Data processing method, device and equipment and readable storage medium

Abstract

The invention discloses a data processing method, apparatus and device, and a readable storage medium. The method comprises the steps of obtaining a service conversion data set; performing conversion data volume statistics on the conversion nodes of the service media according to the service conversion data set to obtain a conversion statistical result, and generating a conversion quality result for the service media according to the conversion statistical result; performing return quality identification on the return process of the service conversion data set to obtain a return quality result; the return process of the business conversion data set refers to a process of returning the business conversion data set by the main object; if the conversion quality result and the return quality result both meet the sample quality condition, providing the service conversion data set as a training sample to a downstream processing node; and the downstream processing node is used for estimating a conversion index aiming at the conversion behavior. By adopting the method and the device, the quality of the business conversion data can be measured so as to ensure the estimation precision of the downstream processing node.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, apparatus, device, and readable storage medium. Background Technology

[0002] Business conversion data refers to the conversion behavior performed by an object targeting a specific business media platform during the business media campaign. This data can be provided by the main object publishing the business media and can be used as training samples for machine learning models within the business media system to train predictive models for estimating conversion metrics. However, business conversion data requires manual uploading (e.g., by the main object) to the business media platform, which can lead to anomalies in the data during the upload process. This can result in insufficient accuracy of the predictive models trained using these abnormal conversion data. Summary of the Invention

[0003] This application provides a data processing method, apparatus, device, and readable storage medium that can ensure the prediction accuracy of downstream processing nodes by measuring the quality of business conversion data.

[0004] One embodiment of this application provides a data processing method, including:

[0005] Obtain the business conversion dataset; the business conversion dataset includes data on the conversion behaviors performed by the object against the business media that have been launched;

[0006] Based on the business conversion dataset, the conversion data volume of the conversion nodes of the business media is statistically analyzed to obtain conversion statistics results. Based on the conversion statistics results, conversion quality results for the business media are generated.

[0007] The process of transmitting the business conversion dataset back is used to identify the transmission quality and obtain the transmission quality results. The transmission process of the business conversion dataset refers to the process of transmitting the business conversion dataset back by the master object. The master object refers to the object that provides the business media.

[0008] If both the conversion quality results and the feedback quality results meet the sample quality conditions, the business conversion dataset will be provided as training samples to the downstream processing nodes; the downstream processing nodes will then be used to predict conversion metrics for conversion behaviors.

[0009] The conversion statistics include first and second statistical results, and conversion nodes include shallow and deep conversion nodes. Based on the business conversion dataset, conversion data volume is statistically analyzed for the conversion nodes of the business media to obtain conversion statistics. Based on these statistics, conversion quality results for the business media are generated, including:

[0010] In the business conversion dataset, the number of business conversion data that have already executed conversion actions in shallow conversion nodes and have also executed conversion actions in deep conversion nodes is counted as the first statistical result; the execution order of business media executing conversion actions in shallow conversion nodes is placed before the execution order of conversion actions executing in deep conversion nodes;

[0011] In the business conversion dataset, the number of business conversion data that have executed conversion behaviors in deep conversion nodes is counted as a second statistical result;

[0012] The ratio of the first statistical result to the second statistical result is determined as the conversion quality result for business media.

[0013] The method also includes:

[0014] If the transformation quality result is greater than or equal to the first proportion threshold, then the transformation quality result is determined to meet the sample quality condition.

[0015] The return quality results include centralized return quality results; the return quality is identified during the return process of the business conversion dataset to obtain return quality results, including:

[0016] Retrieve the first business conversion data belonging to the current time range within the current period in the business conversion dataset, and the second business conversion data belonging to the full historical range time range in each of N historical periods, where N is a positive integer; the full historical range time range includes multiple time ranges used to evenly divide a historical period.

[0017] Obtain the first business conversion value corresponding to the first business conversion data, obtain the first conversion mean and first conversion standard deviation corresponding to the second business conversion data, and generate the month-on-month conversion result for the business conversion dataset based on the first business conversion value, the first conversion mean, and the first conversion standard deviation;

[0018] Obtain the third business conversion data in the business conversion dataset that belongs to the same historical range time period within each of the N historical periods; the same historical range time period includes multiple range time periods that are adjacent to the current range time period in the unit time dimension;

[0019] Obtain the second conversion mean and second conversion standard deviation corresponding to the third business conversion data, and generate year-on-year conversion results for the business conversion dataset based on the first business conversion value, the second conversion mean, and the second conversion standard deviation;

[0020] The month-on-month conversion results and year-on-year conversion results are determined as the centralized quality results for the business conversion dataset;

[0021] The method also includes:

[0022] If both the month-on-month conversion result and the year-on-year conversion result in the centralized back-transmission quality results are less than the numerical threshold, then the centralized back-transmission quality results are determined to meet the sample quality conditions.

[0023] The return quality results include delayed return quality results; the return quality is identified during the return process of the business conversion dataset, and the return quality results are obtained, including:

[0024] Obtain the fourth business conversion data that meets the first business dimension from the business conversion dataset, and calculate the proportion of business conversion data in the fourth business conversion data whose reporting delay is greater than or equal to the delay threshold. Determine the proportion as the delay feedback quality result for the business conversion dataset. The reporting delay is determined based on the background receiving time of the conversion behavior associated with the fourth business conversion data and the occurrence time of the conversion behavior contained in the fourth business conversion data.

[0025] The method also includes:

[0026] If the delayed return quality result is less than the second proportional threshold, then the delayed return quality result is determined to meet the sample quality condition.

[0027] The return quality results include repeated return quality results; the return quality identification process of the business conversion dataset is performed to obtain the return quality results, including:

[0028] Obtain the fifth business conversion data that meets the second business dimension from the business conversion dataset, and calculate the target daily conversion volume corresponding to the fifth business conversion data.

[0029] Obtain the daily conversion volume of H business media domains under the second business dimension. Generate the daily conversion mean and daily conversion standard deviation based on the H daily conversion volumes. Generate the daily conversion volume threshold based on the daily conversion mean and daily conversion standard deviation. Determine the daily conversion volume threshold and the target daily conversion volume as the repeated feedback quality result for the business conversion dataset.

[0030] The method also includes:

[0031] If the target daily conversion rate in the repeated quality return results is less than the daily conversion rate threshold, then the repeated quality return results are determined to meet the sample quality conditions.

[0032] The feedback quality results include those with errors or omissions; the feedback quality is identified during the feedback process of the business conversion dataset to obtain feedback quality results, including:

[0033] In the business conversion dataset, retrieve the business conversion data where the object has performed a conversion action on the landing page of the business media, and use it as the sixth business conversion data;

[0034] The sixth business conversion data is compared with the conversion behavior data of the landing page for business media recorded by the business media platform to obtain the first comparison result;

[0035] The second comparison result is obtained by comparing the number of conversion data of the sixth business with the number of conversion behavior data recorded by the business media platform;

[0036] The first and second comparison results are determined as the quality results of error omissions in the business conversion dataset.

[0037] The method also includes:

[0038] If the first comparison result indicates that the conversion data of the sixth business is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the number of conversion data of the sixth business is equal to the number of conversion behavior data recorded by the business media platform, then it is determined that the quality result of the error omission feedback meets the sample quality condition.

[0039] The downstream processing nodes include a conversion rate prediction node; the method also includes:

[0040] Input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node;

[0041] The conversion rate prediction model extracts features from the business conversion dataset to obtain object features, business media features, and object media interaction features. Based on the object features, business media features, and object media interaction features, the predicted probability of the object performing conversion behavior in response to the business media is generated.

[0042] A first loss value is generated based on the predicted probability and the actual conversion rate corresponding to the business conversion dataset. The model parameters of the conversion rate prediction model are then adjusted based on the first loss value to obtain a target conversion rate prediction model used to predict the probability of the target object converting to business media that have not been advertised.

[0043] The downstream processing nodes include cost control nodes; the method also includes:

[0044] The business conversion data that has already performed conversion behavior in the deep conversion node in the business conversion dataset is determined as the training label, and the business conversion data that has already performed conversion behavior in the shallow conversion node and is associated with the training label in the business conversion dataset is determined as the shallow training sample. The exposure click data, training label and shallow training sample of the business media platform for business media are input into the return flow prediction model in the cost control node.

[0045] In the backflow prediction model, shallow conversion features corresponding to shallow training samples and exposure click features of exposure click data are extracted.

[0046] Based on shallow conversion features and exposure-click features, predict the deep conversion prediction probability corresponding to shallow training samples, and generate a second loss value based on the deep conversion prediction probability and training labels.

[0047] The model parameters of the reflux prediction model are adjusted based on the second loss value to obtain a reflux prediction model for predicting the deep conversion probability of target conversion business data; target conversion business data refers to conversion business data that has already performed conversion behavior in shallow conversion nodes.

[0048] The downstream processing nodes include cold start exploration nodes; the cold start exploration nodes include cold start exploration pool A and cold start exploration pool B; cold start exploration pool A is a network model trained on business conversion data that has already performed conversion behavior in shallow conversion nodes in the business conversion dataset, and cold start exploration pool B is a network model trained on business conversion data that has already performed conversion behavior in deep conversion nodes in the business conversion dataset; the method also includes:

[0049] Input the new business media into the cold start exploration node, and estimate the conversion rate of the new business media for the shallow conversion node based on the cold start exploration pool A to obtain the shallow conversion probability of the new business media; new business media refers to business media that have not been launched.

[0050] If the shallow conversion probability of the new business media is less than the shallow exploration probability threshold, then stop the cold start exploration of the new business media.

[0051] If the shallow conversion probability of the new business media is greater than or equal to the shallow exploration probability threshold, the new business media will be put on trial and the corresponding trial business conversion volume of the new business media during the trial period will be obtained.

[0052] If the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, then based on the cold start exploration pool B, the conversion rate of the new business media for deep conversion nodes will be estimated to obtain the deep conversion probability of the new business media.

[0053] If the probability of deep conversion is greater than or equal to the probability threshold of deep exploration, then the new business media will be officially launched.

[0054] One embodiment of this application provides a data processing apparatus, including:

[0055] The send / receive module is used to acquire business conversion datasets; the business conversion datasets include data on the conversion behaviors performed by objects for the business media that have been deployed;

[0056] The first processing module is used to perform conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset, obtain conversion statistics results, and generate conversion quality results for the business media based on the conversion statistics results.

[0057] The second processing module is used to identify the return quality of the business conversion dataset during the return process and obtain the return quality result. The return process of the business conversion dataset refers to the process of the main object returning the business conversion dataset. The main object refers to the object that provides the business media.

[0058] The transceiver module is also used to provide the business conversion dataset as training samples to the downstream processing nodes if both the conversion quality results and the return quality results meet the sample quality conditions; the downstream processing nodes are used to predict conversion metrics for conversion behavior.

[0059] In one possible implementation, the conversion statistics result includes a first statistical result and a second statistical result, and the conversion nodes include shallow conversion nodes and deep conversion nodes; the first processing module is used to perform conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset to obtain conversion statistics results. When generating conversion quality results for the business media based on the conversion statistics results, it is specifically used to perform the following operations:

[0060] In the business conversion dataset, the number of business conversion data that have already executed conversion actions in shallow conversion nodes and have also executed conversion actions in deep conversion nodes is counted as the first statistical result; the execution order of business media executing conversion actions in shallow conversion nodes is placed before the execution order of conversion actions executing in deep conversion nodes;

[0061] In the business conversion dataset, the number of business conversion data that have executed conversion behaviors in deep conversion nodes is counted as a second statistical result;

[0062] The ratio of the first statistical result to the second statistical result is determined as the conversion quality result for business media.

[0063] The first processing module is also used to perform the following operations:

[0064] If the transformation quality result is greater than or equal to the first proportion threshold, then the transformation quality result is determined to meet the sample quality condition.

[0065] In one possible implementation, the return quality results include centralized return quality results; the second processing module is used to identify the return quality of the business conversion dataset during the return process, and when obtaining the return quality results, it is specifically used to perform the following operations:

[0066] Retrieve the first business conversion data belonging to the current time range within the current period in the business conversion dataset, and the second business conversion data belonging to the full historical range time range in each of N historical periods, where N is a positive integer; the full historical range time range includes multiple time ranges used to evenly divide a historical period.

[0067] Obtain the first business conversion value corresponding to the first business conversion data, obtain the first conversion mean and first conversion standard deviation corresponding to the second business conversion data, and generate the month-on-month conversion result for the business conversion dataset based on the first business conversion value, the first conversion mean, and the first conversion standard deviation;

[0068] Obtain the third business conversion data in the business conversion dataset that belongs to the same historical range time period within each of the N historical periods; the same historical range time period includes multiple range time periods that are adjacent to the current range time period in the unit time dimension;

[0069] Obtain the second conversion mean and second conversion standard deviation corresponding to the third business conversion data, and generate year-on-year conversion results for the business conversion dataset based on the first business conversion value, the second conversion mean, and the second conversion standard deviation;

[0070] The month-on-month conversion results and year-on-year conversion results are determined as the centralized quality results for the business conversion dataset;

[0071] The second processing module is also used to perform the following operations:

[0072] If both the month-on-month conversion result and the year-on-year conversion result in the centralized back-transmission quality results are less than the numerical threshold, then the centralized back-transmission quality results are determined to meet the sample quality conditions.

[0073] In one possible implementation, the return quality results include delayed return quality results; the second processing module is used to identify the return quality of the business conversion dataset during the return process, and when obtaining the return quality results, it is specifically used to perform the following operations:

[0074] Obtain the fourth business conversion data that meets the first business dimension from the business conversion dataset, and calculate the proportion of business conversion data in the fourth business conversion data whose reporting delay is greater than or equal to the delay threshold. Determine the proportion as the delay feedback quality result for the business conversion dataset. The reporting delay is determined based on the background receiving time of the conversion behavior associated with the fourth business conversion data and the occurrence time of the conversion behavior contained in the fourth business conversion data.

[0075] The second processing module is also used to perform the following operations:

[0076] If the delayed return quality result is less than the second proportional threshold, then the delayed return quality result is determined to meet the sample quality condition.

[0077] In one possible implementation, the feedback quality results include repeated feedback quality results; the second processing module is used to identify the feedback quality of the business conversion dataset feedback process, and when obtaining the feedback quality results, it is specifically used to perform the following operations:

[0078] Obtain the fifth business conversion data that meets the second business dimension from the business conversion dataset, and calculate the target daily conversion volume corresponding to the fifth business conversion data.

[0079] Obtain the daily conversion volume of H business media domains under the second business dimension. Generate the daily conversion mean and daily conversion standard deviation based on the H daily conversion volumes. Generate the daily conversion volume threshold based on the daily conversion mean and daily conversion standard deviation. Determine the daily conversion volume threshold and the target daily conversion volume as the repeated feedback quality result for the business conversion dataset.

[0080] The second processing module is also used to perform the following operations:

[0081] If the target daily conversion rate in the repeated quality return results is less than the daily conversion rate threshold, then the repeated quality return results are determined to meet the sample quality conditions.

[0082] In one possible implementation, the feedback quality results include error and omission feedback quality results; the second processing module is used to identify the feedback quality of the business conversion dataset feedback process, and when obtaining the feedback quality results, it is specifically used to perform the following operations:

[0083] In the business conversion dataset, retrieve the business conversion data where the object has performed a conversion action on the landing page of the business media, and use it as the sixth business conversion data;

[0084] The sixth business conversion data is compared with the conversion behavior data of the landing page for business media recorded by the business media platform to obtain the first comparison result;

[0085] The second comparison result is obtained by comparing the number of conversion data of the sixth business with the number of conversion behavior data recorded by the business media platform;

[0086] The first and second comparison results are determined as the quality results of error omissions in the business conversion dataset.

[0087] The second processing module is also used to perform the following operations:

[0088] If the first comparison result indicates that the conversion data of the sixth business is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the number of conversion data of the sixth business is equal to the number of conversion behavior data recorded by the business media platform, then it is determined that the quality result of the error omission feedback meets the sample quality condition.

[0089] In one possible implementation, the downstream processing node includes a conversion rate prediction node; the data processing device also includes a training module, which is used to perform the following operations:

[0090] Input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node;

[0091] The conversion rate prediction model extracts features from the business conversion dataset to obtain object features, business media features, and object media interaction features. Based on the object features, business media features, and object media interaction features, the predicted probability of the object performing conversion behavior in response to the business media is generated.

[0092] A first loss value is generated based on the predicted probability and the actual conversion rate corresponding to the business conversion dataset. The model parameters of the conversion rate prediction model are then adjusted based on the first loss value to obtain a target conversion rate prediction model used to predict the probability of the target object converting to business media that have not been advertised.

[0093] In one possible implementation, the downstream processing node includes a cost control node; the data processing device also includes a training module, which is used to perform the following operations:

[0094] The business conversion data that has already performed conversion behavior in the deep conversion node in the business conversion dataset is determined as the training label, and the business conversion data that has already performed conversion behavior in the shallow conversion node and is associated with the training label in the business conversion dataset is determined as the shallow training sample. The exposure click data, training label and shallow training sample of the business media platform for business media are input into the return flow prediction model in the cost control node.

[0095] In the backflow prediction model, shallow conversion features corresponding to shallow training samples and exposure click features of exposure click data are extracted.

[0096] Based on shallow conversion features and exposure-click features, predict the deep conversion prediction probability corresponding to shallow training samples, and generate a second loss value based on the deep conversion prediction probability and training labels.

[0097] The model parameters of the reflux prediction model are adjusted based on the second loss value to obtain a reflux prediction model for predicting the deep conversion probability of target conversion business data; target conversion business data refers to conversion business data that has already performed conversion behavior in shallow conversion nodes.

[0098] In one possible implementation, the downstream processing nodes include cold start exploration nodes; the cold start exploration nodes include cold start exploration pool A and cold start exploration pool B; cold start exploration pool A is a network model trained on business conversion data that has already performed conversion actions in shallow conversion nodes in the business conversion dataset, and cold start exploration pool B is a network model trained on business conversion data that has already performed conversion actions in deep conversion nodes in the business conversion dataset; the data processing device also includes a cold start module, which is used to perform the following operations:

[0099] Input the new business media into the cold start exploration node, and estimate the conversion rate of the new business media for the shallow conversion node based on the cold start exploration pool A to obtain the shallow conversion probability of the new business media; new business media refers to business media that have not been launched.

[0100] If the shallow conversion probability of the new business media is less than the shallow exploration probability threshold, then stop the cold start exploration of the new business media.

[0101] If the shallow conversion probability of the new business media is greater than or equal to the shallow exploration probability threshold, the new business media will be put on trial and the corresponding trial business conversion volume of the new business media during the trial period will be obtained.

[0102] If the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, then based on the cold start exploration pool B, the conversion rate of the new business media for deep conversion nodes will be estimated to obtain the deep conversion probability of the new business media.

[0103] If the probability of deep conversion is greater than or equal to the probability threshold of deep exploration, then the new business media will be officially launched.

[0104] One embodiment of this application provides a computer device, including: a processor, a memory, and a network interface;

[0105] The processor is connected to a memory and a network interface. The network interface is used to provide data communication functions, and the memory is used to store computer programs. When the computer program is executed by the processor, the computer device performs the method provided in the embodiments of this application.

[0106] One aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded and executed by a processor, so that a computer device having the processor performs the method provided in this application.

[0107] One embodiment of this application provides a computer program product comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the method provided in this application embodiment.

[0108] This application embodiment records the conversion behavior performed by an object on the deployed business media as business conversion data, and records the collected business conversion data as a business conversion dataset. By statistically analyzing the conversion data volume of the conversion nodes of the business media using the business conversion dataset, conversion statistics can be obtained, thereby generating conversion quality results for the business media. Furthermore, by identifying the return quality of the business conversion dataset from the main object, a return quality result can be obtained; where the main object refers to the object providing the business media. By analyzing the conversion quality results and the return quality results, a business conversion dataset that meets the sample quality conditions can be obtained. This can filter out abnormal business conversion data caused by tampering with the main object or transmission failure of the business media system. Therefore, by analyzing the conversion quality results and the return quality results, it can be ensured that accurate and high-quality business conversion datasets are transmitted as training samples to downstream processing nodes, so that the downstream processing nodes can train high-quality prediction models and improve the accuracy of downstream processing nodes in predicting conversion indicators. Attached Figure Description

[0109] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0110] Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application;

[0111] Figure 2 This is a schematic diagram of a data processing scenario provided in an embodiment of this application;

[0112] Figure 3 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 1 ;

[0113] Figure 4 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 2 ;

[0114] Figure 5This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 1 ;

[0115] Figure 6 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 3 ;

[0116] Figure 7 This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 2 ;

[0117] Figure 8 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 4 ;

[0118] Figure 9 This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 3 ;

[0119] Figure 10 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application;

[0120] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0121] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0122] It is understood that in the specific embodiments of this application, the object (object or player) data involved, when applied to specific products or technologies in the above and below embodiments of this application, requires the permission or consent of the object, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant regions.

[0123] If object data needs to be collected in this application, a prompt interface or pop-up window will be displayed before and during the collection process. This prompt interface or pop-up window is used to inform the object that certain data is currently being collected. The data acquisition steps will only begin after the object confirms the prompt interface or pop-up window; otherwise, the process will end. Furthermore, the acquired object data will be used in reasonable and legal scenarios or for legitimate purposes. Optionally, in scenarios where object data needs to be used but the object has not authorized its use, authorization can be requested from the object, and the object data can only be used after authorization is granted.

[0124] Please see Figure 1 , Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application. For example... Figure 1 As shown, the network architecture may include a service server 100 and a terminal device 200. The service server 100 may have a communication connection with the terminal device 200. The communication connection is not limited to a specific method. It may be directly or indirectly connected via wired communication, or directly or indirectly connected via wireless communication, or in other ways. This application does not impose any restrictions on this method.

[0125] It should be understood that, such as Figure 1 The terminal device 200 shown can have a service application installed. When the service application runs on the terminal device 200, it can interact with the aforementioned... Figure 1 The business servers 100 shown interact with each other, enabling each business server 100 to receive business conversion data from the terminal device 200. This business application can be a client application with image, video, or other data information capabilities, such as a game application, video editing application, social application, instant messaging application, live streaming application, short video application, video application, music application, shopping application, novel application, payment application, or browser. This application client can be a standalone client or an embedded sub-client integrated into another client (such as an instant messaging client, social client, or video client); this is not limited here.

[0126] like Figure 1 As shown, the main object can upload a business conversion dataset in the business application via terminal device 200. By uploading the business conversion dataset, terminal device 200 can send a quality identification request containing the business conversion dataset to business server 100. Here, "conversion" is an action defined by the main object, which can include clicks, registrations, activations, payments, etc. Business conversion data represents the data showing that the object successfully achieves a specific action expected by the main object during the business media campaign. The main object refers to the object providing the business media. The format of business conversion data is typically <business media ID, object ID, conversion time, conversion behavior type>. Business media can include one or more media types, such as at least one of text media, image media, audio media, or video media. For example, business media can be a video advertisement.

[0127] For example, if object A clicks on the game business media 1 of the main object at 10:30 AM, the display interface of object A's corresponding terminal device will redirect to the game page indicated by game business media 1. At 10:45 AM, object A registers on that game page. Therefore, since the main object's terminal device 200 can detect conversion behaviors associated with game business media 1 in real time, when object A performs a conversion behavior targeting game business media 1, the main object's terminal device 200 can generate business conversion data based on the detected conversion behavior. This includes generating business conversion data 1 related to the click behavior, displayed as <1, A, 10.30, Click>, and business conversion data 2 related to the registration behavior, displayed as <1, A, 10.45, Register>. The terminal device 200 can generate business conversion data in real time. When the business conversion data meets the periodic upload conditions or the number of business conversion data reaches a threshold, the terminal device 200 can combine the generated multiple business conversion data into a business conversion dataset, which can then be sent to the business server 100. Optionally, the terminal device corresponding to the object performing the conversion behavior can also generate business conversion data related to the conversion behavior being performed, and then send the generated business conversion data to the terminal device 200 corresponding to the main object, so that the terminal device 200 can collect this business conversion data in real time.

[0128] The business server 100 can perform quality identification on the received business conversion dataset to obtain quality identification results associated with the business conversion data. It can be understood that the business server 100 can use business conversion datasets whose quality identification results meet the sample quality conditions as training samples and send them to downstream processing nodes. This allows the downstream processing nodes to train high-quality prediction models based on high-quality and accurate business conversion datasets, thereby improving the accuracy of predictions of conversion metrics. The prediction models in the downstream processing nodes can be used to estimate conversion metrics for any object performing conversion behavior on any business media, such as estimating the conversion rate.

[0129] Please see Figure 2 , Figure 2 This is a schematic diagram of a data processing scenario provided in an embodiment of this application. For example... Figure 2As shown, the terminal device 200 corresponding to the main object can generate business conversion data in real time and display the business conversion data in the display interface corresponding to the business application. For example, the business conversion data 1 can be displayed as <1, A, 10.30, click>, indicating that the business media id corresponding to the business conversion data 1 is 1, the corresponding object id is A, the conversion time of object A for business media 1 is 10:30, and the conversion behavior type of object A for business media 1 is click;..., the business conversion data n can be displayed as <r, S, 11.45, payment>, indicating that the business media id corresponding to the business conversion data n is r, the corresponding object id is S, the conversion time of object S for business media r is 11:45, and the conversion behavior type of object S for business media r is payment. When the business conversion data meets the periodic upload condition or the number of business conversion data reaches the quantity threshold, the terminal device 200 can combine the generated multiple business conversion data into a business conversion data set, and then can send the business conversion data set to the business server 100. Among them, the meaning and source of the business conversion data can be referred to the above Figure 1 Regarding the description of the business conversion data, it will not be elaborated here. Among them, the main object refers to the object that provides business media, such as an advertiser.

[0130] The business server 100 can perform quality identification on the received business conversion data set based on the conversion nodes of the business conversion data, and obtain a conversion quality result associated with the business conversion data set. Among them, the conversion nodes include shallow conversion nodes and deep conversion nodes. The main object can select one of the conversion nodes as the deep conversion node, and the shallow conversion node can be the node whose execution order is before the deep conversion node. For example, if a set of conversion nodes is set for a series of conversion behaviors: click conversion node, registration conversion node, activation conversion node, order placement conversion node, and payment conversion node, the main object can select the order placement conversion node as the deep conversion node, and then select the registration conversion node as the shallow conversion node.

[0131] The process by which the business server 100 generates conversion quality results can be as follows: Based on the received business conversion dataset, the conversion data volume of the conversion nodes of the business media is statistically analyzed to obtain the conversion quality results for the business conversion dataset. For example, the business server 100 can use a counting function to count the number 'a' of business conversion data that is concentrated in shallow conversion nodes and performs conversion actions in deep conversion nodes, and the number 'b' of business conversion data that is concentrated in deep conversion nodes. The conversion quality result corresponding to the business conversion dataset is obtained based on the ratio a / b. The larger the ratio a / b, the higher the accuracy of the received business conversion dataset. This is because the normal process executes shallow conversion nodes first and then deep conversion nodes. Therefore, for a normal business conversion dataset, for business conversion data that performs conversion actions in deep conversion nodes, the corresponding business conversion data that performs conversion actions in shallow conversion nodes should be found in the business conversion dataset (the two corresponding business conversion data have the same object and the same business media, and the conversion time is similar). If a / b is 1, it means that for each business conversion data that performs a conversion behavior in a deep conversion node, its corresponding business conversion data that performs a conversion behavior in a shallow conversion node can be found in the business conversion dataset. In other words, if a business conversion data that performs a conversion behavior in a deep conversion node cannot be found in the business conversion dataset with its corresponding business conversion data that performs a conversion behavior in a shallow conversion node, it means that the business conversion data that performed the conversion behavior in the deep conversion node may be fictitious or tampered with. Therefore, by analyzing the conversion quality results, it is possible to identify whether the main object tampered with the business conversion dataset before the data was sent back.

[0132] The business server 100 can further perform return quality identification on the business conversion dataset during the return process to obtain return quality results. These results can be used to identify whether the main object tampered with the business conversion dataset before return or to identify whether abnormal business conversion data occurred due to transmission failure in the business media system.

[0133] The business server 100 can send a business conversion dataset, in which both the conversion quality results and the return quality results meet the sample quality criteria, as training samples to downstream processing nodes. This allows the business server 100 to train a prediction model on the downstream processing nodes based on the accurate and high-quality business conversion dataset. For example, the downstream processing nodes can train a prediction model for estimating conversion rates using a business conversion dataset that meets the sample quality criteria.

[0134] It is understandable that the business server 100 can statistically analyze the conversion data volume of the conversion nodes of the business media using the business conversion dataset to obtain conversion statistics and generate conversion quality results for the business media. Furthermore, by identifying the return quality during the process of the main object sending back the business conversion dataset, a return quality result can be obtained. Through analysis of the conversion quality results and the return quality results, the business server can obtain a business conversion dataset where both the conversion quality results and the return quality results meet the sample quality conditions. This allows for the filtering out of abnormal business conversion data caused by main object tampering or business media system transmission failures. Therefore, by analyzing the conversion quality results and the return quality results, it is possible to ensure that accurate and high-quality business conversion datasets are used as training samples and transmitted to downstream processing nodes, enabling the downstream processing nodes to train high-quality prediction models and improve the accuracy of downstream processing nodes in predicting conversion metrics.

[0135] Please see Figure 3 , Figure 3 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 1 This data processing method can be executed by a computer device, which can be, for example, Figure 1 The business server 100 is shown. The following description uses the example of this data processing method being executed by a computer device. This data processing method may include at least the following steps S101-S104:

[0136] Step S101: Obtain the business conversion dataset; the business conversion dataset includes data on the conversion behaviors performed by the object on the advertised business media.

[0137] Specifically, the computer device can acquire the business conversion dataset returned by the master object. The business media can include one or more media types, such as at least one of text media, image media, audio media, or video media.

[0138] The computer equipment can use common interfaces, such as APIs (Application Programming Interfaces) and data file transfer interfaces, to access the business conversion dataset. It should also select efficient data transmission technologies to receive the dataset, such as Hypertext Transfer Protocol (HTTP), message queues, and RPC (Remote Procedure Call). Furthermore, the computer equipment can choose data storage technologies such as relational databases, non-relational databases, and distributed file systems to store the business conversion dataset; this embodiment does not impose any limitations on these methods. The meaning of the business conversion dataset, conversion behavior, and main object can be found above. Figure 1 The specific content of the business conversion dataset, "conversion", and main object in the corresponding embodiment will not be described again in this application embodiment.

[0139] Step S102: Perform conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset to obtain conversion statistics results, and generate conversion quality results for the business media based on the conversion statistics results;

[0140] Specifically, the business server 100 can perform quality identification on the received business conversion dataset based on the conversion nodes of the business conversion dataset to obtain conversion quality results associated with the business conversion dataset. Conversion nodes include shallow conversion nodes and deep conversion nodes. The main object can select one of the conversion nodes as a deep conversion node, and the shallow conversion node can be a node whose execution order precedes that of the deep conversion node. For example, if a set of conversion nodes is set for a series of conversion behaviors: click conversion node, registration conversion node, activation conversion node, order placement conversion node, and payment conversion node, the main object can select the order placement conversion node as the deep conversion node and then select the registration conversion node as the shallow conversion node. The execution order of the registration conversion node precedes that of the order placement conversion node.

[0141] Specifically, when the main object selects deep conversion nodes and shallow conversion nodes, the computer device counts the number of business conversion data that have already performed conversion actions in shallow conversion nodes and also in deep conversion nodes, as statistical result C. For example, there is a business conversion data c1 belonging to a shallow conversion node and a business conversion data c2 belonging to a deep conversion node. Business conversion data c1 and business conversion data c2 have the same business media ID and object ID, and their conversion times are similar. Therefore, business conversion data c2 can be included in statistical result C. On the other hand, the computer also counts the number of business conversion data that have already performed conversion actions in deep conversion nodes, as statistical result D. Statistical results C and D are defined as the conversion statistics obtained by counting the conversion data volume of conversion nodes for business media based on the business conversion dataset. The ratio of statistical result C to statistical result D is defined as the conversion quality result for the business media. When the conversion quality result is greater than or equal to the first proportion threshold of 0.95, it indicates that at least 95% of the business conversion data executed at deep conversion nodes can be matched with corresponding business conversion data executed at shallow conversion nodes in the business conversion dataset. This means that less than 5% of the business conversion data may have been fabricated or tampered with. Therefore, a higher conversion quality result indicates higher accuracy of the received business conversion dataset, and when the result is sufficiently high, the computer can determine that the conversion quality meets the sample quality criteria. Conversely, when the conversion quality result is less than the first proportion threshold of 0.95, the computer can determine that the conversion quality does not meet the sample quality criteria, indicating that the conversion quality of the business conversion dataset may be abnormal.

[0142] For example, in a live stream workflow, there are six conversion nodes: "Entering the Live Stream," "Watching the Live Stream," "Clicking on a Product," "Browsing the Product Details Page," "Placing an Order," and "Paying." If "Paying" is selected as the deep conversion node and "Clicking on a Product" as the shallow conversion node, a computer can count the number of business conversion data points where a conversion occurred at the "Clicking on a Product" node and also occurred at the "Paying" node, as statistical result E. Simultaneously, the computer can count the total number of business conversion data points where a conversion occurred at the "Paying" node, as statistical result F. The computer can represent the conversion quality result as G, defined as G = E / F. When G is greater than a first proportional threshold, the computer can determine that the conversion quality result G meets the sample quality condition. Therefore, the conversion quality result G can more intuitively represent the quality anomalies in the business conversion dataset.

[0143] Step S103: Perform backhaul quality identification on the backhaul process of the business conversion dataset to obtain backhaul quality results; the backhaul process of the business conversion dataset refers to the process of the main object backhauling the business conversion dataset; the main object refers to the object that provides the business media.

[0144] Optionally, the feedback quality results may include centralized feedback quality results, which may include month-on-month conversion results and year-on-year conversion results. The specific process of identifying feedback quality during the feedback process of the business conversion dataset to obtain feedback quality results may include: obtaining the first business conversion data belonging to the current time period within the current cycle, and the second business conversion data belonging to the full historical time period within each of N historical cycles, where N is a positive integer; the full historical time period includes multiple time periods used to evenly divide a historical cycle; obtaining the first business conversion value corresponding to the first business conversion data, obtaining the first conversion mean and first conversion standard deviation corresponding to the second business conversion data, and based on the first business conversion value, the first conversion mean, and the first conversion standard deviation... The standard deviation is used to generate month-on-month conversion results for the business conversion dataset; the third business conversion data within the same historical range time period in each of the N historical periods is obtained; the same historical range time period includes multiple range time periods adjacent to the current range time period in the unit time dimension; the second conversion mean and second conversion standard deviation corresponding to the third business conversion data are obtained, and the same conversion result for the business conversion dataset is generated based on the first business conversion value, the second conversion mean, and the second conversion standard deviation; the month-on-month conversion result and the same conversion result are determined as the centralized backhaul quality results for the business conversion dataset; if both the month-on-month conversion result and the same conversion result in the centralized backhaul quality results are less than the numerical threshold, then the centralized backhaul quality results are determined to meet the sample quality conditions.

[0145] For example, a computer device can acquire the first business conversion data within the current 5 minutes and its corresponding first business conversion value miu from a business conversion dataset. Here, the current 5 minutes represents the current time period within the current cycle, and the first business conversion value miu represents the quantity (i.e., conversion volume) of the first business conversion data. In the process of comparing month-on-month conversion results, the computer device can acquire the second business conversion data within each 5-minute period (a 5-minute period can be considered a time slice) over the past 7 days. Based on the second business conversion data, the computer device can calculate the business conversion volume for each time slice. Specifically, based on the quantity of second business conversion data within each of the 2016 time slices used to represent 5 minutes over the past 7 days, the computer device calculates the business conversion volume corresponding to each of the 2016 time slices. Then, it averages these 2016 business conversion volumes to obtain the first conversion mean, mean1. The computer equipment can calculate the difference between each business conversion and the first conversion mean (mean1), resulting in 2016 differences. Each difference is squared to obtain the square value corresponding to each difference. The 2016 square values ​​are then summed to obtain the total square value. The variance can be obtained from the total square value, and the first conversion standard deviation (sigma1) can be obtained by taking the square root of the variance. The computer equipment can define the month-on-month conversion result as (mean1 - mean1) / sigma1. Here, each 5-minute interval in the past 7 days can represent the full historical range time period in each historical period under N historical periods. The value of N is 7. A full historical range time period includes 288 time periods of 5 minutes each within one day. In the year-on-year comparison process to obtain year-on-year conversion results, the computer equipment can acquire third business conversion data within 5 minutes of each hour before and after the same time period in the past 7 days. Further, it calculates the second conversion mean (mean2) and the second conversion standard deviation (sigma2) based on the third business conversion data, thus defining the month-on-month conversion result as (mean2) / sigma2. The calculation methods for the second conversion mean (mean2) and the second conversion standard deviation (sigma2) are the same as those for the calculation of the first conversion mean (mean1) and the first conversion standard deviation (sigma1) mentioned above, and will not be repeated here. Each 5-minute interval within each hour before and after the same time period within a day can represent the historical range time period for year-on-year comparison, that is, multiple time periods adjacent to the current time period in a unit time dimension. For example, if the current time period is 5 minutes from 14:00 to 14:05, then taking yesterday as one of N historical cycles as an example, the 24 time slices representing 5 minutes within yesterday's 13:00 to 15:00 can represent multiple time periods adjacent to the current time period in a unit time dimension.When both the month-on-month and year-on-year conversion results are less than the numerical threshold of 3, i.e., (miu-mean1) / sigma1 < 3 and (miu-mean2) / sigma2 < 3, the computer equipment can determine that the centralized backhaul quality results meet the sample quality conditions. Since a sudden and significant increase in business conversion values ​​within a certain time period can lead to an overly large centralized backhaul quality result when the main object performs centralized backhaul of the business conversion dataset, it indicates that the business conversion dataset may have been delayed in its centralized backhaul. Such a business conversion dataset loses its timeliness, thus affecting its quality. Therefore, the smaller the centralized backhaul quality result, the closer the current business conversion dataset's backhaul situation is to the historical business conversion dataset's backhaul situation within a historical time period, i.e., the smaller the fluctuation, indicating that the main object has performed timely backhaul of the business conversion dataset, and also indicating that the business conversion dataset's quality is good. When (miu-mean1) / sigma1≥3 or (miu-mean2) / sigma2≥3, the computer equipment can determine that the centralized return quality results do not meet the sample quality conditions, which means that the business conversion dataset at this time may not have been returned in a timely manner.

[0146] Optionally, the feedback quality result may include a delayed feedback quality result. The specific process of identifying the feedback quality during the feedback process of the business conversion dataset to obtain the feedback quality result may include: obtaining the fourth business conversion data in the business conversion dataset that meets the first business dimension; calculating the proportion of business conversion data in the fourth business conversion data whose reporting delay is greater than or equal to a delay threshold; and determining the proportion as the delayed feedback quality result for the business conversion dataset; the reporting delay is determined based on the background receiving time of the conversion behavior associated with the fourth business conversion data and the occurrence time of the conversion behavior contained in the fourth business conversion data; if the delayed feedback quality result is less than the second proportion threshold, then it is determined that the delayed feedback quality result meets the sample quality condition.

[0147] Specifically, computer equipment can acquire fourth business conversion data within the business conversion dataset, satisfying the triple dimension of Optimization Objective (OG), Marketing Vehicle, and Traffic. OG is the core guide of the business conversion dataset, clearly defining the expected results for the business media. Marketing Vehicle refers to the means and channels through which the target audience performs conversion actions on the business media. Traffic represents the target audience's target area for the business media. Therefore, fourth business conversion data can represent business conversion data obtained by selecting a specific OG, placing ads through a marketing vehicle, and generating conversions within the target area indicated by the traffic within the business conversion dataset. For example, fourth business conversion data could be business conversion data placed in a designated M region via a mini-program under a specific OG. The computer equipment can monitor the timing patterns of the target audience's data feedback. By analyzing the time intervals and frequencies of historical feedback data, a latency threshold of 10 minutes is established under normal feedback mode. The proportion of business conversion data in the fourth business conversion dataset with a reported latency greater than or equal to the latency threshold is then statistically analyzed, and this proportion is determined as the latency feedback quality result for the business conversion dataset. The reporting delay is determined by the background reception time t1 of the conversion behavior associated with the fourth business conversion data and the occurrence time t2 of the conversion behavior included in the fourth business conversion data. The computer device can define the reporting delay as t1-t2. The delayed feedback quality result can be defined as the ratio K between the number of fourth business conversion data with a delay condition satisfying [(t1-t2)≥10] and the total number of fourth business conversion data. When the ratio K is less than a second ratio threshold, the computer device can determine that the delayed feedback quality result meets the sample quality condition. The specific value of the second ratio threshold can be 0.01. When the ratio K is greater than or equal to the second ratio threshold, for example, when the value of the ratio K is 0.1, the computer device can determine that the delayed feedback quality result does not meet the sample quality condition. By analyzing the delayed feedback quality result, the computer device can accurately measure the timeliness of the business conversion dataset, thus avoiding the inability to receive timely feedback on the conversion behavior performed by the object due to the use of business conversion data with high delay.

[0148] Optionally, the feedback quality results may include repeated feedback quality results. The specific process of identifying feedback quality during the feedback process of the business conversion dataset to obtain feedback quality results may include: obtaining the fifth business conversion data that satisfies the second business dimension in the business conversion dataset, and calculating the target daily conversion volume corresponding to the fifth business conversion data; obtaining the daily conversion volume of H business media domains under the second business dimension, generating the daily conversion mean and daily conversion standard deviation based on the H daily conversion volumes, generating a daily conversion volume threshold based on the daily conversion mean and daily conversion standard deviation, and determining the daily conversion volume threshold and the target daily conversion volume as the repeated feedback quality results for the business conversion dataset; if the target daily conversion volume in the repeated feedback quality results is less than the daily conversion volume threshold, then the repeated feedback quality results are determined to meet the sample quality conditions.

[0149] Specifically, the computer device can obtain the fifth business conversion data in the triple dimension of the optimization goal (OG), main object, and traffic in the business conversion dataset, and thereby count the target single-day conversion volume corresponding to the fifth business conversion data. Here, the main object refers to the object of the advertising business media. The content of OG and traffic can be referred to the specific descriptions of OG and traffic above, which will not be elaborated here. Therefore, the fifth business conversion data can represent the business conversion data obtained by selecting in the business conversion dataset, under this OG, being advertised by this main object, and generating conversion behaviors within the advertising area indicated by this traffic. Specifically, the computer device can extract the conversion time of the fifth business conversion data and define the single-day time range. For example, from 0:00 yesterday to 0:00 today is a single-day time range. The computer device can count the conversion volume of the fifth business conversion data within the single-day time range. This counting process can refer to counting the number of objects that have performed conversion behaviors under a certain conversion node in the fifth business conversion data within the single-day time range to obtain the target single-day conversion volume conv; or, it can also refer to counting the number of objects that have performed conversion behaviors under all conversion nodes in the fifth business conversion data within the single-day time range, that is, counting the number of different objects, so as to obtain the target single-day conversion volume conv. The computer device can obtain the single-day conversion volumes of H business media fields respectively in the triple dimension of the optimization goal (OG), main object, and traffic, and generate the single-day conversion mean mean3 and the single-day conversion standard deviation sigma3 based on the H single-day conversion volumes, and further obtain the single-day conversion volume threshold. Here, the calculation process of the H single-day conversion volumes can be the same as the calculation process of the above target single-day conversion volume conv, the calculation process of the single-day conversion mean mean3 can be the same as the calculation process of the above first conversion mean mean1, and the calculation process of the single-day conversion standard deviation sigma3 can be the same as the calculation process of the above first conversion standard deviation sigma1, which will not be elaborated here. Further, the computer device can define the single-day conversion volume threshold as mean3 + 2 * sigma3 + 1. The computer device can determine the single-day conversion volume threshold and the target single-day conversion volume as the repeated feedback quality result for the business conversion dataset. When the target single-day conversion volume is less than the single-day conversion volume threshold, that is, when conv < mean3 + 2 * sigma3 + 1, the computer device can determine that the repeated feedback quality result meets the sample quality condition; when the target single-day conversion volume is greater than or equal to the single-day conversion volume threshold, that is, conv ≥ mean3 + 2 * sigma3 + 1, the computer device can determine that the repeated feedback quality result does not meet the sample quality condition.It's understandable that when the target daily conversion rate is less than the daily conversion rate threshold, it means the target daily conversion rate can fluctuate within a relatively stable range within the threshold. In this case, it can be assumed that the business conversion dataset likely doesn't contain a large amount of repeatedly submitted business conversion data from the main object, meaning the quality of this business conversion dataset is considered relatively high. Using business conversion datasets with repeated submissions as training samples could significantly impact the accuracy of the trained prediction model. Therefore, analyzing the quality of repeated submissions can effectively ensure the quality of the business conversion dataset, thereby guaranteeing the accuracy of the subsequently trained prediction model.

[0150] Optionally, the feedback quality results may include erroneous feedback quality results. The specific process of identifying feedback quality during the feedback process of the business conversion dataset to obtain feedback quality results may include: in the business conversion dataset, obtaining business conversion data where the object has already performed conversion behavior on the landing page of the business media, as the sixth business conversion data; comparing the sixth business conversion data with the conversion behavior data for the landing page of the business media recorded by the business media platform to obtain a first comparison result; comparing the quantity of the sixth business conversion data with the quantity of conversion behavior data recorded by the business media platform to obtain a second comparison result; determining the first comparison result and the second comparison result as erroneous feedback quality results for the business conversion dataset; if the first comparison result indicates that the sixth business conversion data is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the quantity of the sixth business conversion data is equal to the quantity of conversion behavior data recorded by the business media platform, then the erroneous feedback quality results are determined to meet the sample quality conditions.

[0151] Specifically, the computer device can retrieve the sixth business conversion data from the business conversion dataset, showing that the object has already performed a conversion action on the landing page of the business media. This sixth business conversion data is then compared with the conversion action data for the landing page of the business media recorded by the business media platform. Specifically, the object, business media, and conversion action in a single sixth business conversion data set are compared with the conversion action data for the landing page of the business media recorded by the business media platform. This identifies sixth business conversion data sets and conversion action data sets that share the same object, business media, and conversion action. Next, it is determined whether the conversion time in these two sets of sixth business conversion data sets is consistent. If they are consistent, the two sets of sixth business conversion data sets and conversion action data sets are considered identical; otherwise, they are considered different. This process is repeated to identify all the difference comparison results between the sixth business conversion data sets and the conversion action data for the landing page of the business media recorded by the business media platform, and these results are then identified as the first comparison result. The computer device can also compare the number of sixth business conversion data sets with the number of conversion action data sets recorded by the business media platform to obtain the second comparison result. The first comparison result characterizes the degree of similarity between the sixth business conversion data and the conversion behavior data of the landing page for the business media recorded by the business media platform. The second comparison result characterizes the data completeness of the sixth business conversion data and the conversion behavior data of the landing page for the business media recorded by the business media platform. The computer device can determine the first and second comparison results as the error omission feedback quality results for the business conversion dataset. When the first comparison result indicates that the sixth business conversion data is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the quantity of the sixth business conversion data is equal to the quantity of the conversion behavior data recorded by the business media platform, then the error omission feedback quality results are determined to meet the sample quality conditions. This filters out erroneous business conversion data that cannot match the conversion behavior data recorded by the business media platform, as well as omission business conversion data whose quantity is not equal to the quantity of the conversion behavior data recorded by the business media platform. When the conversion data of the sixth business is not completely the same as the conversion behavior data recorded by the business media platform, or when the number of conversion data of the sixth business is not equal to the number of conversion behavior data recorded by the business media platform, the computer equipment can determine that the quality results of the error omission do not meet the sample quality conditions.

[0152] Step S104: If both the conversion quality result and the return quality result meet the sample quality conditions, the business conversion dataset is provided as a training sample to the downstream processing node; the downstream processing node is used to predict the conversion index for the conversion behavior.

[0153] It is understood that the feedback quality results can include at least one of the above-mentioned centralized feedback quality results, delayed feedback quality results, repeated feedback quality results, or erroneous feedback quality results. For example, if the feedback quality results include centralized feedback quality results, delayed feedback quality results, repeated feedback quality results, and erroneous feedback quality results, then when the computer equipment detects that the conversion quality result is greater than or equal to a preset first proportion threshold, the centralized feedback quality result is less than a preset numerical threshold, the delayed feedback quality result is less than a preset second proportion threshold, the daily conversion volume in the repeated feedback quality results is less than the daily conversion volume threshold, and the first comparison result in the erroneous feedback quality results indicates that the sixth business conversion data is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the quantity of the sixth business conversion data is equal to the quantity of conversion behavior data recorded by the business media platform, it can be determined that the conversion quality results and feedback quality results corresponding to the business conversion dataset both meet the sample quality conditions, and the computer equipment can send the business conversion dataset as a training sample to the downstream processing node. Downstream processing nodes can train prediction models using high-quality, filtered training samples, enabling the trained prediction models to accurately estimate conversion metrics for conversion behaviors.

[0154] Computer equipment can statistically analyze the conversion data volume at conversion nodes of business media. It can collect the number of business conversion data points that have already executed conversion actions at shallow conversion nodes and those at deep conversion nodes, as well as the number of conversion data points executed at deep conversion nodes, as conversion statistics. This data is then used to generate conversion quality results for the business media. Since the execution order of conversion actions at deep conversion nodes follows that at shallow conversion nodes, generally, business conversion data belonging to deep conversion nodes should also have corresponding data belonging to shallow conversion nodes. Therefore, the conversion quality results can effectively characterize the accuracy of the business conversion dataset. By identifying the return quality during the process of the main object returning the business conversion dataset, return quality results can be obtained. Based on these results, computer equipment can filter out business conversion datasets that have experienced concentrated, duplicate, delayed, or erroneously missed returns, ensuring that the business conversion datasets that can be used as training samples have high accuracy. Here, the main object refers to the object providing the business media. By analyzing the conversion quality results and the return quality results, we can obtain a business conversion dataset that meets the sample quality conditions. This allows us to filter out abnormal business conversion data caused by main object tampering or business media system transmission failures. Through the analysis of conversion quality results and return quality results, we can ensure that accurate and high-quality business conversion datasets are transmitted to downstream processing nodes as training samples, so that downstream processing nodes can train high-quality prediction models and improve the accuracy of downstream processing nodes in predicting conversion indicators.

[0155] Please see Figure 4 , Figure 4 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 2 This data processing method can be executed by a computer device, which can be, for example, Figure 1 The business server 100 is shown. The following description uses the example of this data processing method being executed by a computer device. This data processing method may include at least the following steps S201-S202:

[0156] Step S201: Input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node; extract features from the business conversion dataset through the conversion rate prediction model to obtain object features, business media features and object media interaction features; based on the object features, business media features and object media interaction features, generate the predicted probability of the object performing conversion behavior for the business media.

[0157] Specifically, the computer equipment can input a business conversion dataset (i.e., standard-compliant business conversion data) whose conversion quality results and feedback quality results both meet the sample quality conditions into the conversion rate prediction model in the conversion rate prediction node. The computer equipment can extract object features, business media features, and object-media interaction features from the business conversion dataset. Object features can include various attributes related to the object in the business conversion data, such as the object's consumption habits and historical interaction records. Business media features mainly describe the attributes of the business media associated with the business conversion data, such as the content theme and audience positioning of the business media. Object-media interaction features can include object-media statistical features and object long-term conversion sequence features. Object long-term conversion sequence features mainly reflect the interaction between the object and the business media it has interacted with, such as the object's browsing time, click count, and comment behavior on each business media. Object-media statistical features reflect the number of business media that the object has interacted with. Based on the obtained object features, business media features, and object-media interaction features, the computer equipment can generate a predicted probability of the object converting to the business media.

[0158] Please see also Figure 5 , Figure 5 This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 1 ,like Figure 5 As shown, the conversion rate prediction model can include a feature extraction layer, a feature concatenation layer, a feature interaction layer, and a fully connected layer. The computer can extract object features, business media features, and object-media interaction features from the business conversion dataset using a CNN (Convolutional Neural Network) or DNN (Deep Neural Network) model. This is represented by embedding, which transforms high-dimensional sparse feature vectors into low-dimensional dense vector representations. This helps the model better process and understand the features, improving model performance. The computer can then perform pooling operations on the extracted object features, business media features, and object-media interaction features, such as taking the average or maximum value, to obtain a fixed-length feature vector. This reduces the dimensionality of the features while retaining important information.

[0159] Specifically, in the feature concatenation layer, the computer device concatenates the object features, business media features, and object media interaction feature vectors, which have undergone embedding and pooling processing, using the Concat method. This concatenates the feature vectors by merging the number of feature channels, resulting in a higher-dimensional vector representation, thus providing richer input for the subsequent feature interaction layer. For example, if the object feature vector has a dimension of n1, the business media feature vector has a dimension of n2, and the object media interaction feature vector has a dimension of n3, after concatenation, the resulting feature vector has a dimension of n1 + n2 + n3. The computer device can perform concatenation along either the horizontal or vertical dimensions; this application does not impose any restrictions on this.

[0160] Specifically, in the feature interaction layer, the computer device can use a DCN (Deep & Cross Network) cross network to learn the high-order interaction relationships between object features, business media features, and object media interaction features. The process can be represented by formula (1):

[0161] x i+1 =x0x T i w i +b i +x i Formula (1)

[0162] Where, x i and x i+1 These are the output vectors of the i-th and (i+1)-th layers in the intermediate layers of the cross-network, respectively. x0 is the original input feature vector, formed by concatenating a low-dimensional dense vector (transformed from high-dimensional discrete features in the input data) with a continuous vector. The original input feature vector can be composed of object features, business media features, and object media interaction features. w i and b i These are the weights and biases of the i-th layer. Thus, by cross-processing the original input feature vector with the output vector of the current layer and adding linear terms, high-order interactive learning between features is achieved. In the deep layers of the feature interaction layer, the computer device can learn high-order representations of features through multiple layers of nonlinear transformations, thereby capturing complex nonlinear relationships. Each layer of the deep layer consists of linear transformations and nonlinear activation functions, such as the common activation function ReLU (Rectified Linear Unit). By continuously performing linear transformations and nonlinear activations, the deep layer can extract more abstract and higher-level feature representations, improving prediction accuracy.

[0163] Specifically, the fully connected layer receives the output from the feature interaction layer and uses a series of linear transformations and non-linear activation functions to make the final prediction. The fully connected layer maps the learned object features, business media features, and object-media interaction features to the output space, which is the predicted probability that the object will perform a conversion action in response to the business media. For example, after processing by the fully connected layer, the computer outputs a value between 0 and 1, representing the probability that the object will convert in response to the business media. If the output value is close to 1, it means the object is very likely to convert; if the output value is close to 0, it means the object is very unlikely to convert.

[0164] Step S202: Generate a first loss value based on the predicted probability and the actual conversion rate corresponding to the business conversion dataset. Adjust the model parameters of the conversion rate prediction model based on the first loss value to obtain a target conversion rate prediction model for predicting the probability of the target object converting to business media that have not been advertised.

[0165] Specifically, the computer device can generate a first loss value based on the predicted probability generated by the fully connected layer in step S201 and the actual conversion rate corresponding to the business conversion dataset, using a loss function. For example, the computer device can use the MSE (Mean Squared Error) loss function to calculate the difference between the predicted probability and the actual conversion rate, and the process can be expressed by formula (2):

[0166]

[0167] Where n is the number of samples, y j It is the actual conversion rate. This is the predicted probability. The computer calculates the square of the difference between the actual conversion rate and the predicted probability for each business conversion data point. Then, it sums up all the business conversion data and takes the average. The result is the first loss value. The first loss value reflects the gap between the model's prediction and the actual situation. The smaller the loss value, the more accurate the model's prediction.

[0168] Specifically, computer devices can use optimization algorithms to adjust the model parameters of the conversion rate prediction model based on the first loss value. For example, taking stochastic gradient descent as an example, its basic idea is to update the model parameters along the negative gradient direction of the loss function, so that the first loss value gradually decreases. For the model parameter m, the method of updating the model parameters can be defined as follows: Where α is the learning rate. This is the gradient of the loss function with respect to the parameters. By continuously iterating and updating the parameters, the conversion rate prediction model gradually approaches the optimal solution, making the gap between the predicted probability and the actual conversion rate smaller and smaller until convergence is achieved. Thus, computer equipment can obtain a target conversion rate prediction model for predicting the probability of a target audience converting to unreleased media, improving the accuracy of predicting the probability of conversion to unreleased media.

[0169] It is understandable that computer equipment can input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node to extract object features, business media features, and object-media interaction features from the business media dataset. This allows for a more comprehensive understanding of the object and business media situation. The feature embedding layer transforms high-dimensional sparse features into low-dimensional dense vector representations, making the conversion rate prediction model easier to process and understand, while also capturing the potential relationships between features. The feature interaction layer using a DCN structure can learn high-order interaction relationships between features. The fully connected layer maps the learned feature representations to the output space, enabling a more intuitive generation of predicted probabilities of objects performing conversion behaviors for business media. Because the business conversion dataset has sufficient accuracy and reliability, by predicting probabilities and the corresponding actual conversion rates, a more accurate first loss value can be generated to adjust the model parameters of the conversion rate prediction model, improving the accuracy of the conversion rate prediction model in predicting conversions from un-deployed business media, thus reducing the prediction bias of the conversion rate prediction model by 42%.

[0170] Please see Figure 6 , Figure 6 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 3 This data processing method can be executed by a computer device, which can be, for example, Figure 1 The business server 100 is shown. The following description will use the example of this data processing method being executed by a computer device. This data processing method will at least include the following steps S301-S303:

[0171] Step S301: Determine the business conversion data in the business conversion dataset that has already performed conversion behavior in the deep conversion node as training labels, and determine the business conversion data in the business conversion dataset that has already performed conversion behavior in the shallow conversion node and is associated with the training labels as shallow training samples. Input the exposure click data, training labels and shallow training samples of the business media platform for business media into the return flow prediction model in the cost control node.

[0172] The computer device can use business conversion data that has already performed conversion actions in deep conversion nodes within the business conversion dataset as deep conversion data, and determine the deep conversion data as training labels, thereby marking them as the training target values ​​for the repatriation prediction model. The computer device can also use business conversion data that has already performed conversion actions in shallow conversion nodes and is associated with the training labels within the business conversion dataset as shallow conversion data, and determine the shallow conversion data as shallow training samples. The meanings of deep conversion nodes and shallow conversion nodes are explained above. Figure 2 The specific descriptions of deep conversion nodes and shallow conversion nodes in the corresponding embodiments will not be repeated here.

[0173] Please see also Figure 7 , Figure 7 This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 2 ,like Figure 7 As shown, the computer equipment can integrate business media domains, business optimization goals, and link data. It inputs exposure click data, training tags, and shallow training samples into the return flow prediction model at the cost control node. Additionally, it can input data indicating the business media domain and business optimization goals into the return flow prediction model, enabling it to predict conversion return flow for business media and forecast the probability of deep conversions corresponding to shallow training samples—that is, the probability of executing a conversion behavior at a deep conversion node. This allows the target entity to effectively control the cost invested in business media. Exposure click data comes from the business media platform and reflects the target entity's initial contact and response to the business media. Training tags and shallow training samples come from the business conversion dataset. Since the shallow training samples are known, the computer equipment can optimize the original simple conversion return flow prediction problem from exposure clicks to predicting the probability of deep conversions into obtaining the probability of deep conversions based on exposure click data and shallow conversion samples. The computer equipment can analyze the target conversion process in more detail, improving the accuracy of conversion return flow prediction.

[0174] Step S302: In the backflow prediction model, extract the shallow conversion features corresponding to the shallow training samples and the exposure click features of the exposure click data; predict the deep conversion prediction probability corresponding to the shallow training samples based on the shallow conversion features and the exposure click features; and generate a second loss value based on the deep conversion prediction probability and the training labels.

[0175] Specifically, computer equipment can extract features from shallow training samples using a CNN network model to obtain shallow conversion features corresponding to the shallow training samples, and also extract features from exposure and click data to obtain exposure and click features. These exposure and click features can reflect the exposure and click situation of an object for the business media through data such as exposure count, click count, and click time. Then, based on the shallow conversion features and exposure and click features, the probability of predicting deep conversions corresponding to the shallow training samples can be predicted. Optionally, target data features used to indicate the business media domain and business optimization goals can also be extracted simultaneously. Then, based on these target data features, shallow conversion features, and exposure and click features, the probability of predicting deep conversions corresponding to the shallow training samples can be predicted.

[0176] Optional, such as Figure 7 As shown, computer devices can use the Weibull distribution method in a hierarchical statistical model to predict the deep conversion prediction probability corresponding to shallow training samples. The probability density function of the Weibull distribution is shown in equation (3):

[0177]

[0178] Here, t represents the input data (including shallow conversion features and exposure-click features, and may also include the aforementioned target data features), k is a shape parameter that affects the shape of the distribution, and λ is a scale parameter that affects the scale of the distribution. The initial shape parameter k and scale parameter λ are estimated using the input data t. Then, based on the input data t, the initial shape parameter k, and the scale parameter λ, the deep conversion prediction probability f(t) is generated. Based on the deep conversion prediction probability f(t) and the training labels, the loss value can be calculated, and the shape parameter k and scale parameter λ are adjusted based on the loss value until the model converges. The scale parameter λ determines the scale of the Weibull distribution. A larger λ value makes the distribution wider, simplifying the structure of the hierarchical statistical model and reducing the accuracy of the deep conversion prediction probability f(t). A smaller λ value makes the distribution more concentrated, making the structure of the hierarchical statistical model more complex and improving the accuracy of the deep conversion prediction probability f(t).

[0179] Specifically, the computer device can generate a second loss value based on the deep transformation prediction probability and training labels. For ease of understanding, the cross-entropy loss function is used as an example to illustrate the process of calculating the model loss value L.

[0180]

[0181] in, This is the deep conversion prediction probability generated by the reversion prediction model. y represents the prediction probability of the reversion prediction model corresponding to the training label. In binary classification, it can correspond to 0 or 1. When the training label is positive (i.e., the training sample associated with this training label is a positive sample), y = 1, and the loss value is... This means that when the deep transformation prediction probability The closer the value is to 1, the closer the loss value L is to 0, and the more accurate the prediction of the backflow prediction model; when the training label is negative (i.e., the training sample associated with the training label is a negative sample), y = 0, and the loss value is... The computer equipment can calculate the average of the model loss value K corresponding to all training samples to obtain the second loss value, which reflects the prediction accuracy of the backflow prediction model for the probability of deep conversion under the current parameters.

[0182] Step S303: Adjust the model parameters of the reflux prediction model according to the second loss value to obtain the reflux prediction model for predicting the deep conversion prediction probability of the target conversion business data; the target conversion business data refers to the conversion business data that has already performed conversion behavior in the shallow conversion node.

[0183] The computer equipment can adjust the model parameters of the backflow prediction model using optimization algorithms based on the second loss value. For example, taking stochastic gradient descent as an example, its basic idea is to update the model parameters along the negative gradient direction of the loss function, so that the second loss value gradually decreases. For the shape parameter k, the method for updating the model parameters can be defined as follows: For the scaling parameter λ, the method for updating the model parameters can be defined as follows: Where α is the learning rate. It is the gradient of the loss function with respect to the parameters. By iteratively updating the shape parameter k and the scale parameter λ along the opposite direction of the gradient, the second loss value is continuously reduced. When the second loss value is lower than the preset loss threshold, the computer device can determine that the shape parameter k and the scale parameter λ have converged. Thus, a target backflow prediction model can be obtained to predict the probability of an object performing a deep conversion behavior based on the business media and the object that has performed shallow conversion behavior.

[0184] Optionally, computer devices can also predict deep conversion probabilities using deep neural networks. This involves training the deep neural network with training labels and shallow conversion samples. During training, the computer device can predict the deep conversion probability corresponding to the training samples using exposure-click features and shallow conversion features. It then calculates the loss based on the training labels and deep conversion probability, generating a loss function to measure the difference between the deep conversion probability output by the deep neural network and the actual conversion. Optimization algorithms, such as stochastic gradient descent, can be used to continuously adjust the network's weights and biases to minimize the loss function. Through multiple iterative training iterations, the deep neural network gradually learns the mapping relationship between shallow conversion features and deep conversion probability, thus obtaining a backflow prediction model for predicting the deep conversion probability of target conversion business data.

[0185] It is understandable that computer equipment can identify business conversion data that has already performed conversion behavior in deep conversion nodes within the business conversion dataset as training labels, and identify business conversion data that has already performed conversion behavior in shallow conversion nodes associated with the training labels as shallow training samples. The exposure and click data for business media from the business media platform, the training labels, and the shallow training samples are then input into the return flow prediction model in the cost control node. This allows the return flow prediction model to extract the shallow conversion features corresponding to the shallow training samples and the exposure and click features of the exposure and click data. Computer equipment can introduce shallow conversion nodes into the Weibull distribution method, decomposing the original conversion return flow prediction problem "from exposure and click to deep conversion node" into a two-stage conversion return flow prediction problem "from click to shallow node, and then to deep conversion node." This allows for the prediction of the deep conversion probability corresponding to the shallow training samples, thereby refining the analysis of the conversion return flow prediction problem and improving the accuracy of conversion return flow prediction. The computer equipment can generate a second loss value based on the deep conversion prediction probability and training labels, and adjust the model parameters of the reflux prediction model according to the second loss value. The model is continuously optimized until a reflux prediction model for predicting the deep conversion prediction probability of target conversion business data is obtained. This provides the main object with an accurate deep conversion prediction probability and increases the cost achievement rate of the investment in conversion reflux prediction by 7.7%. Here, target conversion business data refers to conversion business data that has already performed conversion behavior in shallow conversion nodes.

[0186] Please see Figure 8 , Figure 8 This is a flowchart illustrating a data processing method provided in an embodiment of this application. Figure 4 This data processing method can be executed by a computer device, which can be, for example, Figure 1The business server 100 is shown. The following description uses the example of this data processing method being executed by a computer device. This data processing method may include at least the following steps S401-S403:

[0187] Step S401: Input the new business media into the cold start exploration node, and estimate the conversion rate of the new business media for the shallow conversion node based on the cold start exploration pool A to obtain the shallow conversion probability of the new business media; new business media refers to business media that have not been launched.

[0188] Please see also Figure 9 , Figure 9 This is a schematic diagram of a model structure provided in an embodiment of this application. Figure 3 ,like Figure 9 As shown, the cold start exploration nodes include Cold Start Exploration Pool A and Cold Start Exploration Pool B. Cold Start Exploration Pool A is a network model trained using business conversion data from the business conversion dataset that has already executed conversion behaviors in shallow conversion nodes, used to explore front-end goals. Cold Start Exploration Pool B is a network model trained using business conversion data from the business conversion dataset that has already executed conversion behaviors in deep conversion nodes, used to explore optimization goals, i.e., to explore high-quality target audiences (such as high-quality paying users). The meanings of deep conversion nodes and shallow conversion nodes can be found above. Figure 2 The specific descriptions of deep conversion nodes and shallow conversion nodes in the corresponding embodiments will not be repeated here.

[0189] Specifically, the computer device can input unlaunched business media, i.e., new business media, into the cold start exploration node. Based on the cold start exploration pool A within the cold start exploration node, the computer device can predict the conversion rate of the new business media for shallow conversion nodes. The cold start exploration pool A can be a network model trained on relevant data from shallow conversion nodes. This relevant data can include performance data of existing similar business media at shallow conversion nodes, target audience behavior characteristics, market trends, etc. The computer device can extract features from the new business media based on the cold start exploration pool A and use these extracted features as input data for the cold start exploration pool A. This allows the cold start exploration pool A to predict the conversion rate of the new business media for shallow conversion nodes with multiple different user group types, obtaining the shallow conversion probability for each user group type. For example, predicting the shallow conversion probability of the new business media for the elderly user group and predicting the shallow conversion probability for the athlete user group.

[0190] Step S402: If the shallow conversion probability of the new business media is less than the shallow exploration probability threshold, then stop the cold start exploration of the new business media; if the shallow conversion probability of the new business media is greater than or equal to the shallow exploration probability threshold, then conduct trial deployment of the new business media and obtain the trial business conversion volume corresponding to the new business media during the trial deployment period.

[0191] Specifically, the shallow exploration probability threshold is a pre-set value determined based on historical business conversion data or specific targets. It can be used to measure whether the performance of a new business media at the shallow conversion node is worth further exploration. For example, if the shallow conversion probability of new business media 1 under user group type Z is less than the shallow exploration probability threshold, it can be determined that the front-end cold start of new business media 1 for user group type Z has failed. The computer equipment can reduce the resources invested in new business media 1 in user group type Z to reduce ineffective exploration, thereby reducing cost investment that cannot bring the expected return. When the shallow conversion probability of new business media 1 for user group type V is greater than or equal to the shallow exploration probability threshold, it can be determined that the front-end cold start of new business media 1 for user group type V has succeeded. Then, the computer equipment can trial-deploy new business media 1 in the actual business media field (such as trial deployment to users of user group type V) within a certain time period, such as within a week, and count the user's reaction and behavior to obtain the trial business conversion volume of new business media 1 in user group type V. The trial business conversion volume can include the number of specific conversion behaviors such as clicks and registrations by users for new business media 1.

[0192] Step S403: If the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, then based on the cold start exploration pool B, continue to estimate the conversion rate of the new business media for deep conversion nodes to obtain the deep conversion probability of the new business media; if the deep conversion probability is greater than or equal to the deep exploration probability threshold, then the new business media will be officially launched.

[0193] Specifically, when the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, the computer equipment can continue to estimate the conversion rate of the new business media for deep conversion nodes based on the cold start exploration pool B, and obtain the deep conversion probability of the new business media. The quantity threshold is a pre-set standard value based on business experience, historical data, or specific business objectives, used to evaluate whether the performance of the new business media in the trial deployment phase meets the requirements of deep conversion analysis. The cold start exploration pool B can be a network model trained based on the performance data of successfully deployed business media in terms of deep conversion, the behavioral characteristics of the target audience in the deep conversion stage, market trends, and other information. For example, for a new business media 1 whose trial conversion rate is greater than or equal to a certain threshold, the computer equipment can enhance the optimization target exploration of the new business media 1 based on the cold start exploration pool B. It can extract features from the cold start exploration pool B and use these extracted features as input data for the cold start exploration pool B. Based on this input data, the cold start exploration pool B can predict the conversion rate of the new business media 1 for deep conversion nodes under user group type V (since it has been previously determined that the new business media 1 has successfully cold-started in user group type V), thus obtaining the deep conversion probability of the new business media 1 in user group type V. When the deep conversion probability of the new business media 1 is greater than or equal to the deep exploration probability threshold, the computer equipment can identify users in user group type V as the optimization target of the new business media 1. That is, it can begin formal deployment processing for users of the new business media 1 in user group type V, meaning the optimization target cold start is successful. The deep exploration probability threshold can be a pre-set value determined based on historical business conversion data or a specific target audience, and can be used to measure whether the new business media is worth formal deployment.

[0194] It is understandable that the computer equipment can input new business media into the cold start exploration node. Based on the cold start exploration pool A, the conversion rate of the new business media for shallow conversion nodes is estimated, resulting in the shallow conversion probability of the new business media. Here, new business media refers to business media that has not yet been launched. When the shallow conversion probability of a new business media is less than the shallow exploration probability threshold, the computer equipment can stop the cold start exploration of the new business media, thereby filtering out new business media that cannot be successfully cold-started at shallow conversion nodes and reducing the waste of cold start exploration resources. When the shallow conversion probability of a new business media is greater than or equal to the shallow exploration probability threshold, the computer equipment can conduct trial launches of the new business media, obtaining the trial business conversion volume corresponding to the new business media during the trial launch period. When the trial business conversion volume corresponding to the new business media is greater than or equal to the quantity threshold, the computer equipment continues to estimate the conversion rate of the new business media for deep conversion nodes based on the cold start exploration pool B, obtaining the deep conversion probability of the new business media, thereby filtering out new business media that failed in trial launches and reducing the waste of launch resources. If the deep conversion probability is greater than or equal to the deep exploration probability threshold, the computer equipment can proceed with the formal launch of the new business media. This can improve the exploration efficiency of deep transformation nodes, reduce the waste of cold start exploration resources, and reduce the support costs invested in cold start exploration by 57.6%.

[0195] Please see Figure 10 , Figure 10 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application. Figure 1 .like Figure 10 As shown, the data processing device includes a transceiver module 1100, a first processing module 1200, a second processing module 1300, a training module 1400, and a cold start module 1500.

[0196] The transceiver module 1100 is used to acquire the business conversion dataset; the business conversion dataset includes data on the conversion behaviors performed by the object on the deployed business media; it is also used to provide the business conversion dataset as training samples to the downstream processing node if both the conversion quality results and the feedback quality results meet the sample quality conditions; the downstream processing node is used to predict the conversion metrics for the conversion behaviors.

[0197] The first processing module 1200 is used to perform conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset, obtain conversion statistics results, and generate conversion quality results for the business media based on the conversion statistics results.

[0198] The second processing module 1300 is used to identify the return quality of the business conversion dataset during the return process and obtain the return quality result; the return process of the business conversion dataset refers to the process of the main object returning the business conversion dataset; the main object refers to the object that provides the business media;

[0199] In one possible implementation, the conversion statistics result includes a first statistical result and a second statistical result, and the conversion nodes include shallow conversion nodes and deep conversion nodes; the first processing module 1200 is used to perform conversion data volume statistics on the conversion nodes of the business media according to the business conversion dataset to obtain conversion statistics results. When generating conversion quality results for the business media based on the conversion statistics results, it is specifically used to perform the following operations:

[0200] In the business conversion dataset, the number of business conversion data that have already executed conversion actions in shallow conversion nodes and have also executed conversion actions in deep conversion nodes is counted as the first statistical result; the execution order of business media executing conversion actions in shallow conversion nodes is placed before the execution order of conversion actions executing in deep conversion nodes;

[0201] In the business conversion dataset, the number of business conversion data that have executed conversion behaviors in deep conversion nodes is counted as a second statistical result;

[0202] The ratio of the first statistical result to the second statistical result is determined as the conversion quality result for business media.

[0203] The first processing module 1200 is also used to perform the following operations:

[0204] If the transformation quality result is greater than or equal to the first proportion threshold, then the transformation quality result is determined to meet the sample quality condition.

[0205] In one possible implementation, the feedback quality results include centralized feedback quality results; the second processing module 1300 is used to identify the feedback quality of the feedback process of the business conversion dataset, and when obtaining the feedback quality results, it is specifically used to perform the following operations:

[0206] Retrieve the first business conversion data belonging to the current time range within the current period in the business conversion dataset, and the second business conversion data belonging to the full historical range time range in each of N historical periods, where N is a positive integer; the full historical range time range includes multiple time ranges used to evenly divide a historical period.

[0207] Obtain the first business conversion value corresponding to the first business conversion data, obtain the first conversion mean and first conversion standard deviation corresponding to the second business conversion data, and generate the month-on-month conversion result for the business conversion dataset based on the first business conversion value, the first conversion mean, and the first conversion standard deviation;

[0208] Obtain the third business conversion data in the business conversion dataset that belongs to the same historical range time period within each of the N historical periods; the same historical range time period includes multiple range time periods that are adjacent to the current range time period in the unit time dimension;

[0209] Obtain the second conversion mean and second conversion standard deviation corresponding to the third business conversion data, and generate year-on-year conversion results for the business conversion dataset based on the first business conversion value, the second conversion mean, and the second conversion standard deviation;

[0210] The month-on-month conversion results and year-on-year conversion results are determined as the centralized quality results for the business conversion dataset;

[0211] The second processing module 1300 is also used to perform the following operations:

[0212] If both the month-on-month conversion result and the year-on-year conversion result in the centralized back-transmission quality results are less than the numerical threshold, then the centralized back-transmission quality results are determined to meet the sample quality conditions.

[0213] In one possible implementation, the return quality result includes a delayed return quality result; the second processing module 1300 is used to identify the return quality of the business conversion dataset during the return process, and when obtaining the return quality result, it is specifically used to perform the following operations:

[0214] Obtain the fourth business conversion data that meets the first business dimension from the business conversion dataset, and calculate the proportion of business conversion data in the fourth business conversion data whose reporting delay is greater than or equal to the delay threshold. Determine the proportion as the delay feedback quality result for the business conversion dataset. The reporting delay is determined based on the background receiving time of the conversion behavior associated with the fourth business conversion data and the occurrence time of the conversion behavior contained in the fourth business conversion data.

[0215] The second processing module 1300 is also used to perform the following operations:

[0216] If the delayed return quality result is less than the second proportional threshold, then the delayed return quality result is determined to meet the sample quality condition.

[0217] In one possible implementation, the feedback quality results include repeated feedback quality results; the second processing module 1300 is used to identify the feedback quality of the business conversion dataset feedback process, and when obtaining the feedback quality results, it is specifically used to perform the following operations:

[0218] Obtain the fifth business conversion data that meets the second business dimension from the business conversion dataset, and calculate the target daily conversion volume corresponding to the fifth business conversion data.

[0219] Obtain the daily conversion volume of H business media domains under the second business dimension. Generate the daily conversion mean and daily conversion standard deviation based on the H daily conversion volumes. Generate the daily conversion volume threshold based on the daily conversion mean and daily conversion standard deviation. Determine the daily conversion volume threshold and the target daily conversion volume as the repeated feedback quality result for the business conversion dataset.

[0220] The second processing module 1300 is also used to perform the following operations:

[0221] If the target daily conversion rate in the repeated quality return results is less than the daily conversion rate threshold, then the repeated quality return results are determined to meet the sample quality conditions.

[0222] In one possible implementation, the feedback quality results include error and omission feedback quality results; the second processing module 1300 is used to identify the feedback quality of the business conversion dataset feedback process, and when obtaining the feedback quality results, it is specifically used to perform the following operations:

[0223] In the business conversion dataset, retrieve the business conversion data where the object has performed a conversion action on the landing page of the business media, and use it as the sixth business conversion data;

[0224] The sixth business conversion data is compared with the conversion behavior data of the landing page for business media recorded by the business media platform to obtain the first comparison result;

[0225] The second comparison result is obtained by comparing the number of conversion data of the sixth business with the number of conversion behavior data recorded by the business media platform;

[0226] The first and second comparison results are determined as the quality results of error omissions in the business conversion dataset.

[0227] The second processing module 1300 is also used to perform the following operations:

[0228] If the first comparison result indicates that the conversion data of the sixth business is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the number of conversion data of the sixth business is equal to the number of conversion behavior data recorded by the business media platform, then it is determined that the quality result of the error omission feedback meets the sample quality condition.

[0229] In one possible implementation, the downstream processing node includes a conversion rate prediction node; the data processing device also includes a training module 1400, which is used to perform the following operations:

[0230] Input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node;

[0231] The conversion rate prediction model extracts features from the business conversion dataset to obtain object features, business media features, and object media interaction features. Based on the object features, business media features, and object media interaction features, the predicted probability of the object performing conversion behavior in response to the business media is generated.

[0232] A first loss value is generated based on the predicted probability and the actual conversion rate corresponding to the business conversion dataset. The model parameters of the conversion rate prediction model are then adjusted based on the first loss value to obtain a target conversion rate prediction model used to predict the probability of the target object converting to business media that have not been advertised.

[0233] In one possible implementation, the downstream processing node includes a cost control node; the data processing device also includes a training module 1400, which is used to perform the following operations:

[0234] The business conversion data that has already performed conversion behavior in the deep conversion node in the business conversion dataset is determined as the training label, and the business conversion data that has already performed conversion behavior in the shallow conversion node and is associated with the training label in the business conversion dataset is determined as the shallow training sample. The exposure click data, training label and shallow training sample of the business media platform for business media are input into the return flow prediction model in the cost control node.

[0235] In the backflow prediction model, shallow conversion features corresponding to shallow training samples and exposure click features of exposure click data are extracted.

[0236] Based on shallow conversion features and exposure-click features, predict the deep conversion prediction probability corresponding to shallow training samples, and generate a second loss value based on the deep conversion prediction probability and training labels.

[0237] The model parameters of the reflux prediction model are adjusted based on the second loss value to obtain a reflux prediction model for predicting the deep conversion probability of target conversion business data; target conversion business data refers to conversion business data that has already performed conversion behavior in shallow conversion nodes.

[0238] In one possible implementation, the downstream processing nodes include cold start exploration nodes; the cold start exploration nodes include cold start exploration pool A and cold start exploration pool B; cold start exploration pool A is a network model trained on business conversion data that has already performed conversion actions in shallow conversion nodes in the business conversion dataset, and cold start exploration pool B is a network model trained on business conversion data that has already performed conversion actions in deep conversion nodes in the business conversion dataset; the data processing device also includes a cold start module 1500, which is used to perform the following operations:

[0239] Input the new business media into the cold start exploration node, and estimate the conversion rate of the new business media for the shallow conversion node based on the cold start exploration pool A to obtain the shallow conversion probability of the new business media; new business media refers to business media that have not been launched.

[0240] If the shallow conversion probability of the new business media is less than the shallow exploration probability threshold, then stop the cold start exploration of the new business media.

[0241] If the shallow conversion probability of the new business media is greater than or equal to the shallow exploration probability threshold, the new business media will be put on trial and the corresponding trial business conversion volume of the new business media during the trial period will be obtained.

[0242] If the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, then based on the cold start exploration pool B, the conversion rate of the new business media for deep conversion nodes will be estimated to obtain the deep conversion probability of the new business media.

[0243] If the probability of deep conversion is greater than or equal to the probability threshold of deep exploration, then the new business media will be officially launched.

[0244] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0245] This application embodiment records the conversion behavior performed by an object on the deployed business media as business conversion data, and records the collected business conversion data as a business conversion dataset. By statistically analyzing the conversion data volume of the conversion nodes of the business media using the business conversion dataset, conversion statistics can be obtained, thereby generating conversion quality results for the business media. Furthermore, by identifying the return quality of the business conversion dataset from the main object, a return quality result can be obtained; where the main object refers to the object providing the business media. By analyzing the conversion quality results and the return quality results, a business conversion dataset that meets the sample quality conditions can be obtained. This can filter out abnormal business conversion data caused by tampering with the main object or transmission failure of the business media system. Therefore, by analyzing the conversion quality results and the return quality results, it can be ensured that accurate and high-quality business conversion datasets are transmitted as training samples to downstream processing nodes, so that the downstream processing nodes can train high-quality prediction models and improve the accuracy of downstream processing nodes in predicting conversion indicators.

[0246] Please see Figure 11 , Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 11 As shown, the computer device 1000 may include a processor 1001, a network interface 1004, and a memory 1005. Furthermore, the computer device 1000 may also include an object interface 1003 and at least one communication bus 1002. The communication bus 1002 is used to enable communication between these components. The object interface 1003 may include a display screen and a keyboard; optionally, the object interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1005 may also be at least one storage device located remotely from the processor 1001. Figure 11 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, an object interface module, and a device control application.

[0247] In such Figure 11 In the computer device 1000 shown, the network interface 1004 can provide network communication elements; the object interface 1003 is mainly used to provide an input interface for objects; and the processor 1001 can be used to call the device control application stored in the memory 1005.

[0248] When the computer device 1000 is a data processing device, it achieves the following:

[0249] Obtain the business conversion dataset; the business conversion dataset includes data on the conversion behaviors performed by the object against the business media that have been launched;

[0250] Based on the business conversion dataset, the conversion data volume of the conversion nodes of the business media is statistically analyzed to obtain conversion statistics results. Based on the conversion statistics results, conversion quality results for the business media are generated.

[0251] The process of transmitting the business conversion dataset back is used to identify the transmission quality and obtain the transmission quality results. The transmission process of the business conversion dataset refers to the process of transmitting the business conversion dataset back by the master object. The master object refers to the object that provides the business media.

[0252] If both the conversion quality results and the feedback quality results meet the sample quality conditions, the business conversion dataset will be provided as training samples to the downstream processing nodes; the downstream processing nodes will then be used to predict conversion metrics for conversion behaviors.

[0253] It should be understood that the computer device 1000 described in the embodiments of this application can execute the foregoing text. Figure 3 , Figure 4 , Figure 6 and Figure 8 The description of the data processing method in any corresponding embodiment will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0254] Furthermore, it should be noted that this application embodiment also provides a computer-readable storage medium, which stores a computer program. When the processor executes the computer program, it can execute the aforementioned... Figure 3 , Figure 4 , Figure 6 and Figure 8 The description of the data processing method in any corresponding embodiment is already provided, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application.

[0255] The aforementioned computer-readable storage medium can be an internal storage unit of the data processing apparatus or computer device provided in any of the foregoing embodiments, such as a hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store data that has been displayed or will be displayed.

[0256] Furthermore, it should be noted that this application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. The processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the aforementioned... Figure 3 , Figure 4 , Figure 6 and Figure 8 The method provided in any of the corresponding embodiments.

[0257] The terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0258] 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, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the foregoing description as a network element. Whether these network elements are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can implement the described network elements using different methods for each specific application, but such implementation should not be considered beyond the scope of this application.

[0259] The methods and related apparatus provided in this application are described with reference to the method flowcharts and / or structural diagrams provided in this application. Specifically, each block of the method flowchart and / or structural diagram, as well as combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable device to create a machine, such that the instructions, which execute via the processor of the computer or other programmable device, generate instructions for implementing the process. Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1A process or multiple processes and / or structures illustrate the steps of the functions specified in one or more boxes.

[0260] The steps in the method of this application embodiment can be adjusted, combined, or deleted according to actual needs.

[0261] The modules in the device of this application embodiment can be merged, divided, and deleted according to actual needs.

[0262] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A data processing method, characterized in that, include: Obtain the business conversion dataset; the business conversion dataset includes data on the conversion behaviors performed by the object on the advertised business media; Based on the business conversion dataset, the conversion data volume of the conversion nodes of the business media is statistically analyzed to obtain conversion statistics results, and conversion quality results for the business media are generated based on the conversion statistics results. The return process of the business conversion dataset is subjected to return quality identification to obtain the return quality result; the return process of the business conversion dataset refers to the process of the main object returning the business conversion dataset. The main object refers to the object that provides the business media; If both the conversion quality result and the feedback quality result meet the sample quality conditions, the business conversion dataset will be provided as a training sample to the downstream processing node; the downstream processing node is used to estimate the conversion index for the conversion behavior.

2. The method according to claim 1, characterized in that, The conversion statistics results include a first statistical result and a second statistical result; the conversion nodes include shallow conversion nodes and deep conversion nodes; the step of performing conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset to obtain conversion statistics results, and generating conversion quality results for the business media based on the conversion statistics results, includes: In the business conversion dataset, the number of business conversion data that have been converted in the shallow conversion nodes and have also been converted in the deep conversion nodes is counted as the first statistical result; the execution order of the business media in the shallow conversion nodes is prior to the execution order of the conversions in the deep conversion nodes. In the business conversion dataset, the number of business conversion data that have performed conversion actions in the deep conversion nodes is counted as a second statistical result; The ratio of the first statistical result to the second statistical result is determined as the conversion quality result for the business media. The method further includes: If the transformation quality result is greater than or equal to the first proportion threshold, then the transformation quality result is determined to meet the sample quality condition.

3. The method according to claim 1, characterized in that, The backhaul quality results include centralized backhaul quality results; the backhaul quality identification process for the business conversion dataset to obtain backhaul quality results includes: Obtain the first business conversion data belonging to the current time range under the current period in the business conversion dataset, and the second business conversion data belonging to the full historical range time range in each historical period under N historical periods, where N is a positive integer; the full historical range time range includes multiple range time periods used to evenly divide a historical period. Obtain the first business conversion value corresponding to the first business conversion data, obtain the first conversion mean and the first conversion standard deviation corresponding to the second business conversion data, and generate the month-on-month conversion result for the business conversion dataset based on the first business conversion value, the first conversion mean and the first conversion standard deviation; Obtain the third business conversion data in the business conversion dataset that belongs to the same historical range time period within each of the N historical periods; the same historical range time period includes multiple range time periods that are adjacent to the current range time period in the unit time dimension; Obtain the second conversion mean and the second conversion standard deviation corresponding to the third business conversion data, and generate year-on-year conversion results for the business conversion dataset based on the first business conversion value, the second conversion mean, and the second conversion standard deviation; The month-on-month conversion results and the year-on-year conversion results are determined as the centralized feedback quality results for the business conversion dataset; The method further includes: If both the month-on-month conversion result and the year-on-year conversion result in the centralized backhaul quality results are less than the numerical threshold, then the centralized backhaul quality results are determined to meet the sample quality conditions.

4. The method according to claim 1, characterized in that, The backhaul quality results include delayed backhaul quality results; the backhaul quality identification process for the backhaul of the business conversion dataset to obtain backhaul quality results includes: Obtain the fourth business conversion data that satisfies the first business dimension from the business conversion dataset, and calculate the proportion of business conversion data in the fourth business conversion data whose reporting delay is greater than or equal to the delay threshold. Determine the proportion as the delay feedback quality result for the business conversion dataset. The reporting delay is determined based on the background receiving time of the conversion behavior associated with the fourth business conversion data and the occurrence time of the conversion behavior contained in the fourth business conversion data. The method further includes: If the delayed return quality result is less than the second proportional threshold, then the delayed return quality result is determined to meet the sample quality condition.

5. The method according to claim 1, characterized in that, The return quality results include repeated return quality results; the process of performing return quality identification on the return process of the business conversion dataset to obtain return quality results includes: Obtain the fifth business conversion data that satisfies the second business dimension from the business conversion dataset, and calculate the target daily conversion volume corresponding to the fifth business conversion data; Obtain the daily conversion volume of H business media domains under the second business dimension, generate the daily conversion mean and daily conversion standard deviation based on the H daily conversion volumes, generate a daily conversion volume threshold based on the daily conversion mean and daily conversion standard deviation, and determine the daily conversion volume threshold and the target daily conversion volume as the repeated backhaul quality result for the business conversion dataset; The method further includes: If the target daily conversion rate in the repeated feedback quality results is less than the daily conversion rate threshold, then the repeated feedback quality results are determined to meet the sample quality conditions.

6. The method according to claim 1, characterized in that, The feedback quality results include erroneous or omitted feedback quality results; the feedback quality identification process for the feedback of the business conversion dataset to obtain feedback quality results includes: In the business conversion dataset, the business conversion data in which the object has performed a conversion action on the landing page of the business media is obtained and used as the sixth business conversion data; The sixth business conversion data is compared with the conversion behavior data of the landing page of the business media recorded by the business media platform to obtain the first comparison result; The quantity of the sixth business conversion data is compared with the quantity of conversion behavior data recorded by the business media platform to obtain a second comparison result; The first comparison result and the second comparison result are determined as the error omission feedback quality results for the business conversion dataset; The method further includes: If the first comparison result indicates that the sixth business conversion data is the same as the conversion behavior data recorded by the business media platform, and the second comparison result indicates that the quantity of the sixth business conversion data is equal to the quantity of conversion behavior data recorded by the business media platform, then it is determined that the error omission feedback quality result meets the sample quality condition.

7. The method according to claim 1, characterized in that, The downstream processing node includes a conversion rate prediction node; the method further includes: Input the business conversion dataset into the conversion rate prediction model in the conversion rate prediction node; The conversion rate prediction model is used to extract features from the business conversion dataset to obtain object features, business media features, and object media interaction features. Based on the object features, business media features, and object media interaction features, a predicted probability of the object performing conversion behavior in response to the business media is generated. A first loss value is generated based on the predicted probability and the actual conversion rate corresponding to the business conversion dataset. The model parameters of the conversion rate prediction model are adjusted based on the first loss value to obtain a target conversion rate prediction model for predicting the probability of a target object converting to business media that have not been advertised.

8. The method according to claim 2, characterized in that, The downstream processing node includes a cost control node; the method further includes: The business conversion data in the business conversion dataset that has already performed conversion behavior in the deep conversion node is determined as training labels, and the business conversion data in the business conversion dataset that has already performed conversion behavior in the shallow conversion node and is associated with the training labels is determined as shallow training samples. The exposure click data for the business media in the business media platform, the training labels, and the shallow training samples are input into the return flow prediction model in the cost control node. In the backflow prediction model, the shallow conversion features corresponding to the shallow training samples and the exposure click features of the exposure click data are extracted. Based on the shallow conversion features and the exposure-click features, predict the deep conversion prediction probability corresponding to the shallow training sample, and generate a second loss value based on the deep conversion prediction probability and the training label; The model parameters of the reflux prediction model are adjusted according to the second loss value to obtain a reflux prediction model for predicting the deep conversion prediction probability of target conversion business data; the target conversion business data refers to the conversion business data that has already performed conversion behavior in the shallow conversion node.

9. The method according to claim 2, characterized in that, The downstream processing node includes a cold start exploration node; the cold start exploration node includes a cold start exploration pool A and a cold start exploration pool B; the cold start exploration pool A is a network model trained on business conversion data that has already performed conversion behavior in the shallow conversion nodes of the business conversion dataset, and the cold start exploration pool B is a network model trained on business conversion data that has already performed conversion behavior in the deep conversion nodes of the business conversion dataset; the method further includes: The new business media is input into the cold start exploration node, and the conversion rate of the new business media for the shallow conversion node is estimated based on the cold start exploration pool A to obtain the shallow conversion probability of the new business media; the new business media refers to business media that has not been launched. If the shallow conversion probability of the new business media is less than the shallow exploration probability threshold, then the cold start exploration of the new business media shall be stopped. If the shallow conversion probability of the new business media is greater than or equal to the shallow exploration probability threshold, then the new business media will be put on trial to obtain the trial business conversion volume corresponding to the new business media during the trial period. If the conversion rate of the trial business corresponding to the new business media is greater than or equal to the quantity threshold, then based on the cold start exploration pool B, the conversion rate of the new business media for the deep conversion node will be estimated to obtain the deep conversion probability of the new business media. If the deep conversion probability is greater than or equal to the deep exploration probability threshold, then the new business media will be officially launched.

10. A data processing apparatus, characterized in that, include: The sending and receiving module is used to acquire business conversion datasets; the business conversion datasets include data on the conversion behaviors performed by the object on the advertised business media. The first processing module is used to perform conversion data volume statistics on the conversion nodes of the business media based on the business conversion dataset, obtain conversion statistics results, and generate conversion quality results for the business media based on the conversion statistics results; The second processing module is used to identify the return quality of the business conversion dataset during the return process and obtain the return quality result; the return process of the business conversion dataset refers to the process of the main object returning the business conversion dataset; the main object refers to the object that provides the business media; The transceiver module is further configured to provide the business conversion dataset as a training sample to the downstream processing node if both the conversion quality result and the return quality result meet the sample quality conditions; the downstream processing node is configured to estimate the conversion index for the conversion behavior.

11. A computer device, characterized in that, include: Processor, memory, and network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide data communication functions, the memory is used to store computer programs, and the processor is used to call the computer programs so that the computer device executes the method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded and executed by a processor to cause a computer device having the processor to perform the method of any one of claims 1-9.

13. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium and adapted to be read and executed by a processor so that a computer device having the processor performs the method of any one of claims 1-9.

Images