ad is inserted into the refined ad queue corresponding to the target exposure opportunity based on its estimated eCPM_ad. The ad with the lowest eCPM in the refined ad queue corresponding to the target exposure opportunity is removed to ensure that the maximum queue of the refined ad queue is 200.

[0124] When estimating advertising exposure results, the exposure win rate estimation model can be used for each previously determined target exposure opportunity to determine the win rate of the target advertisement in the target exposure opportunity based on the bid information and advertising characteristics of the target advertisement (such as the size of the advertising creative, the attractiveness of the creative, etc.), the viewing audience characteristics of the target exposure opportunity (such as gender, age, address, etc.), and the exposure condition characteristics of the target exposure opportunity (such as the website, position, time, etc.); then, the win rates of the target advertisement in each target exposure opportunity are added up to obtain the estimated exposure result corresponding to the target advertisement. Figure 9 The figure shows a schematic diagram of the implementation principle of the process of estimating advertising exposure results.

[0125] It should be noted that during the training phase of the exposure win rate prediction model, a training sample set can be constructed using all historical exposure opportunities that occurred in the previous hour. For each historical exposure opportunity, only one candidate ad can be exposed through that historical exposure opportunity. Therefore, based on the refined ad queue corresponding to that historical exposure opportunity, 200 training samples can be constructed, of which 199 training samples include a historical exposure result of 0, and 1 training sample includes a historical exposure result of 1. If a total of N historical exposure opportunities occurred in the previous hour, the training sample set should include N*200 training samples.

[0126] Each training sample primarily includes the following data: the candidate ad's bid information and ad features (such as ad creative size and appeal), the audience characteristics of the historical exposure opportunity (such as gender, age, and address), the exposure condition characteristics of the historical exposure opportunity (such as website, placement, and time), and the candidate ad's exposure results in that historical exposure opportunity. During model processing, the bid information and ad features, the audience characteristics, and the exposure condition characteristics pass through the Embedding layer, then the FM layer and the DNN layer, respectively, to output a feature scalar. The feature scalars are then summed and applied to the Sigmoid function to calculate the win rate.

[0127] Compared with the exposure estimation method based on the advertising dimension in the related art, the advertising exposure estimation method provided in the embodiment of the present application has an accuracy rate increased from 70% to 80% when the mean square error (MSE) is less than 0.2, achieving a relatively good exposure estimation effect.

[0128] With respect to the advertising exposure estimation method described above, the present application also provides a corresponding advertising exposure estimation device to enable the above-mentioned advertising exposure estimation method to be applied and implemented in practice.

[0129] See also Figure 10 , Figure 10 It is the above Figure 3 The structure diagram of an advertisement exposure estimation device 1000 corresponding to the advertisement exposure estimation method shown in FIG. Figure 10 As shown, the advertising exposure estimation device 1000 includes:

[0130] The exposure inventory determination module 1001 is configured to determine a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions;

[0131] A win rate estimation module 1002 is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, the win rate of the target advertisement in the candidate exposure opportunity using an exposure win rate estimation model based on the bid information and ad features of the target advertisement, as well as the opportunity features of the candidate exposure opportunity. The exposure win rate estimation model is trained based on the bid information and ad features of each of multiple candidate advertisements in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target advertisement being exposed through the candidate exposure opportunity.

[0132] The exposure result determination module 1003 is configured to determine an exposure estimation result corresponding to the target advertisement according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0133] Optional, in Figure 10 Based on the advertising exposure estimation device shown, see Figure 11 , Figure 11 This is a structural diagram of another advertising exposure estimation device 1100 provided in an embodiment of the present application. Figure 11 As shown, the advertising exposure estimation device also includes:

[0134] The refined ranking queue determination module 1101 is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, a historically mapped exposure opportunity corresponding to the candidate exposure opportunity; and obtain a refined ranking advertisement queue corresponding to the historically mapped exposure opportunity as the refined ranking advertisement queue corresponding to the candidate exposure opportunity.

[0135] The exposure opportunity screening module 1102 is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, whether the target advertisement meets the reference revenue condition of the refined ranking advertisement queue corresponding to the candidate exposure opportunity, and if so, determine the candidate exposure opportunity as the target exposure opportunity;

[0136] The winning rate estimation module 1002 is specifically used to:

[0137] For each target exposure opportunity, determining the winning rate of the target advertisement in the target exposure opportunity by using the exposure winning rate prediction model according to the bid information and advertisement characteristics of the target advertisement and the opportunity characteristics of the target exposure opportunity;

[0138] The exposure result determination module 1003 is specifically used to:

[0139] An exposure estimation result corresponding to the target advertisement is determined according to the winning rate of the target advertisement in each of the target exposure opportunities.

[0140] Optional, in Figure 11 Based on the advertising exposure estimation device shown, the exposure opportunity screening module 1102 is specifically used to:

[0141] If the target advertisement is included in the refined advertisement queue corresponding to the candidate exposure opportunity, determining that the target advertisement meets the reference revenue condition, and determining the candidate exposure opportunity as the target exposure opportunity;

[0142] When the target ad is not included in the precise ad queue corresponding to the candidate exposure opportunity, the estimated revenue per thousand impressions (eCPM) of the target ad is determined based on the bid information, estimated click-through rate, and estimated conversion rate of the target ad; and a judgment is made as to whether the estimated eCPM of the target ad is greater than the lowest eCPM of the precise ad queue corresponding to the candidate exposure opportunity. If so, it is determined that the target ad meets the reference revenue condition, and the candidate exposure opportunity is determined as the target exposure opportunity.

[0143] Optional, in Figure 10 Based on the advertising exposure estimation device shown, the winning rate estimation module 1002 is specifically used to:

[0144] The exposure winning rate prediction model is used to determine the winning rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertisement features of the target advertisement, the viewing object features corresponding to the candidate exposure opportunities, and the exposure condition features corresponding to the candidate exposure opportunities.

[0145] Optional, in Figure 10 Based on the advertising exposure estimation device shown, the exposure win rate estimation model includes an Embedding layer, a FM layer, and a DNN layer; the win rate estimation module 1002 is specifically used to:

[0146] Determining, through the Embedding layer, dense feature vectors corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features;

[0147] The FM layer and the DNN layer process the dense feature vectors corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features, respectively, to obtain feature scalars corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features, respectively;

[0148] Adding the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features to the corresponding feature scalars to obtain a target scalar;

[0149] The winning rate of the target advertisement in the candidate exposure opportunities is determined according to the target scalar through the Sigmoid function.

[0150] Optional, in Figure 10 Based on the advertising exposure estimation device shown, the exposure inventory determination module 1001 is specifically used to:

[0151] The candidate exposure inventory is predicted based on the targeting conditions of the target advertisement through a deep spatiotemporal tensor decomposition model; the deep spatiotemporal tensor decomposition model is used to predict the exposure inventory of the target advertisement within a preset time period through tensor decomposition based on the attribute combination characteristics of the targeting conditions and the temporal characteristics of historical advertisement delivery data on the target advertisement delivery platform; the target advertisement delivery platform is an advertisement delivery platform used to deliver the target advertisement.

[0152] Optional, in Figure 10 Based on the advertising exposure estimation device shown, see Figure 12 , Figure 12 This is a structural diagram of another advertising exposure estimation device 1200 provided in an embodiment of the present application. Figure 12 As shown, the advertising exposure estimation device also includes:

[0153] Model training module 1201 is used to obtain a training sample set; the training sample set includes multiple training samples; the training samples include bidding information and advertising features of candidate advertisements in historical exposure opportunities, opportunity features of the historical exposure opportunities, and exposure results of the candidate advertisements in the historical exposure opportunities; based on the training samples in the training sample set, the exposure win rate prediction model is trained.

[0154] Optional, in Figure 12 Based on the advertising exposure estimation device shown, the model training module 1201 is specifically used to:

[0155] Determine a preset period of time before the model training moment as the target historical period;

[0156] For each of the historical exposure opportunities occurring within the target historical period, obtaining a refined advertising queue corresponding to the historical exposure opportunity;

[0157] For each candidate advertisement in the refined advertisement queue corresponding to each historical exposure opportunity, a training sample corresponding to the candidate advertisement in the historical exposure opportunity is formed using the bid information and advertisement features of the candidate advertisement, the opportunity features of the historical exposure opportunity, and the exposure results of the candidate advertisement in the historical exposure opportunity;

[0158] The training sample set is formed by utilizing the training samples corresponding to each of the candidate advertisements in each of the historical exposure opportunities within the target historical period.

[0159] Optional, in Figure 10 Based on the advertising exposure estimation device shown, see Figure 13 , Figure 13 This is a structural diagram of another advertising exposure estimation device 1300 provided in an embodiment of the present application. Figure 13 As shown, the advertising exposure estimation device also includes:

[0160] The bid modification request response module 1301 is used to receive a bid modification request initiated by an advertising delivery client for the target advertisement; the bid modification request includes the updated bid information of the target advertisement; for the candidate exposure opportunities in the candidate exposure inventory, the updated winning rate of the target advertisement in the candidate exposure opportunities is determined through the exposure winning rate prediction model according to the updated bid information and advertisement characteristics of the target advertisement, as well as the opportunity characteristics of the candidate exposure opportunities; and the updated exposure prediction result corresponding to the target advertisement is determined according to the updated winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory; and the updated exposure prediction result is sent to the advertising delivery client.

[0161] Optional, in Figure 10 Based on the advertising exposure estimation device shown, see Figure 14 , Figure 14 This is a structural diagram of another advertising exposure estimation device 1400 provided in an embodiment of the present application. Figure 14 As shown, the advertising exposure estimation device also includes:

[0162] The reference information feedback module 1401 is used to send advertisement delivery reference information to the advertisement delivery client; the advertisement delivery reference information includes prompt information corresponding to potential advertisements, and the prompt information is used to indicate a bid adjustment reference strategy corresponding to the potential advertisements.

[0163] Optional, in Figure 10 Based on the advertising exposure estimation device shown, the target advertisement corresponds to a plurality of continuous candidate bidding information; see Figure 15 , Figure 15 This is a schematic diagram of the structure of another advertising exposure estimation device 1500 provided in an embodiment of the present application. In the advertising exposure estimation device 1500:

[0164] The winning rate estimation module 1002 is specifically used for:

[0165] For each type of candidate bid information, determine the winning rate of the target advertisement in the candidate exposure opportunity when the candidate bid information is used, using an exposure winning rate prediction model based on the candidate bid information and advertisement characteristics, as well as the opportunity characteristics of the candidate exposure opportunity;

[0166] The exposure result determination module 1003 is specifically configured to:

[0167] For each type of candidate bid information, determining an estimated exposure result of the target advertisement when using the candidate bid information based on a winning rate of the target advertisement among the candidate exposure opportunities included in the candidate exposure inventory when using the candidate bid information;

[0168] The advertising exposure estimation device further includes:

[0169] The exposure change trend feedback module 1501 is used to generate an exposure change trend based on the exposure estimation results when the target advertisement uses various candidate bidding information; the exposure change trend is used to characterize the change trend of the exposure estimation results corresponding to the target advertisement under the multiple candidate bidding information; and send the exposure change trend to the advertising delivery client corresponding to the target advertisement.

[0170] The above-mentioned advertising exposure prediction device innovatively starts from the dimension of exposure opportunities and determines the exposure prediction result corresponding to the target advertisement based on the winning rate of the target advertisement in the candidate exposure opportunities estimated by the exposure winning rate prediction model. Since the exposure winning rate prediction model is trained based on the bidding information and advertising characteristics of multiple candidate advertisements in historical exposure opportunities, the opportunity characteristics of historical exposure opportunities, and the exposure results of historical exposure opportunities, the exposure winning rate prediction model can learn which advertisements are more likely to win in the competition for exposure opportunities based on the training sample data during the training process; accordingly, when estimating the winning rate of the target advertisement in the candidate exposure opportunities, the exposure winning rate prediction model will also comprehensively consider the competition between the target advertisement and other advertisements; furthermore, the exposure prediction result corresponding to the target advertisement is determined based on the winning rate estimated by the exposure winning rate prediction model, which can ensure the accuracy of the determined exposure prediction result.

[0171] The embodiment of the present application also provides a device for estimating advertising exposure, which may specifically be a terminal device or a server. The terminal device and server provided in the embodiment of the present application will be introduced below from the perspective of hardware implementation.

[0172] See also Figure 16 , Figure 16 This is a schematic diagram of the structure of the terminal device provided in the embodiment of the present application. Figure 16For ease of explanation, only the parts related to the embodiments of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiments of the present application. The terminal can be any terminal device including a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sales (POS), a car computer, etc. For example, the terminal is a computer:

[0173] Figure 16 FIG2 is a block diagram showing a partial structure of a computer related to a terminal provided in an embodiment of the present application. Figure 16 The computer includes components such as a radio frequency (RF) circuit 1610, a memory 1620, an input unit 1630, a display unit 1640, a sensor 1650, an audio circuit 1660, a wireless fidelity (WiFi) module 1670, a processor 1680, and a power supply 1690. Those skilled in the art will appreciate that Figure 16 The computer structure shown in the figure does not constitute a limitation of the computer, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0174] Memory 1620 can be used to store software programs and modules. Processor 1680 executes the various computer functions and data processing by running the software programs and modules stored in memory 1620. Memory 1620 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function). The data storage area may store data generated based on computer use (such as audio data, a phone book, etc.). Memory 1620 may also include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state memory device.

[0175] Processor 1680 is the computer's control center, connecting all computer components using various interfaces and circuits. By running or executing software programs and / or modules stored in memory 1620 and accessing data stored in memory 1620, it performs various computer functions and processes data, thereby providing overall computer monitoring. Optionally, processor 1680 may include one or more processing units. Preferably, processor 1680 may integrate an application processor and a modem processor, with the application processor primarily processing the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 1680.

[0176] In the embodiment of the present application, the processor 1680 included in the terminal further has the following functions:

[0177] Determining a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions;

[0178] For the candidate exposure opportunities in the candidate exposure inventory, an exposure win rate prediction model is used to determine the target ad's win rate in the candidate exposure opportunities based on the bid information and ad features of the target ad, as well as the opportunity features of the candidate exposure opportunities. The exposure win rate prediction model is trained based on the bid information and ad features of multiple candidate ads in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target ad being exposed through the candidate exposure opportunities.

[0179] An exposure estimation result corresponding to the target advertisement is determined according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0180] Optionally, the processor 1680 is further configured to execute steps of any implementation of the advertising exposure estimation method provided in the embodiments of the present application.

[0181] See also Figure 17 , Figure 17A structural diagram of a server 1700 provided for an embodiment of the present application. The server 1700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1722 (for example, one or more processors) and a memory 1732, and one or more storage media 1730 (for example, one or more massive storage devices) for storing application programs 1742 or data 1744. Among them, the memory 1732 and the storage medium 1730 may be temporary storage or permanent storage. The program stored in the storage medium 1730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1722 may be configured to communicate with the storage medium 1730 to execute a series of instruction operations in the storage medium 1730 on the server 1700.

[0182] The server 1700 may also include one or more power supplies 1726, one or more wired or wireless network interfaces 1750, one or more input and output interfaces 1758, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0183] The steps performed by the server in the above embodiment can be based on the Figure 17 The server structure shown.

[0184] The CPU 1722 is configured to execute the following steps:

[0185] Determining a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions;

[0186] For the candidate exposure opportunities in the candidate exposure inventory, an exposure win rate prediction model is used to determine the target ad's win rate in the candidate exposure opportunities based on the bid information and ad features of the target ad, as well as the opportunity features of the candidate exposure opportunities. The exposure win rate prediction model is trained based on the bid information and ad features of multiple candidate ads in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target ad being exposed through the candidate exposure opportunities.

[0187] An exposure estimation result corresponding to the target advertisement is determined according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

[0188] Optionally, the CPU 1722 can also be configured to execute the steps of any of the implementation manners of the advertisement exposure estimation method provided in the embodiments of the present application.

[0189] The embodiments of the present application further provide a computer readable storage medium for storing a computer program, and the computer program is used for executing any of the implementation manners of the advertisement exposure estimation method.

[0190] The embodiments of the present application further provide a computer program product or a computer program, and the computer program product or the computer program includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes any of the implementation manners of the advertisement exposure estimation method.

[0191] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0192] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented by other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0193] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the embodiments of the present application.

[0194] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of software functional unit.

[0195] If the integrated unit 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 is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), random access memory (English full name: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store computer programs.

[0196] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0197] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for estimating advertising exposure, characterized in that: The method comprises: Determining a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions; For the candidate exposure opportunities in the candidate exposure inventory, an exposure win rate prediction model is used to determine the target ad's win rate in the candidate exposure opportunities based on the bid information and ad features of the target ad, as well as the opportunity features of the candidate exposure opportunities. The exposure win rate prediction model is trained based on the bid information and ad features of multiple candidate ads in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target ad being exposed through the candidate exposure opportunities. An exposure estimation result corresponding to the target advertisement is determined according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

2. The method according to claim 1, characterized in that The method further comprises: For each candidate exposure opportunity in the candidate exposure inventory, determine a historically mapped exposure opportunity corresponding to the candidate exposure opportunity; obtain a refined advertising queue corresponding to the historically mapped exposure opportunity as the refined advertising queue corresponding to the candidate exposure opportunity; For each candidate exposure opportunity in the candidate exposure inventory, determine whether the target advertisement meets the reference revenue condition of the refined advertisement queue corresponding to the candidate exposure opportunity; if so, determine the candidate exposure opportunity as the target exposure opportunity; The step of determining the winning rate of the target advertisement in the candidate exposure opportunities in the candidate exposure inventory by using an exposure winning rate estimation model according to the bid information and advertisement features of the target advertisement and the opportunity features of the candidate exposure opportunities includes: For each target exposure opportunity, determining the winning rate of the target advertisement in the target exposure opportunity by using the exposure winning rate prediction model according to the bid information and advertisement characteristics of the target advertisement and the opportunity characteristics of the target exposure opportunity; The determining of the exposure estimation result corresponding to the target advertisement according to the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory includes: An exposure estimation result corresponding to the target advertisement is determined according to the winning rate of the target advertisement in each of the target exposure opportunities.

3. The method according to claim 2, characterized in that The determining whether the target advertisement satisfies the reference revenue condition of the refined advertisement queue corresponding to the candidate exposure opportunity, and if so, determining the candidate exposure opportunity as the target exposure opportunity includes: If the target advertisement is included in the refined advertisement queue corresponding to the candidate exposure opportunity, determining that the target advertisement meets the reference revenue condition, and determining the candidate exposure opportunity as the target exposure opportunity; When the target ad is not included in the precise ad queue corresponding to the candidate exposure opportunity, the estimated revenue per thousand impressions (eCPM) of the target ad is determined based on the bid information, estimated click-through rate, and estimated conversion rate of the target ad; and a judgment is made as to whether the estimated eCPM of the target ad is greater than the lowest eCPM of the precise ad queue corresponding to the candidate exposure opportunity. If so, it is determined that the target ad meets the reference revenue condition, and the candidate exposure opportunity is determined as the target exposure opportunity.

4. The method according to claim 1, wherein The exposure win rate prediction model determines the win rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertisement features of the target advertisement and the opportunity features of the candidate exposure opportunities, including: The exposure winning rate prediction model is used to determine the winning rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertisement features of the target advertisement, the viewing object features corresponding to the candidate exposure opportunities, and the exposure condition features corresponding to the candidate exposure opportunities.

5. The method according to claim 4, characterized in that The exposure win rate prediction model includes a feature map Embedding layer, a factorization machine FM layer and a deep neural network DNN layer; The exposure win rate prediction model determines the win rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertisement features of the target advertisement and the opportunity features of the candidate exposure opportunities, including: Determining, through the Embedding layer, dense feature vectors corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features; The FM layer and the DNN layer process the dense feature vectors corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features, respectively, to obtain feature scalars corresponding to the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features, respectively; Adding the bid information and advertisement features of the target advertisement, the viewing object features, and the exposure condition features to the corresponding feature scalars to obtain a target scalar; The winning rate of the target advertisement in the candidate exposure opportunities is determined according to the target scalar through the Sigmoid function.

6. The method according to claim 1, characterized in that Determining candidate exposure inventory based on the targeting conditions of the target advertisement includes: The candidate exposure inventory is predicted based on the targeting conditions of the target advertisement through a deep spatiotemporal tensor decomposition model; the deep spatiotemporal tensor decomposition model is used to predict the exposure inventory of the target advertisement within a preset time period through tensor decomposition based on the attribute combination characteristics of the targeting conditions and the temporal characteristics of historical advertisement delivery data on the target advertisement delivery platform; the target advertisement delivery platform is an advertisement delivery platform used to deliver the target advertisement.

7. The method according to claim 1, characterized in that The impression win rate prediction model is trained in the following way: Obtaining a training sample set; the training sample set includes a plurality of training samples; the training samples include bid information and advertisement features of candidate advertisements in historical exposure opportunities, opportunity features of the historical exposure opportunities, and exposure results of the candidate advertisements in the historical exposure opportunities; The exposure win rate prediction model is trained based on the training samples in the training sample set.

8. The method according to claim 7, characterized in that The obtaining of a training sample set includes: Determine a preset period of time before the model training moment as the target historical period; For each of the historical exposure opportunities occurring within the target historical period, obtaining a refined advertising queue corresponding to the historical exposure opportunity; For each candidate advertisement in the refined advertisement queue corresponding to each historical exposure opportunity, a training sample corresponding to the candidate advertisement in the historical exposure opportunity is formed using the bid information and advertisement features of the candidate advertisement, the opportunity features of the historical exposure opportunity, and the exposure results of the candidate advertisement in the historical exposure opportunity; The training sample set is formed by utilizing the training samples corresponding to each of the candidate advertisements in each of the historical exposure opportunities within the target historical period.

9. The method according to claim 1, characterized in that The method further comprises: receiving a bid modification request initiated by an advertisement delivery client for the target advertisement; the bid modification request includes updated bid information of the target advertisement; For the candidate exposure opportunities in the candidate exposure inventory, determine the updated winning rate of the target advertisement in the candidate exposure opportunities using the exposure winning rate estimation model based on the updated bid information and advertisement features of the target advertisement, as well as the opportunity features of the candidate exposure opportunities; and determine an updated exposure estimation result corresponding to the target advertisement based on the updated winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory; Sending the updated exposure estimation result to the advertising delivery client.

10. The method according to claim 1, characterized in that The method further comprises: Sending advertisement delivery reference information to an advertisement delivery client; the advertisement delivery reference information includes prompt information corresponding to a potential advertisement, and the prompt information is used to indicate a bid adjustment reference strategy corresponding to the potential advertisement.

11. The method according to claim 1, wherein The target advertisement corresponds to a plurality of continuous candidate bidding information; The exposure win rate prediction model determines the win rate of the target advertisement in the candidate exposure opportunities based on the bid information and advertisement features of the target advertisement and the opportunity features of the candidate exposure opportunities, including: For each type of candidate bid information, determine the winning rate of the target advertisement in the candidate exposure opportunity when the candidate bid information is used, using an exposure winning rate prediction model based on the candidate bid information and advertisement characteristics, as well as the opportunity characteristics of the candidate exposure opportunity; The determining of the exposure estimation result corresponding to the target advertisement according to the winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory includes: For each type of candidate bid information, determining an estimated exposure result of the target advertisement when using the candidate bid information based on a winning rate of the target advertisement among the candidate exposure opportunities included in the candidate exposure inventory when using the candidate bid information; The method further comprises: Generate an exposure change trend based on the exposure estimation results of the target advertisement when using the various candidate bidding information; the exposure change trend is used to represent the change trend of the exposure estimation results corresponding to the target advertisement under the multiple candidate bidding information; The exposure change trend is sent to an advertisement delivery client corresponding to the target advertisement.

12. An advertising exposure estimation device, characterized in that: The device comprises: An exposure inventory determination module is configured to determine a candidate exposure inventory based on the targeting conditions of the target advertisement; the candidate exposure inventory includes candidate exposure opportunities that match the targeting conditions; A win rate estimation module is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, the win rate of the target advertisement in the candidate exposure opportunities using an exposure win rate estimation model, based on the bid information and ad features of the target advertisement, and the opportunity features of the candidate exposure opportunities. The exposure win rate estimation model is trained based on the bid information and ad features of each of multiple candidate advertisements in historical exposure opportunities, the opportunity features of the historical exposure opportunities, and the exposure results of the historical exposure opportunities. The win rate is used to represent the probability of the target advertisement being exposed through the candidate exposure opportunities. The exposure result determination module is configured to determine an exposure estimation result corresponding to the target advertisement according to a winning rate of the target advertisement in the candidate exposure opportunities included in the candidate exposure inventory.

13. The device according to claim 12, characterized in that The device further comprises: A refined ranking queue determination module is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, a historically mapped exposure opportunity corresponding to the candidate exposure opportunity; and obtain a refined ranking advertisement queue corresponding to the historically mapped exposure opportunity as the refined ranking advertisement queue corresponding to the candidate exposure opportunity. An exposure opportunity screening module is configured to determine, for each candidate exposure opportunity in the candidate exposure inventory, whether the target advertisement meets the reference revenue condition of the refined ranking advertisement queue corresponding to the candidate exposure opportunity, and if so, determine the candidate exposure opportunity as the target exposure opportunity; The winning rate estimation module is specifically used for: For each target exposure opportunity, determining the winning rate of the target advertisement in the target exposure opportunity by using the exposure winning rate prediction model according to the bid information and advertisement characteristics of the target advertisement and the opportunity characteristics of the target exposure opportunity; The exposure result determination module is specifically used to: An exposure estimation result corresponding to the target advertisement is determined according to the winning rate of the target advertisement in each of the target exposure opportunities.

14. A device, characterized in that The device includes a processor and a memory; The memory is used to store computer programs; The processor is configured to execute the advertisement exposure estimation method according to any one of claims 1 to 11 according to the computer program.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the advertising exposure estimation method according to any one of claims 1 to 11.

Images