Method and device for determining advertisement bid

Abstract

The invention provides an advertisement bidding method and equipment, which can estimate parameters in a probability distribution function obeyed by data in a data set of an advertisement through a parameter estimation model, thereby obtaining a probability of delayed conversion of the advertisement after a bidding moment. Therefore, the bid of the advertisement at the bid moment can be obtained based on the advertisement on-time conversion quantity and the delay conversion probability. In other words, the future conversion number in the advertisement downloading scene can be estimated, so that the bid of the advertisement is adaptively adjusted, and the income of the advertisement platform can be guaranteed.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method and apparatus for determining advertising bids. Background Technology

[0002] Optimized cost per action (oCPA) is a smart, automated bidding strategy for performance advertising. Advertisers (e.g., an application, an app) predetermine their conversion goals (e.g., user registration, user usage, user payment) and provide their desired bids and historical conversion data to the advertising platform (e.g., media outlets). The advertising platform uses machine learning to predict conversion data and automatically bids, then charges based on the number of ad impressions, user clicks, or user downloads on the media.

[0003] However, in some cases, conversion events do not occur immediately after a user downloads the app (e.g., user registration, user usage, user payment, etc.), or because the advertiser does not synchronize conversion data with the advertising platform in real time, the advertising platform cannot obtain the conversion data in a timely manner, resulting in a discrepancy between the final bid and the actual situation, thus causing the advertising platform to suffer revenue losses. Summary of the Invention

[0004] This application provides a method for determining advertising bids. By predicting possible conversions after the bid time, it can encourage advertisers to spend their time and protect the revenue of the advertising platform.

[0005] Firstly, a method for determining an advertising bid is provided. This method can be performed by a device for determining advertising bids, or by a component of that device (e.g., a chip or circuit). For example, the device can be a computing device.

[0006] The method includes: estimating a first parameter based on the advertisement's dataset and a first model, the first parameter being a parameter in a probability distribution function following the conversion data after the advertisement is downloaded; the first model being a model used for parameter estimation, the input of the first model being the dataset, and the output of the first model being the first parameter; determining the probability of delayed conversion of the advertisement after the bidding time based on the first parameter and the probability distribution function; and determining the advertisement bid at the bidding time based on the total cost of the advertisement collected at the bidding time, the number of timely conversions of the advertisement, and the probability of delayed conversion of the advertisement.

[0007] In this application, "conversion" includes "on-time conversion" and "delayed conversion". On-time conversion occurs before the bidding time, while delayed conversion occurs after the bidding time.

[0008] In this application, the dataset of the advertisement can be understood as a dataset to be tested or a dataset to be inferred. The dataset of the advertisement may include at least one of the following: information about the advertisement, information about users who downloaded the advertisement, the number of conversions after the advertisement was downloaded, and the time when the dataset of the advertisement was collected. Among them, "information about the advertisement" may include, for example, the name of the advertisement, the size of the advertisement (e.g., the advertisement is for a certain app), the type of advertisement, etc.

[0009] In this application, the "first model" is a model used for parameter estimation, the input of which is the dataset of the advertisement, and the output of which is the first parameter.

[0010] In this application, the probability distribution function of the data converted after the advertisement is downloaded is an exponential distribution function.

[0011] In this application, the parameter estimation model includes at least one of the following: maximum likelihood function estimation model, Bayesian parameter estimation model, least squares estimation model, and maximum a posteriori probability estimation model.

[0012] Based on the above technical solution, in this application, the probability of delayed conversion of the advertisement after the bidding time can be determined by the first parameter and probability distribution function estimated by the first model. Then, based on the number of timely conversions and the probability of delayed conversions, the advertisement bid at the bidding time can be determined, which can promote the advertiser's consumption and ensure the revenue of the advertising platform.

[0013] In conjunction with the first aspect, in one possible implementation, the method further includes: training a deep learning model based on a training dataset to obtain the first model, wherein the first model is the deep learning model after training, wherein the input of the deep learning model is the training dataset, the output of the deep learning model is the parameters in the probability distribution function of the data converted after the advertisement is downloaded in the training dataset, and the deep learning model is a model for parameter estimation.

[0014] Based on the above technical solution, in this application, a first model for parameter estimation can be trained. This first model can be pre-trained or trained before executing this method. This application does not limit the timing of training the first model. Furthermore, the model can be trained on the computing device implementing the technical solution provided in this application, or on other computing devices. This application does not limit the computing device used for training the model.

[0015] In conjunction with the first aspect, in one possible implementation, the probability of delayed conversion of the ad after the bidding time is determined based on the first parameter and the probability distribution function, including: determining the probability of delayed conversion of the ad between the bidding time and the ad delivery time, assuming no conversion occurs at the bidding time, based on the first parameter and the probability distribution function.

[0016] In this application, the probability of delayed conversion can be understood as a conditional probability. The probability of delayed conversion of an advertisement can be obtained by calculating the conditional probability.

[0017] In conjunction with the first aspect, in one possible implementation, the ad bid at the bidding time is determined based on the total ad fee collected at the bidding time, the number of timely conversions, and the probability of delayed conversions. This includes: determining the average conversion cost of the ad based on the total ad fee collected at the bidding time, the number of timely conversions, and the probability of delayed conversions; and determining the ad bid at the bidding time based on the average conversion cost and the advertiser's expected conversion cost.

[0018] In this application, when calculating the ad bid at the bidding moment based on the number of timely conversions and the probability of delayed conversions, the average conversion cost of the ad can be calculated first. Based on the average conversion cost and the expected conversion cost determined by the advertiser, the ad bid at the bidding moment can be determined.

[0019] Secondly, this application proposes an apparatus for determining advertising bids, the computing apparatus being used to perform the method described in the first aspect above. Specifically, the apparatus may include units and / or modules for performing the method proposed in this application, such as a parameter estimation module and / or a determination module.

[0020] Thirdly, this application provides an apparatus for determining an advertising bid, the apparatus comprising: at least one processor for executing a computer program or instructions stored in a memory to execute the advertising bid apparatus described in the first aspect. Optionally, the apparatus further comprises a memory for storing the computer program or instructions. Optionally, the apparatus further comprises a communication interface through which the processor reads the computer program or instructions stored in the memory.

[0021] In one implementation, the device is a device for implementing the functions of the above-described method in a chip.

[0022] In another implementation, the computing device is a chip, chip system, or circuit used to implement the functions described above in a chip.

[0023] Fourthly, this application provides a processor, including: an input circuit, an output circuit, and a processing circuit. The processing circuit is used to receive signals through the input circuit and to transmit signals through the output circuit, causing the processor to execute the method described in the first aspect.

[0024] In specific implementation, the processor can be one or more chips, the input circuit can be input pins, the output circuit can be output pins, and the processing circuit can be transistors, gate circuits, flip-flops, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a transceiver, and the signal output by the output circuit can be, for example, but not limited to, output to and transmitted by a transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as both the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.

[0025] Unless otherwise specified, or if it does not contradict its actual function or internal logic in the relevant description, the transmission and acquisition / reception operations involved in the processor can be understood as processor output and reception, input and other operations, or as transmission and reception operations performed by radio frequency circuits and antennas. This application does not limit them in this regard.

[0026] Fifthly, a processing apparatus is provided, including a processor and a memory. The processor is used to read instructions stored in the memory and to receive signals via a transceiver and transmit signals via a transmitter to execute the method described in the first aspect.

[0027] Optionally, the processor may be one or more, and the memory may be one or more.

[0028] Optionally, the memory may be integrated with the processor, or the memory may be separated from the processor.

[0029] In specific implementation, the memory can be a non-transitory memory, such as read-only memory (ROM), which can be integrated with the processor on the same chip or set on different chips. The embodiments of this application do not limit the type of memory or the way the memory and processor are set.

[0030] Specifically, the processor can output data to the transmitter, and the processor can receive input data from the transceiver. The transmitter and transceiver can be collectively referred to as the transceiver.

[0031] The processing device mentioned in the fifth aspect above can be one or more chips. The processor in the processing device can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, integrated circuit, etc.; when implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0032] In a sixth aspect, a computing cluster is provided, including at least one computing device, each computing device including a processor and a memory; the processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device, such that the computing device cluster performs the method described in any possible implementation of the first aspect.

[0033] Optionally, the processor can be a general-purpose processor, which can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, integrated circuit, etc.; when implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0034] In a seventh aspect, a computer-readable storage medium is provided that stores program code for execution by a device, the program code including the method described in the first aspect.

[0035] Eighthly, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the method described in the first aspect.

[0036] A ninth aspect provides a chip system including a processor for calling and running a computer program from a memory, causing a device equipped with the chip system to perform the method of the first aspect described above. Attached Figure Description

[0037] Figure 1 This is a schematic block diagram illustrating the model training and inference provided in this application.

[0038] Figure 2 This is a schematic flowchart of the advertising bidding process provided in this application.

[0039] Figure 3 This is a schematic diagram illustrating the advertising conversion provided in this application.

[0040] Figure 4 This is a schematic flowchart of a method 400 for determining automatic advertising bids provided in this application.

[0041] Figure 5 This is a schematic block diagram of an advertising bidding device 500 provided in this application.

[0042] Figure 6 This is a schematic block diagram of an advertising bidding device 600 provided in this application. Detailed Implementation

[0043] To facilitate understanding of the technical solution of this application, the following is a brief introduction to the technical terms used in this application.

[0044] 1. Model Training

[0045] Figure 1 This is a schematic diagram of the model training and inference provided in this application, such as... Figure 1 As shown, during the training phase, a training set for the deep learning model needs to be constructed based on the objective. The training set includes multiple training data points, each labeled. The label of a training data point represents the correct answer to a specific question, and the label can indicate the objective of training the deep learning model using the training data. For example, to train a deep learning model that can identify different animals, the training set can include images of multiple different animals (i.e., training data). Each image can have a label identifying the type of animal it contains, such as cat or dog. In this example, the type of animal corresponding to each image is the label of that training data.

[0046] When training a deep learning model, training data can be input into the model in batches after parameter initialization. The deep learning model performs calculations (i.e., inference) on the training data to obtain prediction results. The prediction results obtained through inference, along with the corresponding labels of the training data, are used to calculate the loss based on the loss function. The loss function is used during the model training phase to calculate the difference (i.e., the loss value) between the model's prediction results on the training data and the labels of that training data. Loss functions can be implemented using different mathematical functions; commonly used expressions for loss functions include: mean squared error loss function, logarithmic loss function, least squares method, etc.

[0047] The loss value calculated based on the loss function can be used to update the parameters of the deep learning model. The gradient descent method is commonly used for parameter updating. Model training is a repetitive iterative process. Each iteration performs inference on different training data and calculates the loss value. The goal of multiple iterations is to continuously update the parameters of the deep learning model and find the parameter configuration that minimizes or stabilizes the loss value of the loss function.

[0048] Once trained, a deep learning model can be used to infer from input data. In the inference phase, data from real-world application scenarios is typically used as input. The trained deep learning model then infers the results. The inference phase is the practical application of the trained deep learning model, allowing for the rapid use of AI capabilities to solve specific technical problems. Today, AI has numerous applications, and deep learning model inference can be used in various scenarios, such as personnel identification in access control and security systems, video content moderation (including pornography and violence detection), and express delivery tracking number detection and recognition. For example, historical user data can be analyzed and modeled using machine learning algorithms to predict new user requests, thus providing personalized inference results.

[0049] The above only introduces the training of the most typical deep learning model as an example. The training of other types of models has slight differences, but the principle is similar. Most of them involve inference on the training data and adjusting the parameters in the model based on the inference results, with the goal of obtaining a parameter combination that makes the model performance stable.

[0050] 2. Parameter estimation

[0051] Probability studies the prediction of the likelihood of an event (e.g., new input data) given conditions (e.g., a known model and parameters); statistics studies the inductive reasoning of how data was generated (e.g., the model and parameters) given data (e.g., training data). After collecting data, a mathematical data model is typically built to make statistical inferences. This model describes the distribution of the collected data. In statistics, the process of estimating unknown parameters of a model based on a random sample drawn from the population is called parameter estimation. Parameter estimation methods aim to infer the values ​​of population parameters from collected sample data, thereby better describing the distribution characteristics of the population.

[0052] Commonly used parameter estimation methods include: maximum likelihood function method, Bayesian parameter estimation method, least squares estimation method, and maximum a posteriori probability estimation method. Among them, Bayesian parameter estimation method and least squares estimation method are used for parameter estimation of functional models, while maximum likelihood estimation method and maximum a posteriori estimation method are commonly used for parameter estimation of probabilistic models.

[0053] In this application, "model training" and "parameter estimation" are linked together, which can also be understood as the model training process being the parameter estimation process.

[0054] 3. Advertising Placement

[0055] Advertisers (such as applications, apps, or mini-games) place ads on advertising platforms (such as app stores). When users click and / or download, the advertiser needs to pay the advertising platform an advertising fee. Typically, advertising platforms use recommendation algorithms to rank and display ads to users. After seeing an ad, a user may trigger a series of actions, and the advertiser needs to pay the advertising platform for each user's actions.

[0056] 4. Optimized cost per action (oCPA)

[0057] In advertising scenarios, advertisers set conversion bids for each ad campaign ("conversion" can refer to actions such as activation, usage, and payment), and ultimately bid and execute ads based on the bids and the predicted click-through rate (PCTR) and predicted conversion rate (PCVR) estimated by the recommendation system. The industry currently widely uses oCPA (automatic cost-per-click) bidding. The oCPA process is briefly introduced below.

[0058] like Figure 2 As shown, once a user arrives, the ad placement is activated, and different ads need to bid to determine which ad is displayed in that position. Advertisers can set conversion goals, such as registration, and the expected cost per conversion, such as 10 yuan, as well as the number of days for the campaign (usually, the ad campaign duration is fixed). The advertising platform will automatically bid in each auction based on the advertiser's expected cost per conversion (the "bid" can also be understood as the fee that needs to be paid to the advertising platform after a user clicks and / or downloads), in order to achieve the advertiser's expected cost per conversion. After the bid is set, the advertising platform sorts the ads according to the PCTR, PCVR, and bid provided by the recommendation system, and finally displays the recommended list to the user. After the user sees it, they will click and / or download, and the platform will charge the advertiser according to the bid. If the user subsequently converts, the conversion count is incremented by 1. The average cost per conversion can be calculated based on the charges and the number of conversions. Based on this average cost per conversion and the user's expected cost per conversion, the bid is adjusted again to determine the final bid, and the bidding ends. At the end of the bidding process, the platform needs to ensure that the actual cost of each conversion is between 80% and 120% of the expected conversion cost; otherwise, the advertising platform needs to compensate the advertiser.

[0059] Typically, the average cost per conversion is calculated based on the total fees charged by the advertiser and the total number of conversions, as shown in the following formula (1):

[0060]

[0061] Among them, b i c represents the cumulative number of downloads. i The numerator in formula (1) represents the "cumulative number of downloads * expected conversion cost per download", which can also be understood as the total fee charged to the advertiser; the denominator in formula (1) represents the "total number of conversions", and CPA represents the average conversion cost up to the present. Under normal circumstances, the advertising platform needs to adjust the bid based on the current CPA. For example, when the current CPA is lower than the expected conversion cost, the advertising platform can increase the bid to promote the placement and increase the platform's revenue; when the current CPA is higher than the expected conversion cost, the bid should be reduced to protect the cost and prevent compensation to the advertiser. The real-time bid can usually be calculated based on the expected conversion cost and the average conversion cost. For example, the real-time bid can be calculated using the following formulas (2) to (4).

[0062] e i =v-cpa i Formula (2)

[0063] In the above formula (2), e i Used to characterize the difference between expected cost and actual average cost, where v represents expected conversion cost and cpa represents actual average cost. i Characterizes the actual average conversion cost.

[0064]

[0065] Formula (3) above is a general formula for proportional integral derivative controllers (PID), where K P For proportional units, K i For the integral unit, K d It is a differential unit (derivation).

[0066] b i =φ(u i ) Formula (4)

[0067] Formula (4) above represents the mapping rule from PID output to actual change amount bi, which can be adjusted according to the actual business situation.

[0068] As shown above, existing solutions determine the average cost of conversion based on the number of timely conversions, thus setting the bid accordingly. However, in some cases, a user doesn't immediately convert after downloading (e.g., user registration, user usage, user payment), or the advertiser fails to promptly synchronize conversion data with the advertising platform. In these situations, the advertising platform cannot obtain conversion data in a timely manner, leading to a discrepancy between the final bid and the actual situation, resulting in revenue loss for the platform. Figure 3 As shown in (a) above, since the conversion time is earlier than the bidding time, this can be understood as "on-time conversion". Figure 3 As shown in (b), since the conversion time is later than the bidding time, this can be understood as a "delayed conversion". For example, due to the delay, during peak periods, the advertising platform does not obtain the number of user conversions in time, resulting in a lower bid; during off-peak periods, the advertising platform obtains the number of user conversions from the previous peak period, resulting in a higher bid.

[0069] In view of this, this application provides a method for determining automatic advertising bids. The method estimates the parameters in the probability distribution function of the advertisement based on a first model, thereby predicting the number of delayed conversions after the advertisement is downloaded. Based on the number of on-time conversions and the number of delayed conversions, the advertising bid is determined, thereby ensuring the revenue of the advertising platform.

[0070] Figure 4 This is a schematic flowchart illustrating a method 400 for determining automatic advertising bids provided in this application, such as... Figure 4 As shown, the method includes:

[0071] 410. Based on the ad dataset and the first model, estimate the first parameter, which is the parameter in the probability distribution function that the converted data after the ad is downloaded follows.

[0072] In this application, the dataset of the advertisement can be understood as a dataset to be tested or a dataset to be inferred. The dataset of the advertisement may include at least one of the following: information about the advertisement, information about users who downloaded the advertisement, the number of conversions after the advertisement was downloaded, and the time when the dataset of the advertisement was collected. Among them, "information about the advertisement" may include, for example, the name of the advertisement, the size of the advertisement, the type of the advertisement, etc.

[0073] In this application, the "first model" is a model used for parameter estimation, the input of which is the dataset of the advertisement, and the output of which is the first parameter.

[0074] As mentioned above, "conversion" in this application includes timely conversion and delayed conversion, wherein the conversion time of timely conversion is earlier than the bidding time, and the conversion time of delayed conversion is later than the bidding time.

[0075] In one possible scenario, the method 400 further includes: training a deep learning model based on a training dataset to obtain the first model. The first model is the trained deep learning model, whose input is the training dataset, and whose output is the parameters of the probability distribution function following the conversion data after ad downloads in the training dataset. This deep learning model is used for parameter estimation.

[0076] The "model training" process in this application can be understood by referring to the machine learning process in neural networks in the current field of artificial intelligence.

[0077] For example, assuming that the conversion data after the advertisement is downloaded in the training dataset follows an exponential function of probability distribution, the following is a brief introduction to the idea and general process of model training in this application. For detailed steps of model training, please refer to the existing model training process for understanding, and it will not be described in detail here.

[0078] If we consider all the sample data in the training dataset as independent events, then the training dataset includes only two types of events: events that are transformed on time (represented by x1) or events that are transformed with a delay (represented by x2). See Table 1 below (refer to...). Figure 3 The conversion probability corresponding to an event that converts on time is "1", and the conversion probability corresponding to a delayed conversion time is a conditional probability. This conditional probability is used to characterize the probability of conversion from the time of bidding to the time of placement, provided that no conversion has occurred at the time of bidding.

[0079] Table 1

[0080] Conversion type Conversion probability Transform on time: t≥t′ 1 Delayed transformation: t≤t′ P(d≤Tt″|d≥tt″)

[0081] This can also be understood as follows: once user #1 downloads application software #1, the time when they register, activate, or use application software #1 is fixed. This time is either earlier than or equal to the bidding time, which is considered an on-time conversion; or later than the bidding time, which is considered a delayed conversion. In this application, the training dataset is given. Now, a probability distribution model is defined to describe the distribution of data in the training dataset. The goal is to find the probability model parameters that best match the data distribution of the training dataset.

[0082] This can also be understood as follows: the event has already occurred, and now we are looking for the parameters that maximize the probability of that event occurring. The model training process is about finding the most suitable parameters so that the final determined parameters best fit the training dataset. In other words, the model training process is about finding the most suitable probability distribution parameters based on the input training dataset and the model's internal optimization algorithm. When the most suitable probability distribution parameters are found, the model training can be considered complete. This model can then be used to predict the parameters of the probability distribution function of any input inference dataset (e.g., an advertising dataset). Therefore, deep learning models can be trained using typical parameter estimation methods.

[0083] In this application, the "parameter estimation model" includes at least one of the following: maximum likelihood function estimation model, Bayesian parameter estimation model, least squares estimation model, maximum a posteriori probability estimation model, etc.

[0084] For example, in the scenario of ad download conversion, assuming that the conversion data follows an exponential function distribution, the following will introduce the maximum likelihood function as the parameter estimation method. Assuming that the training dataset includes n sample data, the corresponding maximum likelihood function is as follows (5):

[0085] L=(x1,x2…x i …x n ;λ)=p(x1,λ)·p(x2,λ)…p(x i ,λ)…p(x n Formula (5) ,λ)

[0086] Assuming a total of 100 samples are collected, after extracting features from each sample, the following information can be represented: user characteristics, advertisement characteristics, whether conversion was completed before bidding, the delay time d from the download time to the conversion time, and the time k between the download time and the time of collecting training data. Since the conversion probability is a conditional probability in the case of delayed conversion, this conditional probability can be calculated based on the following formulas (6) to (8):

[0087] The probability density function of the indicator distribution function is: D(x) = λe -λx Formula (6)

[0088] The distribution function is:

[0089]

[0090] In the above formula, T is the launch time, t is the bidding time, and t″ is the download time.

[0091] Assume that the collected training sample set is Where, xi This can be understood as the feature of each sample, d i This can be understood as a delay time. If the sample belongs to an on-time conversion event, then... The value is 1; if the sample belongs to a delayed conversion event, then The value is 0. Based on this, the maximum likelihood loss function can be constructed as shown in the following formula (9):

[0092]

[0093] In the above formula (9), k i This can be understood as the time from the download moment to the time of collecting the training data.

[0094] Based on the collected samples, the above function has a total of 100 terms, each of which is either... (at this time, For 1), or is (at this time, (e.g., d = 0), and in each term d i and k i It is known. In general, it is only necessary to find the λ corresponding to the maximum value of logL, which indicates that the exponential function has the best fit with the samples in the training dataset. For simple first-order and second-order functions, the above formula (9) may be directly obtained by taking the derivative. However, the amount of data used for training sample datasets is quite large. Therefore, the model parameters can be iteratively optimized by combining model training with loss function to find the λ that best fits the training dataset. For example, the loss function can be the maximum likelihood loss function. When the loss value of the loss function is minimized or no longer changes, the model can be considered to have been trained. Subsequently, when using the first model, the advertising dataset is input into the first model. The first model will output λ, so that the specific exponential function that the advertising dataset follows can be obtained.

[0095] 420. Based on the first parameter and the probability distribution function, determine the probability of delayed conversion of the ad after the bidding time.

[0096] In one possible implementation, the probability of delayed conversion of the ad between the bidding time and the ad placement time is determined based on the first parameter and the probability distribution function, assuming no conversion occurs at the bidding time. For example, the probability of delayed conversion can be calculated using a conditional function, as shown in formula (8) above.

[0097] 430. Determine the ad bid at the time of bidding based on the total cost of the ad collected at the time of bidding, the number of timely conversions of the ad, and the probability of delayed conversions of the ad.

[0098] In one possible implementation, the average cost of conversion for an ad is determined based on the total cost of the ad collected at the time of bidding, the number of timely conversions, and the probability of delayed conversions; the ad bid at the time of bidding is determined based on the average cost of conversion and the advertiser's expected cost of conversion.

[0099] For example, based on the specific exponential function obtained, the expected value E of the conversion quantity can be calculated using the following formula (10). conve The average conversion cost is calculated. For example, the number of downloads by users in the data set can be used to determine whether each download belongs to "on-time conversion" or "delayed conversion". If it belongs to "on-time conversion", the conversion probability is 1; if it belongs to "delayed conversion", the conversion probability is calculated based on the specific indicator function obtained. Then, the probabilities corresponding to each download are added together to obtain the expected number of conversions. The average conversion cost is the ratio of the total fee charged to the advertiser at the bidding time (number of downloads × advertising bid) to the expected number of conversions. After obtaining the average conversion cost, the advertising bid at the bidding time can be calculated based on the aforementioned formulas (2) to (4).

[0100]

[0101] Table 2 below shows the revenue of the advertising platform after using the method of this application (which can also be understood as the increase in advertiser spending). As shown in Table 2, by using the method of this application and based on data collected over a certain 5 days, it was found that the advertiser spending for each day showed an upward trend, thereby promoting the revenue of the advertising platform. Specifically, the data from the first day increased the spending of each advertiser by 2.5%, the data from the second day increased the spending of each advertiser by 4.6%, the data from the third day increased the spending of each advertiser by 1.7%, the data from the fourth day increased the spending of each advertiser by 3%, and the data from the fifth day increased the spending of each advertiser by 1.29%.

[0102] Table 2

[0103]

[0104]

[0105] Based on the above technical solution, this application uses a parameter estimation model to estimate the parameters in the probability distribution function that the data in the advertising dataset follows. This allows for the estimation of the probability of delayed conversion after the bidding time, thus enabling the determination of the advertising bid at the bidding time based on the number of timely conversions and the probability of delayed conversions. In other words, this application can predict the future conversion volume in an advertising download scenario, thereby adaptively adjusting the advertising bid and ensuring the advertising platform's revenue.

[0106] It should be understood that the method provided in this application can be applied to scenarios where advertising bidding is applicable, as well as other scenarios where future conversion revenue is predicted. For example, the solution provided in this application can be applied to machine learning systems that are mainly recommendation systems.

[0107] Those skilled in the art will recognize that, based on the units and algorithm steps described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is implemented in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0108] This application embodiment can divide the advertising bidding device into functional modules according to the above method example. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated modules can be implemented in hardware or software. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods. The following description uses the division of functional modules according to each function as an example.

[0109] Figure 5 This is a schematic block diagram of a computing device 500 provided in an embodiment of this application. As shown in the figure, the device 500 may include: a parameter estimation module 510 and a determination module 520. Optionally, it may also include a training module 530.

[0110] The modules described above are used to execute the respective steps of the methods mentioned above, which will not be elaborated here.

[0111] It should also be understood that the device 500 here is embodied in the form of a functional unit. The term "unit" here may refer to application-specific integrated circuits (ASICs), electronic circuits, processors (e.g., shared processors, proprietary processors, or group processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functions.

[0112] The devices 500 in each of the above-described schemes have the function of implementing the corresponding steps of the methods 400 described above. These functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions; for example, the determining unit can be replaced by a processor to execute the transmit / receive operations and related processing operations in each method embodiment. Furthermore, the determining unit can be a processing circuit.

[0113] It should be pointed out that, Figure 5 The device mentioned can be a device used to execute the aforementioned method embodiments, or it can be a chip or chip system corresponding to a computing device (or a cloud management platform, or a communication device used in a cloud management platform), such as a system-on-a-chip (SoC). The processing module is a processor, microprocessor, or integrated circuit integrated on the chip. No limitation is made here.

[0114] Figure 6 This is a schematic block diagram of another device 600 for advertising bidding provided in this application embodiment. As shown, the device 600 includes: at least one processor 620. The processor 620 is coupled to a memory for executing instructions stored in the memory to transmit and / or receive signals. Optionally, the device 600 also includes a memory 630 for storing instructions. Optionally, the device 600 also includes a transceiver 610, which the processor 620 controls to transmit and / or receive signals.

[0115] It should be understood that the processor 620 and memory 630 described above can be combined into a single processing device, with the processor 620 executing the program code stored in the memory 630 to achieve the aforementioned functions. In specific implementations, the memory 630 can be integrated into the processor 620 or independent of the processor 620.

[0116] It should also be understood that transceiver 610 may include a transceiver (or receiver) and a transmitter (or transmitter). The transceiver may further include an antenna, and the number of antennas may be one or more. Transceiver 610 may have a communication interface or interface circuitry.

[0117] Specifically, the processor 620 in device 600 may correspond to the parameter estimation module 510 and the determination module 520 in device 500. In some scenarios, when the device can be used for model training, the processor 620 in the device may also correspond to the model training module 530 in device 500.

[0118] As one option, the device 600 is used to implement the various steps in the above-described method 400 embodiment.

[0119] For example, processor 620 is used to execute computer programs or instructions stored in memory 630 to implement the various steps in method 400 above.

[0120] In this embodiment, a computer program product containing instructions is also provided. The computer program product may be a software or program product containing instructions capable of running on a computing device cluster or stored on any available medium. When run by the computing device cluster, it causes the computing device cluster to perform the methods provided above, or causes the computing device cluster to implement the functions of the apparatus provided above.

[0121] In this embodiment, a computer-readable storage medium is also provided. This computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions that, when executed on a computing device, cause the computing device to perform the method described above.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0125] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0126] In addition, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0127] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining advertising bids, characterized in that, include: Based on the ad dataset and the first model, a first parameter is estimated. The first parameter is a parameter in the probability distribution function that the conversion data after the ad is downloaded follows. The first model is a model for parameter estimation. The input of the first model is the dataset, and the output of the first model is the first parameter. The conversion includes on-time conversion and delayed conversion. The on-time conversion occurs earlier than the bidding time, and the delayed conversion occurs later than the bidding time. Based on the first parameter and the probability distribution function, determine the probability of delayed conversion of the advertisement after the bidding time; The ad bid at the bidding time is determined based on the total cost of the ad collected at the bidding time, the number of timely conversions of the ad, and the probability of delayed conversions of the ad.

2. The method according to claim 1, characterized in that, The method further includes: The deep learning model is trained based on the training dataset to obtain the first model, which is the deep learning model after training. The input of the deep learning model is the training dataset, and the output of the deep learning model is the parameters in the probability distribution function of the data converted after the advertisement is downloaded in the training dataset. The deep learning model is a model for parameter estimation.

3. The method according to claim 1 or 2, characterized in that, Determining the probability of delayed conversion of the advertisement after the bidding time based on the first parameter and the probability distribution function includes: Based on the first parameter and the probability distribution function, determine the probability of delayed conversion of the advertisement between the bidding time and the advertisement placement time, assuming no conversion occurs at the bidding time.

4. The method according to any one of claims 1 to 3, characterized in that, The ad bid at the bidding time is determined based on the total cost of the ad collected at the bidding time, the number of timely conversions of the ad, and the probability of delayed conversions of the ad, including: The average conversion cost of the advertisement is determined based on the total cost of the advertisement charged at the time of bidding, the number of timely conversions of the advertisement, and the probability of delayed conversion of the advertisement. The ad bid at the bidding moment is determined based on the average conversion cost and the advertiser's expected conversion cost.

5. The method according to any one of claims 1 to 4, characterized in that, The data converted after the advertisement is downloaded follows an exponential distribution function.

6. The method according to any one of claims 1 to 5, characterized in that, The parameter estimation model includes at least one of the following: maximum likelihood function estimation model, Bayesian parameter estimation model, least squares estimation model, and maximum a posteriori probability estimation model.

7. The method according to any one of claims 1 to 6, characterized in that, The dataset for the advertisement includes one or more of the following: information about the advertisement, information about users who downloaded the advertisement, the number of users who converted after the advertisement was downloaded, and the time when the dataset was collected.

8. A device for determining advertising bids, characterized in that, include: The parameter estimation module and the determination module, wherein, The parameter estimation module is used to estimate a first parameter based on the ad dataset and a first model. The first parameter is a parameter in the probability distribution function that the conversion data after the ad is downloaded follows. The first model is a model used for parameter estimation. The input of the first model is the dataset, and the output of the first model is the first parameter. The conversion includes on-time conversion and delayed conversion. The on-time conversion occurs earlier than the bidding time, and the delayed conversion occurs later than the bidding time. The determining module is used to determine the probability of delayed conversion of the advertisement after the bidding time based on the first parameter and the probability distribution function. The determining module is used to determine the advertising bid at the bidding time based on the total cost of the advertising collected at the bidding time, the number of timely conversions of the advertising, and the probability of delayed conversions of the advertising.

9. The device according to claim 8, characterized in that, The device also includes a training module. The training module is used to train a deep learning model based on a training dataset and obtain the first model, which is the deep learning model after training. The input of the deep learning model is the training dataset, and the output of the deep learning model is the parameters in the probability distribution function of the data converted after the advertisement is downloaded in the training dataset. The deep learning model is a model for parameter estimation.

10. The device according to claim 8 or 9, characterized in that, The determining module is configured to determine, based on the first parameter and the probability distribution function, the probability of delayed conversion of the advertisement after the bidding time, including: The determining module is used to determine, based on the first parameter and the probability distribution function, the probability of delayed conversion of the advertisement between the bidding time and the advertisement placement time, under the condition that no conversion occurs at the bidding time.

11. The device according to any one of claims 8 to 10, characterized in that, The determining module is used to determine the ad bid at the bidding time based on the total cost of the ad collected at the bidding time, the number of timely conversions of the ad, and the probability of delayed conversion of the ad, including: The determining module is used to determine the average conversion cost of the advertisement based on the total cost of the advertisement collected at the bidding time, the number of timely conversions of the advertisement, and the probability of delayed conversions of the advertisement. The determining module is used to determine the ad bid at the bidding time based on the average conversion cost and the expected conversion cost determined by the advertiser.

12. The device according to any one of claims 8 to 11, characterized in that, The data converted after the advertisement is downloaded follows an exponential distribution function.

13. The device according to any one of claims 8 to 12, characterized in that, The parameter estimation model includes at least one of the following: maximum likelihood function estimation model, Bayesian parameter estimation model, least squares estimation model, and maximum a posteriori probability estimation model.

14. The device according to any one of claims 8 to 13, characterized in that, The dataset for the advertisement includes one or more of the following: information about the advertisement, information about users who downloaded the advertisement, the number of users who converted after the advertisement was downloaded, and the time when the dataset was collected.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

16. A computer program product, characterized in that, Includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images