Content push processing methods, devices, electronic devices and storage media

Abstract

This disclosure relates to a content push processing method, apparatus, electronic device, and storage medium. The method includes: obtaining account information and context information corresponding to a content push request; obtaining multiple first interaction indicator prediction information under multiple preset queue lengths based on the account information, context information, and multiple preset interaction indicator levels; determining multiple interaction indicator information under multiple preset queue lengths based on interaction indicator reference information of multiple preset interaction indicator levels and multiple first interaction indicator prediction information; determining a target queue length corresponding to the content push request from multiple preset queue lengths based on the multiple interaction indicator information; and, in response to the content push request, determining target content to be pushed from candidate content based on the target queue length, and pushing the target content. The technical solution provided by the embodiments of this disclosure can dynamically adjust the number of target contents pushed, improving the utilization efficiency of computing resources.

Description

Technical Field

[0001] This disclosure relates to the field of Internet technology, and in particular to push processing methods, devices, electronic devices and storage media. Background Technology

[0002] In internet content push services, the server responds to the request sent by the user terminal by performing multiple rounds of selection of candidate content. In these multiple rounds of selection, the queue length representing the number of candidate content in the selection results gradually decreases until a limited number of target contents to be pushed to the user terminal are determined from the candidate content.

[0003] However, in the same round of selection, candidate content is selected based on the same queue length for different requests. That is, the computing resources allocated to different requests are the same, which leads to a large consumption of computing resources and low utilization efficiency. Summary of the Invention

[0004] This disclosure provides a content push processing method, apparatus, electronic device, and storage medium to at least solve the problems of high computing power consumption and low utilization efficiency in related technologies. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a content push processing method is provided, including:

[0006] Obtain the account information corresponding to the content push request and the context information corresponding to the content push request;

[0007] Based on the account information, the context information, and multiple preset interaction indicator levels, multiple first interaction indicator prediction information is obtained under multiple preset queue lengths; the multiple preset queue lengths correspond one-to-one with the multiple first interaction indicator prediction information; the preset queue length is used to determine the number of target content to be pushed, and the first interaction indicator prediction information represents the predicted probability that the content interaction indicator under the corresponding preset queue length is at one of the multiple preset interaction indicator levels; the content interaction indicator under the corresponding preset queue length is used to describe the interaction between the account and the target content when the number of target content corresponds to the preset queue length.

[0008] Based on the interactive indicator reference information of the multiple preset interactive indicator levels and the multiple first interactive indicator prediction information, determine multiple interactive indicator information under the multiple preset queue lengths;

[0009] Based on the multiple interaction indicator information, determine the target queue length corresponding to the content push request from the multiple preset queue lengths;

[0010] In response to the content push request, the target content to be pushed is determined from the candidate content based on the target queue length, and the target content is pushed.

[0011] Optionally, the method further includes:

[0012] Obtain historical interaction metrics information for each of the multiple historical requests within the target time period;

[0013] The historical interaction indicator information of each of the multiple historical requests is processed by equal-frequency binning to obtain multiple datasets;

[0014] Based on the multiple datasets, the multiple preset interactive indicator levels are determined, and the multiple datasets correspond one-to-one with the multiple preset interactive indicator levels.

[0015] Optionally, determining the multiple interaction indicator information under the multiple preset queue lengths based on the interaction indicator reference information of the multiple preset interaction indicator levels and the multiple first interaction indicator prediction information includes:

[0016] Based on the historical interaction indicator information of each historical request in the multiple datasets, determine the average value of the multiple preset interaction indicator levels.

[0017] Based on the average value of the multiple preset interactive indicator levels and the predicted information of the first interactive indicator under the preset queue length, a weighted average process is performed to obtain the interactive indicator information under the preset queue length.

[0018] Optionally, determining the target queue length corresponding to the content push request from the plurality of preset queue lengths based on the plurality of interaction indicator information includes:

[0019] Based on the multiple interactive indicator information and the multiple preset queue lengths, determine multiple unit interactive indicator information under the multiple preset queue lengths;

[0020] The target unit interaction indicator information is determined from the plurality of unit interaction indicator information, wherein the target unit interaction indicator information is the maximum value among the plurality of unit interaction indicator information.

[0021] The target queue length is determined based on the preset queue length corresponding to the target unit interaction indicator information.

[0022] Optionally, the method further includes:

[0023] The system obtains request feature information of the sample request and tag information of the sample request. The request feature information includes sample account information corresponding to the sample request, sample context information corresponding to the sample request, and identifier information of the sample queue length corresponding to the sample request. The sample queue length is one of the multiple preset queue lengths. The tag information represents the actual interaction index information of the sample request under the sample queue length.

[0024] The sample account information, sample context information and identification information are input into the model to be trained, and interactive indicator level prediction processing is performed to obtain second interactive indicator prediction information. The second interactive indicator prediction information represents the prediction probability of the content interactive indicator under the sample queue length under the multiple preset interactive indicator levels.

[0025] Based on the second interaction index prediction information and the label information, the model to be trained is adjusted to obtain the interaction index prediction model.

[0026] After obtaining the account information corresponding to the content push request and the context information corresponding to the content push request, the account information and the context information are input into the interaction indicator prediction model to perform interaction indicator level prediction processing, thereby obtaining the multiple first interaction indicator prediction information under the multiple preset queue lengths.

[0027] Optionally, obtaining the request feature information of the sample request includes:

[0028] Randomly sample online requests to obtain a sample request set, wherein the sample request set includes multiple sample requests with the same sample queue length;

[0029] Based on the length of the sample queue, determine the identification information of each sample request in the sample request set.

[0030] Optionally, obtaining the label information of the sample request includes:

[0031] In response to multiple sample requests, content push processing is performed based on the sample queue length to obtain multiple sample interaction indicator information; the multiple sample interaction indicator information corresponds one-to-one with the multiple sample requests.

[0032] Based on the plurality of sample interaction index information, a target sample interaction index information corresponding to the sample queue length is determined, wherein the target sample interaction index information is the maximum value among the plurality of sample interaction index information.

[0033] Based on the multiple preset interaction indicator levels, the interaction indicator information of the target sample is discretized to obtain the tag information of each sample request in the sample request set.

[0034] According to a second aspect of the present disclosure, a push processing apparatus is provided, comprising:

[0035] The acquisition module is configured to acquire the account information corresponding to the content push request and the context information corresponding to the content push request;

[0036] The interaction metric tier prediction module is configured to execute, based on the account information, the context information, and multiple preset interaction metric tiers, to obtain multiple first interaction metric prediction information under multiple preset queue lengths; the multiple preset queue lengths correspond one-to-one with the multiple first interaction metric prediction information; the preset queue lengths are used to determine the quantity of target content to be pushed; the first interaction metric prediction information represents the predicted probability that the content interaction metric under the corresponding preset queue length is at one of the multiple preset interaction metric tiers; the content interaction metric under the corresponding preset queue length is used to describe the interaction between the account and the target content when the quantity of the target content corresponds to the preset queue length;

[0037] The interaction indicator determination module is configured to determine multiple interaction indicator information under multiple preset queue lengths based on the interaction indicator reference information of the multiple preset interaction indicator levels and the multiple first interaction indicator prediction information.

[0038] The queue length determination module is configured to determine the target queue length corresponding to the content push request from the multiple preset queue lengths based on the multiple interaction indicator information;

[0039] The push processing module is configured to respond to the content push request by determining the target content to be pushed from the candidate content based on the target queue length, and then pushing the target content.

[0040] Optionally, the device further includes:

[0041] The historical request acquisition unit is configured to acquire historical interaction indicator information for each of multiple historical requests within a target time period.

[0042] The bucketing processing unit is configured to perform equal-frequency bucketing processing on the historical interaction indicator information of the multiple historical requests to obtain multiple datasets.

[0043] The tier division unit is configured to determine the tiers of the multiple preset interactive indicators based on the multiple datasets, wherein the multiple datasets correspond one-to-one with the multiple preset interactive indicator tiers.

[0044] Optionally, the interaction indicator determination module includes:

[0045] The average value determination unit is configured to determine the average value of the multiple preset interaction indicator levels based on the historical interaction indicator information of each historical request in the multiple datasets.

[0046] The statistical analysis unit is configured to perform weighted averaging processing based on the average value of the multiple preset interactive indicator levels and the first interactive indicator prediction information under the preset queue length, to obtain the interactive indicator information under the preset queue length.

[0047] Optionally, the queue length determination module includes:

[0048] The unit interaction index determination unit is configured to determine multiple unit interaction index information under the multiple preset queue lengths based on the multiple interaction index information and the multiple preset queue lengths.

[0049] The target unit interaction indicator determination unit is configured to perform the task of determining target unit interaction indicator information from the plurality of unit interaction indicator information, wherein the target unit interaction indicator information is the maximum value among the plurality of unit interaction indicator information;

[0050] The target queue length determination unit is configured to determine the target queue length based on a preset queue length corresponding to the target unit interaction index information.

[0051] Optionally, the device further includes:

[0052] The training sample acquisition unit is configured to acquire request feature information of a sample request and acquire tag information of the sample request. The request feature information includes sample account information corresponding to the sample request, sample context information corresponding to the sample request, and identifier information of the sample queue length corresponding to the sample request. The sample queue length is one of the plurality of preset queue lengths. The tag information characterizes the actual interaction index information of the sample request under the sample queue length.

[0053] The training prediction unit is configured to input the sample account information, the sample context information and the identification information into the model to be trained, perform interactive indicator level prediction processing, and obtain second interactive indicator prediction information. The second interactive indicator prediction information represents the prediction probability of the content interactive indicator under the sample queue length under the multiple preset interactive indicator levels.

[0054] The model adjustment unit is configured to adjust the model to be trained based on the second interaction index prediction information and the label information to obtain the interaction index prediction model.

[0055] The interaction metric level prediction unit is configured to, after obtaining the account information corresponding to the content push request and the context information corresponding to the content push request, input the account information and the context information into the interaction metric prediction model, perform interaction metric level prediction processing, and obtain the multiple first interaction metric prediction information under the multiple preset queue lengths.

[0056] Optionally, the training sample acquisition unit includes:

[0057] The request sampling subunit is configured to perform random sampling of online requests to obtain a sample request set, wherein the sample request set includes multiple sample requests with the same sample queue length.

[0058] The identifier determination subunit is configured to perform the task of determining the identifier information of each sample request in the sample request set based on the sample queue length.

[0059] Optionally, the training sample acquisition unit further includes:

[0060] The first push subunit is configured to respond to multiple sample requests by performing content push processing based on the sample queue length to obtain multiple sample interaction indicator information; the multiple sample interaction indicator information corresponds one-to-one with the multiple sample requests.

[0061] The target sample interaction index determination subunit is configured to determine the target sample interaction index information corresponding to the sample queue length based on the plurality of sample interaction index information, wherein the target sample interaction index information is the maximum value among the plurality of sample interaction index information;

[0062] The tag determination subunit is configured to discretize the target sample interaction indicator information according to the multiple preset interaction indicator levels to obtain the tag information of each sample request in the sample request set.

[0063] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the content push processing method described in any one of the first aspects of the present disclosure.

[0064] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein when instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform a content push processing method as described in any one of the first aspects of the present disclosure.

[0065] According to a fifth aspect of the present disclosure, a computer program product is provided, including computer instructions that, when executed by a processor, implement the content push processing method as described in any one of the first aspects of the present disclosure.

[0066] The technical solutions provided by the embodiments of this disclosure bring at least the following beneficial effects:

[0067] The embodiments of this disclosure perform interaction indicator level prediction processing based on account information and context information of content push requests. This allows for the generation of multiple first interaction indicator prediction information under multiple preset queue lengths, with each preset queue length corresponding one-to-one with the first interaction indicator prediction information. The first interaction indicator prediction information represents the predicted probability that the content interaction indicator is at one of the multiple preset interaction indicator levels under the corresponding preset queue length. The preset queue length is used to determine the quantity of target content to be pushed. The content interaction indicator under the corresponding preset queue length describes the interaction between the account and the target content when the quantity of target content corresponds to the preset queue length. By combining the interaction indicator reference information of multiple preset interaction indicator levels and the multiple first interaction indicator prediction information, multiple interaction indicator information under multiple preset queue lengths can be determined. The interaction indicator information can indicate the interaction effect under the preset queue length. The multiple interaction indicator prediction information corresponds one-to-one with the multiple preset queue lengths, thus establishing a correlation between the preset queue length and the interaction effect of the content push request. Furthermore, based on the multiple interaction indicator information, the target queue length corresponding to the content push request can be determined from the multiple preset queue lengths. In response to the content push request, the target content to be pushed is determined from the candidate content based on the target queue length, and the target content is then pushed. This embodiment quantifies the interactive benefits of content push requests through interactive index information. By establishing a correlation between the preset queue length and the interactive benefits, a suitable target queue length can be configured for the content push request. That is, the target queue length configured for content push requests with high interactive benefits will be different from those with low interactive benefits, thus realizing personalized settings for the request queue length. Compared with using a fixed queue length, configuring the queue length personalized according to the interactive target benefits can improve the utilization efficiency of computing resources during the push processing, reduce the overall computing power consumption, avoid computing power waste, and maximize the interactive target benefits.

[0068] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0069] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0070] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment;

[0071] Figure 2 This is a flowchart illustrating a content push processing method according to an exemplary embodiment;

[0072] Figure 3 This is a schematic diagram illustrating the architecture of an interactive indicator prediction model according to an exemplary embodiment;

[0073] Figure 4 This is a flowchart illustrating a model training method according to an exemplary embodiment;

[0074] Figure 5 This is a schematic diagram illustrating the architecture of a model to be trained according to an exemplary embodiment;

[0075] Figure 6 This is a block diagram illustrating a content push processing apparatus according to an exemplary embodiment;

[0076] Figure 7 This is a block diagram illustrating an electronic device for implementing a content push processing method according to an exemplary embodiment;

[0077] Figure 8 This is a block diagram illustrating another electronic device for implementing a content push processing method according to an exemplary embodiment. Detailed Implementation

[0078] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0079] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0080] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0081] Please see Figure 1 The diagram illustrates an application environment for a content push processing method according to an exemplary embodiment. The application environment may include a terminal 110 and a server 120, which can be connected via a wired network or a wireless network.

[0082] Terminal 110 can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. Terminal 110 may have an application (App) installed. This application can be a standalone application or a subroutine within a standalone application. Users of terminal 110 can log in to the application using pre-registered account information, which may include a username and password. Server 120 can be a server providing background services for the application in terminal 110, or it can be another server connected and communicating with the application's background server. It can be a single server or a server cluster consisting of multiple servers.

[0083] In the embodiments of this disclosure, terminal 110 can send a content push request for a target service to server 120 in response to a user's content push operation. Server 120 receives the content push request and obtains the account information and context information of the content push request based on the content push request. Based on the account information, context information, and multiple preset interaction indicator levels, server 120 obtains multiple first interaction indicator prediction information for the content push request under multiple preset queue lengths. Combining the interaction indicator reference information of multiple preset interaction indicator levels, server 120 can determine multiple interaction indicator information for the content push request under multiple preset queue lengths. Furthermore, based on the multiple interaction indicator information, server 120 can determine the target queue length of the content push request from multiple preset queue lengths. Thus, in response to the content push request, server 120 can determine the target content to be pushed from candidate content based on the target queue length and push the target content. The embodiments of this disclosure realize personalized settings for the request queue length. Compared with using a fixed queue length, configuring the queue length personalized according to the interaction value can improve the utilization efficiency of computing resources during the push processing, reduce overall computing power consumption, avoid computing power waste, and maximize the interaction value.

[0084] Figure 2 This is a flowchart illustrating a content push processing method according to an exemplary embodiment, such as... Figure 2 As shown, the content push processing method used in Internet content push services may include the following steps:

[0085] In step S210, the account information and context information corresponding to the content push request are obtained.

[0086] This disclosure applies to internet content push services, including but not limited to search, advertising, and recommendation services. In internet content push services, the server responds to a content push request sent by a user terminal by performing multiple rounds of selection from candidate content. During these rounds, the queue length representing the number of candidate content items gradually decreases until a limited number of target content items to be pushed to the user terminal are determined from the candidate content. Candidate content can include advertisements, products, social media updates, images, videos, etc.

[0087] In this embodiment of the disclosure, a content push request represents a service request sent from the terminal logged into by the user account to the server. The feature information of the content push request includes the account information corresponding to the content push request and the context information corresponding to the content push request. The account information may include, but is not limited to, account identification information, gender information, age information, regional information, interest preference information, etc., and the context information may include, but is not limited to, terminal device identification information, network identification information, terminal device location information, system platform information running on the terminal device, application version information, and the time information of the last service request, etc.

[0088] In step S220, based on account information, context information, and multiple preset interaction indicator levels, multiple first interaction indicator prediction information under multiple preset queue lengths is obtained.

[0089] Among them, multiple preset queue lengths correspond one-to-one with multiple first interaction indicator prediction information; the preset queue length is used to determine the number of target content to be pushed, and the first interaction indicator prediction information represents the predicted probability that the content interaction indicator under the corresponding preset queue length is in multiple preset interaction indicator levels. The content interaction indicator under the corresponding preset queue length is used to describe the interaction between the account and the target content when the number of target content corresponds to the preset queue length.

[0090] In this embodiment of the disclosure, the queue length represents the number of candidate content in any content extraction stage of the Internet content push service, and can be used to finally determine the number of target content to be pushed. For example, the advertising push service mainly includes stages such as targeting, recall, coarse ranking, fine ranking, and push display. In the targeting stage, advertisers generally set the scope they need to place their ads, including audience targeting and geographic targeting. In the recall stage, after the system receives the advertising push request, it selects a portion of the candidate ads from all the candidate ads that have passed the targeting based on the information in the advertising request for subsequent sorting stages to filter. The queue length in the recall stage is the number of candidate ads for coarse ranking. In the coarse ranking stage, the candidate ads brought in by the recall stage are estimated based on metrics such as clicks and views, and then a certain number of candidate ads are selected for fine ranking. That is, the queue length in the coarse ranking stage is the number of candidate ads selected after coarse ranking, and also the number of candidate ads for fine ranking. Multiple preset queues have unequal lengths, which can be discretized using length levels. For example, length level 1 corresponds to a preset queue length of 50, length level 2 corresponds to a queue length of 100, length level 3 corresponds to a queue length of 150, and so on, with length level 10 corresponding to a preset queue length of 300. The higher the length level (1 < 2 < 3 ... < 10), the longer the preset queue length.

[0091] In this embodiment, content interaction metrics can also describe the interactive benefits that content push requests may bring. Content interaction metrics can be behavioral data indicators or conversion data indicators of the user account's interactive processing in response to the pushed content, such as click count, browsing time, conversion cost, and conversion revenue. Conversion revenue can be, for example, ECPM (Effective Cost Per Mille, which refers to the revenue earned per thousand impressions). Interaction metric levels characterize the statistical distribution of content interaction metrics, and each interaction metric level corresponds to a data range of the content interaction metric. The data ranges corresponding to multiple interaction metric levels do not overlap. In this embodiment, multiple preset interaction metric levels serve as a reference for quantifying the interactive benefits of content push requests.

[0092] In one embodiment of this disclosure, statistical analysis is performed using historical interaction indicator information from historical requests to determine multiple preset interaction indicator levels. Specifically, this may include the following steps:

[0093] In step S211, the historical interaction indicator information of each of the multiple historical requests within the target time period is obtained.

[0094] Specifically, the target time period can be a specific period, such as a day or a week. The historical interaction metrics information for multiple historical requests is of the same type. Historical requests are content push requests that have been processed and are in a completed state.

[0095] In step S212, the historical interaction indicator information of each of the multiple historical requests is processed by equal frequency binning to obtain multiple datasets.

[0096] Specifically, when historical interaction metrics are continuous data, bucketing can discretize this information, making it easier to calculate and store. Equal-frequency bucketing, also known as quantile bucketing, involves each data bucket (i.e., dataset) containing the same number of historical requests.

[0097] In step S213, multiple preset interactive indicator levels are determined based on multiple datasets, and each dataset corresponds one-to-one with the multiple preset interactive indicator levels.

[0098] A preset interaction indicator level can be determined based on a dataset, and the level can be labeled. Further, the data interval boundary values ​​and the level mean of the corresponding preset interaction indicator level can be calculated based on the historical interaction indicator information of each historical request within the dataset. For example, the data interval of preset interaction indicator level G1 is [0, 50), and the level mean is 30; the data interval of preset interaction indicator level G2 is [50, 80), and the level mean is 69... The data intervals corresponding to multiple interaction indicator levels do not overlap. The higher the level of the preset interaction indicator (G1 < G2 < ...), the higher the minimum value, maximum value, and level mean of the corresponding data interval all show an increasing trend.

[0099] In the above embodiments, the multiple preset interaction indicator levels determined by the historical interaction indicator information of historical requests can serve as a reference for quantifying the interaction benefits of content push requests. At the same time, the multiple preset interaction indicator levels are also the result of discretizing the historical interaction indicator information, which is convenient for calculation and storage.

[0100] In one embodiment of this disclosure, the account information and context information of the content push request are input into the interaction metric prediction model. Based on the account information and context information of the content push request, the interaction metric prediction model predicts the probability that the content interaction metric will be at each preset interaction metric level for each preset queue length. That is, the resulting multiple first interaction metric predictions correspond one-to-one with multiple preset queue lengths, and each first interaction metric prediction can include a probability value of the content interaction metric being at each preset interaction metric level for the corresponding preset queue length. For example... Figure 3As shown, when there are N (L1, L2...Ln) preset queue lengths and M preset interaction indicator levels (distributed as 1, 2...m), the interaction indicator prediction model will output the first interaction indicator prediction data D1 = {P} under the preset queue length L1. 11 P 12 ...P 1m}, The first interactive indicator prediction data D2 = {P} under the preset queue length L2 21 P 22 ...P 2m ...The first interaction metric prediction data Dn = {P} under the preset queue length Ln. n1 P n2 ...P nm}, where P ij This represents the probability that the content interaction index is at the j-th (1≤j≤m, j is an integer) preset interaction index level under a preset queue length Li (1≤i≤n, i is an integer).

[0101] In one embodiment of this disclosure, the interaction metric prediction model can adopt a multi-task learning architecture and can be built and trained based on a deep neural network (DNN) to obtain the interaction metric prediction model. The model training phase can be detailed in subsequent embodiments and will not be elaborated here.

[0102] In step S230, based on the interactive indicator reference information of multiple preset interactive indicator levels and the prediction information of multiple first interactive indicators, multiple interactive indicator information under multiple preset queue lengths are determined.

[0103] In this embodiment of the disclosure, the interaction indicator reference information is used to describe the content interaction indicator value corresponding to the interaction indicator level, such as the level average, level maximum, level median, etc.; the interaction indicator information represents the content interaction indicator value under the corresponding preset queue length, which can be used to specifically describe the interaction effect of the content push request. Multiple interaction indicator prediction information corresponds one-to-one with multiple preset queue lengths, that is, a correlation relationship is constructed between the preset queue length and the interaction effect of the content push request.

[0104] Considering that in step S220, the first interaction index prediction information output by the model is the predicted probability of the content interaction index being in multiple preset interaction index levels, in this embodiment of the disclosure, multiple preset interaction index levels are used as a reference, and multiple first interaction index prediction information are used to estimate and restore the content interaction index value under each preset queue length.

[0105] In this embodiment, the interaction index information is positively correlated with the preset queue length. That is, the larger the preset queue length, the higher the value of the content interaction index represented by the interaction index information. When sorted by length unit within the preset queue length, the corresponding interaction index information shows an increasing trend. This is consistent with the basic understanding in Internet content push services that the more content pushed, the more interactive behaviors and the higher the interaction benefits.

[0106] Based on the embodiments disclosed in steps S211 to S213, step S230 can be specifically implemented as follows:

[0107] In step S231, the average value of multiple preset interaction indicator levels is determined based on the historical interaction indicator information of each historical request in multiple datasets.

[0108] That is, the average value of historical interaction metrics information of historical requests included in each dataset is calculated as the average value of the preset interaction metric level corresponding to each dataset.

[0109] In step S232, a weighted average is performed based on the average value of multiple preset interactive indicator levels and the predicted information of the first interactive indicator under the preset queue length to obtain the interactive indicator information under the preset queue length.

[0110] Specifically, the predicted probability of the content interaction indicator under each preset interaction indicator level in the first interaction indicator prediction information is multiplied by the average value of each preset interaction indicator level, which is equivalent to multiplying and summing the average value of the level as a weight to obtain the interaction indicator information under the preset queue length. It can also be understood as calculating the expected value of the mathematical value based on the average value of each preset interaction indicator level and the predicted probability of the content interaction indicator under each preset interaction indicator level, and the calculation result is used as the interaction indicator information under the preset queue length.

[0111] For example, when there are M (1, 2, ..., m) preset interaction indicator levels, with a preset queue length of Li, the predicted data for the first interaction indicator is Di = {P}. i1 P i2 ...P im}, then the interaction index information Ti under the preset queue length Li can be estimated according to formula (1):

[0112]

[0113] Among them, P ij V represents the probability that the content interaction index is at the j-th (1≤j≤m, j is an integer) preset interaction index level under the preset queue length Li. j The mean value of the j-th preset interactive indicator level.

[0114] In the above embodiments, based on the average value of each preset interaction indicator level and the predicted probability of the content interaction indicator being at each preset interaction indicator level, the expected value is calculated to obtain the interaction indicator information under the preset queue length. This can accurately quantify the interaction benefits of the content push request and clarify the correlation between the preset queue length and the interaction benefits of the content push request. At the same time, based on the setting of multiple preset queue lengths and multiple preset interaction indicator levels, the interaction indicator information has order preservation, which is consistent with the results of real-world applications.

[0115] In step S240, the target queue length corresponding to the content push request is determined from multiple preset queue lengths based on multiple interaction indicator information.

[0116] In this embodiment, the target queue length matching the content push request can be determined based on multiple interaction metrics. Specifically, the higher the interaction benefit of the content push request indicated by the interaction metrics, the larger the matching target queue length, and correspondingly, the more computing resources are allocated to the content push request; conversely, the lower the interaction benefit of the content push request indicated by the interaction metrics, the smaller the matching target queue length, and correspondingly, the less computing resources are allocated to the content push request. Compared to using a fixed queue length, dynamically adjusting the queue length based on interaction benefits can improve the utilization efficiency of computing resources during push processing, avoid computing waste, and maximize interaction benefits.

[0117] Multiple first interactive indicator prediction information corresponds one-to-one with multiple preset queue lengths, and the multiple first interactive indicator prediction information is order-preserving, that is, the larger the preset queue length, the larger the value represented by the corresponding first interactive indicator prediction information. In one embodiment of this disclosure, step S240 can be specifically implemented as follows:

[0118] In step S241, multiple unit interactive indicator information under multiple preset queue lengths is determined based on multiple interactive indicator information and multiple preset queue lengths.

[0119] Specifically, as shown in formula (2), based on multiple interaction index information Ti (1≤i≤n) under multiple preset queue lengths and multiple preset queue lengths Li (1≤i≤n), multiple unit interaction index information under multiple preset queue lengths can be calculated:

[0120] ROIi=Ti / Li(2)

[0121] In step S242, the target unit interaction indicator information is determined from multiple unit interaction indicator information, and the target unit interaction indicator information is the maximum value among the multiple unit interaction indicator information.

[0122] In step S243, the target queue length is determined based on the preset queue length corresponding to the target unit interaction index information.

[0123] In the above embodiments, considering that the interaction indicator information has order preservation, the preset queue length corresponding to the largest interaction indicator information is also the largest. Directly determining the queue length based on the interaction indicator information cannot achieve dynamic adjustment of the queue length. Therefore, based on the interaction indicator information, unit interaction indicator information is further determined. Unit interaction indicator information can characterize the interaction benefits that an account can generate with a single pushed content, and can more accurately depict the interaction value of the content push request. The target queue length of the content push request is determined based on the maximum value among multiple unit interaction indicator information minus the target unit interaction indicator information. This can achieve personalized configuration of the queue length, avoid wasting computing power, and maximize the interaction benefits.

[0124] In another embodiment of this disclosure, the weight coefficients of each preset queue length can be configured according to the overall business needs and computing power deployment. The target queue length for content push requests is determined by multiplying the weight coefficient of the preset queue length with the corresponding unit interaction indicator information.

[0125] In step S250, in response to the content push request, the target content to be pushed is determined from the candidate content based on the target queue length, and the target content to be pushed is also determined.

[0126] In this embodiment of the disclosure, the push processing may include, but is not limited to, recall, coarse ranking, and fine ranking stages. For example, in the coarse ranking stage, the number of clicks, views, and other metrics of the candidate advertisements sent in the recall are estimated, and then the target content of the number indicated by the target queue length is selected for fine ranking. The target queue length is also the number of target content to be fine-ranked.

[0127] As can be seen from the technical solutions provided by the embodiments of this disclosure above, the embodiments of this disclosure perform interaction indicator level prediction processing based on account information and context information of content push requests, and can obtain multiple first interaction indicator prediction information under multiple preset queue lengths, and the multiple preset queue lengths correspond one-to-one with the multiple first interaction indicator prediction information. The first interaction indicator prediction information represents the predicted probability that the content interaction indicator is in multiple preset interaction indicator levels under the corresponding preset queue length. The preset queue length is used to determine the number of target content to be pushed, and the content interaction indicator under the corresponding preset queue length is used to describe the interaction between the account and the target content when the number of target content corresponds to the preset queue length. The interaction status is analyzed; by combining the interaction indicator reference information of multiple preset interaction indicator levels and the prediction information of multiple first interaction indicators, multiple interaction indicator information under multiple preset queue lengths can be determined. The interaction indicator information can indicate the interaction effect under the preset queue length. The multiple interaction indicator prediction information corresponds one-to-one with the multiple preset queue lengths, that is, a correlation relationship is constructed between the preset queue length and the interaction effect of the content push request; then, based on the multiple interaction indicator information, the target queue length corresponding to the content push request can be determined from the multiple preset queue lengths, and in response to the content push request, the target content to be pushed can be determined from the candidate content based on the target queue length, and the target content can be pushed. This embodiment quantifies the interactive benefits of content push requests through interactive index information. By establishing a correlation between the preset queue length and the interactive benefits, a suitable target queue length can be configured for the content push request. That is, the target queue length configured for content push requests with high interactive benefits will be different from those with low interactive benefits, thus realizing personalized settings for the request queue length. Compared with using a fixed queue length, configuring the queue length personalized according to the interactive target benefits can improve the utilization efficiency of computing resources during the push processing, reduce the overall computing power consumption, avoid computing power waste, and maximize the interactive target benefits.

[0128] In one embodiment of this disclosure, the above-described content push processing method may further include model training and application phases, such as... Figure 4 As shown, it can be implemented as follows:

[0129] In step S310, the request feature information of the sample request and the tag information of the sample request are obtained. The request feature information includes the sample account information corresponding to the sample request, the sample context information corresponding to the sample request, and the identifier information of the sample queue length corresponding to the sample request. The sample queue length is one of multiple preset queue lengths.

[0130] Among them, the label information represents the actual interaction metrics of sample requests within the sample queue length.

[0131] Specifically, the request feature information of a sample request includes the sample account information corresponding to the sample request and the sample context information corresponding to the sample request. The sample account information may include, but is not limited to, account identifier information, gender information, age information, geographic information, and interest preference information. The sample context information may include, but is not limited to, terminal device identifier information, network identifier information, terminal device location information, system platform information running on the terminal device, application version information, and the time of the last service request. The sample queue length identifier information indicates the sample queue length corresponding to the sample request and can be a one-hot encoded vector.

[0132] It should be noted that, considering that it is not easy to obtain multiple real interaction indicator information under multiple preset queue lengths as a sample request, but it is relatively easy to obtain real interaction indicator information under a certain preset queue length, the sample request set can be divided according to the preset queue length in this embodiment of the disclosure.

[0133] In one feasible implementation, relevant information about sample requests is obtained using real online requests and the actual results of pushing these online requests. Specifically, this may include the following steps:

[0134] In step S311, online requests are randomly sampled to obtain a sample request set, and the sample request set includes multiple sample requests with the same sample queue length.

[0135] For example, with 10 preset queue lengths, 10% of the online requests are randomly sampled as full sample requests. The sample queue lengths are then evenly distributed across these sample requests. Specifically, 1% of the sample requests are assigned to length tier 1 (preset queue length 50), 1% to length tier 2 (preset queue length 100), 1% to length tier 3 (preset queue length 150), 1% to length tier 4 (preset queue length 200), and so on. A total of 10% of the online requests are evenly distributed across different preset queue lengths, resulting in a sample request set corresponding to each preset queue length. In other feasible embodiments, the number of samples corresponding to different sample queue lengths can also be unequal.

[0136] In step S312, the identification information of each sample request in the sample request set is determined according to the sample queue length.

[0137] For a single sample request, its corresponding sample queue length has only one possible result. However, the sample queue lengths of multiple sample requests from different sample request sets are different, requiring identification information to mark the sample queue lengths of the sample requests. For example, in the case of 10 preset queue lengths, the identification information corresponding to the preset queue lengths of length levels 1 to 10 can be represented as: [1,0,0,0,0,0,0,0,0,0], [0,1,0,0,0,0,0,0,0,0], [0,0,1,0,0,0,0,0,0,0]...[0,0,0,0,0,0,0,0,0,1], where [1,0,0,0,0,0,0,0,0,0] indicates that this sample request comes from length level 1, i.e., the corresponding preset queue length is 50. Each preset queue length corresponds to multiple sample requests, and these multiple sample requests have the same sample queue length and the same identification information.

[0138] In the above embodiments, sampling using real online requests makes it easier to obtain sample requests and their feature information, avoiding the process of manually constructing training samples and improving training efficiency.

[0139] In step S313, in response to multiple sample requests, content push processing is performed based on the corresponding sample queue length to obtain multiple sample interaction indicator information; the multiple sample interaction indicator information corresponds one-to-one with the multiple sample requests.

[0140] By using the sample queue length as the queue length of each sample request in a certain stage of the push processing, and using the result obtained in the push processing as the sample interaction indicator information corresponding to the sample request, the sample interaction indicator information represents the interaction indicator value under the sample queue length, which can be used to truly describe the interaction effect of the sample request. For example, after the coarse ranking stage of the advertising system, the fine ranking stage estimates the CTR (Click-Through-Rate) and CVR (Conversion Rate) of multiple candidate ads generated by the coarse ranking, and calculates ECPM = CTR * CVR * BID, where BID (bidprice) represents the ad bid. The ECPM of the candidate ads is aligned in terms of dimensions, so the ECPM determined by the fine ranking stage can be used as the sample interaction indicator information of the sample request in the coarse ranking stage.

[0141] It is understandable that when sampling from online requests, if the number of sample requests contained in sample request set 1 corresponding to length 1 and sample request set 2 corresponding to length 2 are the same and sufficiently large, it can be assumed that the sample request distribution of sample request set 1 and sample request set 2 is consistent. That is, for a sample request A in sample request set 1, a sample request B can be found in sample request set 2 such that sample request A is closest to sample request B. It can be approximated that the sample queue length of the same sample request is set to different preset queue lengths, and sample interaction index information under different preset queue lengths is obtained through push processing.

[0142] In step S314, based on multiple sample interaction index information, the target sample interaction index information corresponding to the sample queue length is determined. The target sample interaction index information is the maximum value among the multiple sample interaction index information.

[0143] In step S315, the interactive indicator information of the target sample is discretized according to multiple preset interactive indicator levels to obtain the label information of each sample request in the sample request set.

[0144] In other words, for each preset queue length, it is necessary to determine the target sample interaction index information corresponding to the preset queue length. When the number of sample requests is large enough, as the sample queue length increases, the target sample interaction index information corresponding to the sample queue length also shows an increasing trend, that is, there is order preservation, which is consistent with the push business, and this is what the model needs to learn. However, the sample interaction index information of sample requests in the sample request set corresponding to the sample queue length may not have order preservation, and there may be a situation where the sample interaction index information of a sample request corresponding to a lower sample queue length is greater than that of a sample request corresponding to a higher sample queue length. Therefore, in this embodiment, the sample interaction index information is not directly used as the label information of the corresponding sample request. Instead, a unified label information for all sample requests in the sample request set corresponding to the sample queue length is determined based on the target sample interaction index information to ensure that the model can learn the above-mentioned order preservation during training and improve the accuracy of model training.

[0145] When the target sample interaction indicator information is continuous data, its discretized result needs to be used as label information. Specifically, the target sample interaction indicator information is encoded using multiple preset interaction indicator levels obtained after binning. For example, if there are 10 preset queue length levels and 10 preset interaction indicator levels, then the target sample interaction indicator information corresponding to the preset queue length level 1 is discretized as (1,0,0,0,0,0,0,0,0,0), the target sample interaction indicator information corresponding to the length level 2 is discretized as (1,1,0,0,0,0,0,0,0,0), the target sample interaction indicator information corresponding to the length level 3 is discretized as (1,1,1,0,0,0,0,0,0,0), and the target sample interaction indicator information corresponding to the length level 4 is... The target sample interaction index information, after discretization, is (1,1,1,1,0,0,0,0,0,0), and the target sample interaction index information corresponding to length level 5, after discretization, is (1,1,1,1,1,0,0,0,0,0)... The label information is not a one-hot encoded vector. The label information (1,1,0,0,0,0,0,0,0,0) of each sample request corresponding to length level 2 can indicate that the real interaction index information of the sample request can be in the preset interaction index level 1 or the preset interaction index level 2.

[0146] In the above embodiments, by utilizing real online requests and the actual results of pushing these requests, the label information of sample requests can be obtained relatively easily, improving model training efficiency. Simultaneously, based on the target sample interaction index information, unified label information is determined for all sample requests in the sample request set corresponding to the length of the sample queue. This ensures that the model can learn the order-preserving nature of the target sample interaction index information during training, improving the accuracy of model training. When the target sample interaction index information is continuous data, using the discretized result as the label information facilitates calculation and storage, and also meets the requirements of model data processing.

[0147] In step S320, the sample account information, sample context information, and identification information are input into the model to be trained, and interactive indicator level prediction processing is performed to obtain the second interactive indicator prediction information. The second interactive indicator prediction information represents the prediction probability of the content interactive indicator under multiple preset interactive indicator levels under the sample queue length.

[0148] Feasibly, the model to be trained adopts a multi-task learning approach and a three-layer DNN structure, with each layer containing 1024, 512, and 128 neurons respectively. For example... Figure 5As shown, level 1 - Task 1, level 2 - Task level 2, ..., level 10 - Task 10 represent the outputs of the tasks corresponding to the 10 length levels. Since the sample queue length of the sample request corresponds to only one length level, the second interaction index prediction information output by the model to be trained only includes the predicted probability of the content interaction index being in each preset interaction index level under the sample queue length. With 10 preset queue lengths, when the sample request identifier is [1,0,0,0,0,0,0,0,0,0], the model to be trained will only output the output result of the task corresponding to length level 1; when the sample request identifier is [0,1,0,0,0,0,0,0,0,0], the model to be trained will only output the output result of the task corresponding to length level 2.

[0149] In step S330, the model to be trained is adjusted based on the second interaction index prediction information and the label information to obtain the interaction index prediction model.

[0150] Based on the second interaction metric prediction information and label information, the cross-wrap loss value can be calculated, and the parameters of the model to be trained can be adjusted according to the loss value. By performing the above operation on the sample request sets corresponding to multiple preset queue lengths, an interaction metric prediction model can be obtained. This model can predict the probability of the content interaction metric being at each preset interaction metric level under each preset queue length, based on the feature information of the content push request (the feature information of the content push request does not have label information).

[0151] Furthermore, after obtaining the account information and context information corresponding to the content push request, the account information and context information are input into the interaction indicator prediction model to perform interaction indicator level prediction processing, thereby obtaining multiple first interaction indicator prediction information for the content push request under multiple preset queue lengths.

[0152] In the above embodiments, multiple sample request sets are divided according to a preset queue length. The label information of the sample requests in the sample request set only represents the interaction index information of the sample requests under the sample queue length. The easy acquisition of label information improves the training efficiency of the model. At the same time, the multi-task model is trained by using multiple request sets corresponding to multiple preset queue lengths. The resulting interaction index prediction model can predict the probability that the content interaction index is at each preset interaction index level under each preset queue length, which meets the requirement of dynamically adjusting the queue length corresponding to the content push request.

[0153] Figure 6 This is a block diagram illustrating a content push processing apparatus according to an exemplary embodiment. (Refer to...) Figure 6 The device includes:

[0154] The acquisition module 610 is configured to acquire the account information corresponding to the content push request and the context information corresponding to the content push request.

[0155] The interaction metric level prediction module 620 is configured to perform operations based on the account information, the context information, and multiple preset interaction metric levels to obtain multiple first interaction metric prediction information under multiple preset queue lengths; the multiple preset queue lengths correspond one-to-one with the multiple first interaction metric prediction information; the preset queue lengths are used to determine the quantity of target content to be pushed; the first interaction metric prediction information represents the predicted probability that the content interaction metric is at one of the multiple preset interaction metric levels under the corresponding preset queue length; the content interaction metric under the corresponding preset queue length is used to describe the interaction between the account and the target content when the quantity of the target content corresponds to the preset queue length;

[0156] The interactive indicator determination module 630 is configured to determine multiple interactive indicator information under multiple preset queue lengths based on the multiple preset interactive indicator reference information and the multiple first interactive indicator prediction information.

[0157] The queue length determination module 640 is configured to determine the target queue length corresponding to the content push request from the multiple preset queue lengths based on the multiple interaction indicator information.

[0158] The push processing module 650 is configured to respond to the content push request by determining the target content to be pushed from the candidate content based on the target queue length, and pushing the target content.

[0159] Optionally, the device further includes:

[0160] The historical request acquisition unit is configured to acquire historical interaction indicator information for each of multiple historical requests within a target time period.

[0161] The bucketing processing unit is configured to perform equal-frequency bucketing processing on the historical interaction indicator information of the multiple historical requests to obtain multiple datasets.

[0162] The tier division unit is configured to determine the tiers of the multiple preset interactive indicators based on the multiple datasets, wherein the multiple datasets correspond one-to-one with the multiple preset interactive indicator tiers.

[0163] Optionally, the interaction indicator determination module 630 includes:

[0164] The average value determination unit is configured to determine the average value of the multiple preset interaction indicator levels based on the historical interaction indicator information of each historical request in the multiple datasets.

[0165] The statistical analysis unit is configured to perform weighted averaging processing based on the average value of the multiple preset interactive indicator levels and the first interactive indicator prediction information under the preset queue length, to obtain the interactive indicator information under the preset queue length.

[0166] Optionally, the queue length determination module 640 includes:

[0167] The unit interaction index determination unit is configured to determine multiple unit interaction index information under the multiple preset queue lengths based on the multiple interaction index information and the multiple preset queue lengths.

[0168] The target unit interaction indicator determination unit is configured to perform the task of determining target unit interaction indicator information from the plurality of unit interaction indicator information, wherein the target unit interaction indicator information is the maximum value among the plurality of unit interaction indicator information;

[0169] The target queue length determination unit is configured to determine the target queue length based on a preset queue length corresponding to the target unit interaction index information.

[0170] Optionally, the device further includes:

[0171] The training sample acquisition unit is configured to acquire request feature information of a sample request and acquire tag information of the sample request. The request feature information includes sample account information corresponding to the sample request, sample context information corresponding to the sample request, and identifier information of the sample queue length corresponding to the sample request. The sample queue length is one of the plurality of preset queue lengths. The tag information characterizes the actual interaction index information of the sample request under the sample queue length.

[0172] The training prediction unit is configured to input the sample account information, the sample context information and the identification information into the model to be trained, perform interactive indicator level prediction processing, and obtain second interactive indicator prediction information. The second interactive indicator prediction information represents the prediction probability of the content interactive indicator under the sample queue length under the multiple preset interactive indicator levels.

[0173] The model adjustment unit is configured to adjust the model to be trained based on the second interaction index prediction information and the label information to obtain the interaction index prediction model.

[0174] The interaction metric level prediction unit is configured to, after obtaining the account information corresponding to the content push request and the context information corresponding to the content push request, input the account information and the context information into the interaction metric prediction model, perform interaction metric level prediction processing, and obtain the multiple first interaction metric prediction information under the multiple preset queue lengths.

[0175] Optionally, the training sample acquisition unit includes:

[0176] The request sampling subunit is configured to perform random sampling of online requests to obtain a sample request set, wherein the sample request set includes multiple sample requests with the same sample queue length.

[0177] The identifier determination subunit is configured to perform the task of determining the identifier information of each sample request in the sample request set based on the sample queue length.

[0178] Optionally, the training sample acquisition unit further includes:

[0179] The first push subunit is configured to respond to multiple sample requests by performing content push processing based on the sample queue length to obtain multiple sample interaction indicator information; the multiple sample interaction indicator information corresponds one-to-one with the multiple sample requests.

[0180] The target sample interaction index determination subunit is configured to determine the target sample interaction index information corresponding to the sample queue length based on the plurality of sample interaction index information, wherein the target sample interaction index information is the maximum value among the plurality of sample interaction index information;

[0181] The tag determination subunit is configured to discretize the target sample interaction indicator information according to the multiple preset interaction indicator levels to obtain the tag information of each sample request in the sample request set.

[0182] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0183] Figure 7 This is a block diagram illustrating an electronic device for implementing a content push processing method according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a content push processing method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0184] Figure 8 This is a block diagram illustrating an electronic device for implementing a content push processing method according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a content push processing method.

[0185] Those skilled in the art will understand that Figure 7 and Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0186] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the content push processing method as described in the embodiments of this disclosure.

[0187] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the content push processing method of the present disclosure embodiments.

[0188] In an exemplary embodiment, a computer program product is also provided, including computer instructions that, when executed by a processor, implement the content push processing method of this disclosure embodiment.

[0189] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0190] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0191] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A content push processing method characterized by comprising: The method comprises: obtaining account information corresponding to a content push request and context information corresponding to the content push request; based on the account information, the context information and a plurality of preset interaction indicator ranges, obtaining a plurality of first interaction indicator prediction information under a plurality of preset queue lengths; the plurality of preset queue lengths correspond one-to-one to the plurality of first interaction indicator prediction information; the preset queue length is used to determine the number of target content to be pushed, and the first interaction indicator prediction information represents the predicted probability of the content interaction indicator under the corresponding preset queue length being in the plurality of preset interaction indicator ranges, and the interaction between the account and the target content when the number of target content corresponds to the preset queue length; determining a plurality of interaction indicator information under the plurality of preset queue lengths according to the interaction indicator reference information of the plurality of preset interaction indicator ranges and the plurality of first interaction indicator prediction information; determining a plurality of unit interaction indicator information under the plurality of preset queue lengths according to the plurality of interaction indicator information and the plurality of preset queue lengths; determining target unit interaction indicator information from the plurality of unit interaction indicator information, the target unit interaction indicator information being the maximum value in the plurality of unit interaction indicator information; determining a target queue length corresponding to the content push request based on the preset queue length corresponding to the target unit interaction indicator information; in response to the content push request, determining the target content to be pushed from the candidate content based on the target queue length, and pushing the target content.

2. The method of claim 1, wherein, The method further comprises: obtaining historical interaction indicator information of a plurality of historical requests in a target time period; performing equal-frequency bucketing processing on the historical interaction indicator information of the plurality of historical requests to obtain a plurality of data sets; determining the plurality of preset interaction indicator ranges according to the plurality of data sets, the plurality of data sets corresponding one-to-one to the plurality of preset interaction indicator ranges.

3. The method of claim 2, wherein, The determination of the plurality of interaction indicator information under the plurality of preset queue lengths according to the interaction indicator reference information of the plurality of preset interaction indicator ranges and the plurality of first interaction indicator prediction information comprises: determining the average value of the plurality of preset interaction indicator ranges according to the historical interaction indicator information of each historical request in the plurality of data sets; performing weighted average processing according to the average value of the plurality of preset interaction indicator ranges and the first interaction indicator prediction information under the preset queue length to obtain the interaction indicator information under the preset queue length.

4. The method of claim 1, wherein, The method further comprises: obtain request feature information of a sample request, and obtain label information of the sample request, the request feature information including sample account information corresponding to the sample request, sample context information corresponding to the sample request, and identification information of a sample queue length corresponding to the sample request; the sample queue length is one of the plurality of preset queue lengths; the label information represents real interaction indicator information of the sample request under the sample queue length; input the sample account information, the sample context information, and the identification information into a to-be-trained model to perform interaction indicator gear prediction processing, and obtain second interaction indicator prediction information, the second interaction indicator prediction information representing a predicted probability of the content interaction indicator under the plurality of preset interaction indicator gears under the sample queue length; adjust the to-be-trained model based on the second interaction indicator prediction information and the label information to obtain an interaction indicator prediction model; after obtaining the account information corresponding to the content push request and the context information corresponding to the content push request, input the account information and the context information into the interaction indicator prediction model to perform interaction indicator gear prediction processing, and obtain the plurality of first interaction indicator prediction information under the plurality of preset queue lengths.

5. The method of claim 4, wherein, The request feature information of the sample request includes: randomly sample online requests to obtain a sample request set, the sample request set including a plurality of sample requests with the same sample queue length; determine the identification information of each sample request in the sample request set according to the sample queue length.

6. The method of claim 5, wherein, The label information of the sample request includes: perform content push processing based on the corresponding sample queue length in response to a plurality of sample requests to obtain a plurality of sample interaction indicator information; the plurality of sample interaction indicator information corresponds one-to-one to the plurality of sample requests; determine target sample interaction indicator information corresponding to the sample queue length according to the plurality of sample interaction indicator information, the target sample interaction indicator information being the maximum value in the plurality of sample interaction indicator information; discretize the target sample interaction indicator information according to the plurality of preset interaction indicator gears to obtain label information of each sample request in the sample request set.

7. A content push processing apparatus characterized by comprising: The device includes: an obtaining module configured to obtain account information corresponding to a content push request and context information corresponding to the content push request; The interaction index level prediction module is configured to perform, based on the account information, the context information, and a plurality of preset interaction index levels, obtaining a plurality of first interaction index prediction information under a plurality of preset queue lengths; the plurality of preset queue lengths correspond to the plurality of first interaction index prediction information one by one; the preset queue length is used to determine the number of target content to be pushed, and the first interaction index prediction information represents the prediction probability of the content interaction index under the corresponding preset queue length in the plurality of preset interaction index levels, and the interaction between the account and the target content when the number of target content corresponds to the preset queue length; The interaction index determination module is configured to perform, according to the interaction index reference information of the plurality of preset interaction index levels and the plurality of first interaction index prediction information, determine a plurality of interaction index information under the plurality of preset queue lengths; The queue length determination module is configured to perform, according to the plurality of interaction index information and the plurality of preset queue lengths, determine a plurality of unit interaction index information under the plurality of preset queue lengths; determine target unit interaction index information from the plurality of unit interaction index information, the target unit interaction index information is the maximum value in the plurality of unit interaction index information; determine the target queue length corresponding to the content push request based on the preset queue length corresponding to the target unit interaction index information; The push processing module is configured to perform, in response to the content push request, determine the target content to be pushed from the candidate content based on the target queue length, and push the target content.

8. The apparatus of claim 7, wherein, The device further comprises: The historical request acquisition unit is configured to perform obtaining a plurality of historical request each historical interaction index information in a target time period; The bucket processing unit is configured to perform equal-frequency bucket processing on the historical interaction index information of each of the plurality of historical requests to obtain a plurality of data sets; The level division unit is configured to perform, according to the plurality of data sets, determine the plurality of preset interaction index levels, the plurality of data sets correspond to the plurality of preset interaction index levels one by one.

9. The apparatus of claim 8, wherein, The interaction index determination module comprises: The level mean determination unit is configured to perform, according to the historical interaction index information of each historical request in the plurality of data sets, determine the level mean of the plurality of preset interaction index levels; The statistical analysis unit is configured to perform, according to the level mean of the plurality of preset interaction index levels and the first interaction index prediction information under the preset queue length, weighted average processing to obtain the interaction index information under the preset queue length.

10. The apparatus of claim 7, wherein, The device further comprises: The training sample obtaining unit is configured to perform obtaining request feature information of a sample request, and obtaining label information of the sample request, the request feature information including sample account information corresponding to the sample request, sample context information corresponding to the sample request, and identification information of a sample queue length corresponding to the sample request; the sample queue length is one of the plurality of preset queue lengths; the label information represents real interaction indicator information of the sample request under the sample queue length; The training prediction unit is configured to perform inputting the sample account information, the sample context information, and the identification information into a to-be-trained model, performing interaction indicator gear prediction processing, and obtaining second interaction indicator prediction information, the second interaction indicator prediction information representing a predicted probability of the content interaction indicator under the plurality of preset interaction indicator gears under the sample queue length; The model adjustment unit is configured to perform adjusting the to-be-trained model based on the second interaction indicator prediction information and the label information, and obtaining an interaction indicator prediction model; The interaction indicator gear prediction unit is configured to perform inputting account information corresponding to a content push request and context information corresponding to the content push request into the interaction indicator prediction model after obtaining the account information and the context information corresponding to the content push request, performing interaction indicator gear prediction processing, and obtaining the plurality of first interaction indicator prediction information under the plurality of preset queue lengths.

11. The apparatus of claim 10, wherein, The training sample obtaining unit includes: The request sampling subunit is configured to perform random sampling on online requests to obtain a sample request set, the sample request set including a plurality of sample requests with the same sample queue length; The identification determining subunit is configured to determine the identification information of each sample request in the sample request set according to the sample queue length.

12. The apparatus of claim 10, wherein, The training sample obtaining unit includes: The first push subunit is configured to perform content push processing based on the sample queue length in response to a plurality of sample requests to obtain a plurality of sample interaction indicator information; the plurality of sample interaction indicator information corresponds to the plurality of sample requests one by one; The target sample interaction indicator determining subunit is configured to determine target sample interaction indicator information corresponding to the sample queue length according to the plurality of sample interaction indicator information, the target sample interaction indicator information being the maximum value in the plurality of sample interaction indicator information; The label determining subunit is configured to perform discretization of the target sample interaction indicator information according to the plurality of preset interaction indicator gears to obtain label information of each sample request in the sample request set.

13. An electronic device, comprising: include: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the content push processing method of any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the content pushing processing method according to any one of claims 1 to 6.

Images