Push processing method, related device and medium

Through the prediction model combining the target object characteristics, the most appropriate push period and content is determined, which solves the problem of disconnection between content and timing in the prior art, and improves the accuracy and user experience of content push.

CN119071352APending Publication Date: 2024-12-03TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202310634821.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-31
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing content push platform lacks accuracy when determining the push timing and content, resulting in poor pushing results, especially failure to push high-quality content at a specific time, causing the content to be disconnected from the timing and affecting the user experience.

Method used

By obtaining the content set to be pushed and the time period set to be pushed, combining the characteristics of the target object, input the task completion probability prediction model, determine the most suitable push time period and content, and achieve matching content and timing.

Benefits of technology

Improve the accuracy and user experience of content push, ensure that the pushed content is achieved at the most appropriate time, and enhance users' attention and interactivity with the pushed content.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119071352A_ABST
    Figure CN119071352A_ABST
Patent Text Reader

Abstract

The invention provides a push processing method, a related device and a medium. The push processing method comprises the following steps: acquiring a to-be-pushed content set, wherein the to-be-pushed content set comprises a plurality of to-be-pushed contents; obtaining a to-be-pushed time period set, wherein the to-be-pushed time period set comprises a plurality of to-be-pushed time periods divided by a predetermined pushing period; performing first combination on the feature of each to-be-pushed content in the to-be-pushed content set and each to-be-pushed time period in the to-be-pushed time period set, and inputting each first combination and the object feature of the target object into a task completion probability prediction model to obtain a predicted first task completion probability; determining a first target combination in the plurality of first combinations based on the first task completion probability of each first combination; and taking the to-be-pushed time period in the first target combination as a target time period, and pushing a reminding message corresponding to the to-be-pushed content to the target object in the target time period. According to the embodiment of the invention, the content pushing accuracy is improved. The embodiment of the invention is applied to big data, data processing and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of big data, and in particular to a push processing method, related devices and media. Background Art

[0002] In order to avoid disturbing the target, certain frequency control measures are generally taken on content push platforms, such as pushing two pieces of content (such as short video reminder messages) to the target every day. When pushing, the push timing and push content are determined according to different models, resulting in a disconnect between the determined push content and the push timing. Some content must be pushed at a specific time to achieve a better push effect. The better content pushed without considering the push timing is often quite different from the better content pushed with the timing taken into account, resulting in inaccurate push. Summary of the invention

[0003] The embodiments of the present disclosure provide a push processing method, a related device and a medium, which can improve the accuracy of content push.

[0004] According to one aspect of the present disclosure, a push processing method is provided, including:

[0005] Acquire a set of content to be pushed, where the set of content to be pushed includes a plurality of content to be pushed;

[0006] Acquire a set of time periods to be pushed, wherein the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle;

[0007] Performing a first combination of the features of each of the contents to be pushed in the set of contents to be pushed and each of the time periods to be pushed in the set of time periods to be pushed, and inputting each of the first combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted first task completion probability, wherein the first task completion probability is a probability that the target object completes the task for the contents to be pushed in the time period to be pushed;

[0008] determining a first target combination among a plurality of the first combinations based on the first task completion probability of each of the first combinations;

[0009] The time period to be pushed in the first target combination is used as a target time period, and a reminder message corresponding to the content to be pushed is pushed to the target object during the target time period.

[0010] According to one aspect of the present disclosure, a push processing device is provided, comprising:

[0011] A first acquisition unit is used to acquire a set of content to be pushed, wherein the set of content to be pushed includes a plurality of content to be pushed;

[0012] A second acquisition unit is used to acquire a set of time periods to be pushed, wherein the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle;

[0013] A first input unit is used to perform a first combination of the features of each of the to-be-pushed content in the to-be-pushed content set and each of the to-be-pushed time periods in the to-be-pushed time period set, and input each of the first combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted first task completion probability, wherein the first task completion probability is a probability that the target object completes the task for the to-be-pushed content in the to-be-pushed time period;

[0014] a first determining unit, configured to determine a first target combination among a plurality of the first combinations based on the first task completion probability of each of the first combinations;

[0015] The pushing unit is used to use the time period to be pushed in the first target combination as a target time period, and push a reminder message corresponding to the content to be pushed to the target object during the target time period.

[0016] Optionally, the first input unit is specifically used to: input each of the first combinations together with the object characteristics of the target object and the scene characteristics of the target object into a task completion probability prediction model to obtain a predicted first task completion probability; wherein, the first task completion probability is the probability that the target object completes the task for the content to be pushed in the scene during the period to be pushed.

[0017] Optionally, the pushing unit is specifically used for:

[0018] Acquire a sub-period set to be pushed of the target period, wherein the sub-period set to be pushed includes a plurality of sub-periods to be pushed;

[0019] Acquire the first combination with the target time period from among the plurality of the first combinations, and select a screened first combination from among the first combinations with the target time period based on the first task completion probability;

[0020] Performing a second combination on each of the first combinations after screening and each of the sub-periods to be pushed in the sub-periods to be pushed, inputting each of the second combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted second task completion probability, wherein the second task completion probability is a probability that the target object completes the task for the content to be pushed in the sub-period to be pushed of the period to be pushed;

[0021] determining a second target combination among a plurality of the second combinations based on the second task completion probability of each of the second combinations;

[0022] The time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content, and in the target sub-time period of the target time period, a reminder message corresponding to the target content is pushed to the target object.

[0023] Optionally, the push unit is specifically used to: input each of the second combinations together with the object characteristics of the target object and the scene characteristics of the target object into a task completion probability prediction model to obtain a predicted second task completion probability; wherein, the second task completion probability is the probability that the target object completes the task for the content to be pushed, in the sub-period to be pushed of the period to be pushed, and in the scenario.

[0024] Optionally, the push processing device further includes:

[0025] A third acquisition unit is specifically used to acquire additional content generated on the content push platform between the generation time of the content set to be pushed and the current time;

[0026] The adding unit is specifically used to add the combination of the feature of the added content and the target time period to the first combination after screening.

[0027] Optionally, the pushing unit is specifically used for:

[0028] Determining, among the first combinations with the target time period, a first number of the first combinations having a probability of completing the first task greater than a first threshold;

[0029] If the first number is not less than the second threshold, the first number of the first combinations is taken as the first combination after screening; otherwise, the first combinations before the second threshold are taken in descending order of the probability of completing the first task as the first combination after screening.

[0030] Optionally, the task completion probability prediction model is trained in the following manner:

[0031] Acquire a sample push event set, wherein the sample push event set includes a plurality of sample push events, the sample push event includes a sample object feature, a sample content feature to be pushed, a sample to be pushed period, and a sample to be pushed sub-period, the sample push event has a task training label, and the task training label indicates whether the sample object completes the task for the sample content to be pushed and the sample to be pushed sub-period in the sample to be pushed period;

[0032] Extracting a first sample and a second sample from the sample push event, wherein the first sample includes the sample object feature, the sample content feature to be pushed, and the sample to be pushed period, and the second sample includes the sample object feature, the sample content feature to be pushed, the sample to be pushed period, and the sample to be pushed sub-period;

[0033] The first sample and the second sample extracted from each of the sample push events are input into the task completion probability prediction model for training.

[0034] Optionally, the step of inputting the first sample and the second sample extracted from each sample push event into the task completion probability prediction model for training includes:

[0035] For each of the first sample and the second sample, for a single task, based on the training label of the single task and the first task completion probability predicted by the task completion probability prediction model, determine an error subfunction of the single task;

[0036] Determining an error function of the sample based on a plurality of error sub-functions of the tasks;

[0037] Based on the error function of each of the samples, the task completion probability prediction model is trained.

[0038] Optionally, the task completion probability prediction model includes a vectorization layer, a cascade layer, a plurality of expert sub-models, a plurality of gating nodes corresponding to the plurality of tasks, and a plurality of prediction sub-models;

[0039] The first input unit is specifically used for:

[0040] Vectorize each of the first combinations and the object feature of the target object, and concatenate the vectorization results to obtain a first feature vector;

[0041] Inputting the first feature vector into a plurality of the expert sub-models;

[0042] Through the gating nodes corresponding to each of the tasks, the outputs of the plurality of expert sub-models are weighted and summed to obtain a second feature vector corresponding to the task;

[0043] The second feature vector corresponding to the task is input into the prediction model corresponding to the task to obtain the first task completion probability corresponding to the task.

[0044] Optionally, the first input unit is specifically used for:

[0045] Obtaining a weight vector of the gating node, wherein the weight vector indicates weights of a plurality of the expert sub-models;

[0046] The outputs of the plurality of expert sub-models are weighted and summed using the weight vector through the gating node to obtain a second feature vector corresponding to the task.

[0047] Optionally, the first input unit is specifically used for:

[0048] Obtain a first matrix, wherein the number of rows of the first matrix is ​​the number of the expert sub-models, and the number of columns is the dimension of the first eigenvector;

[0049] The first matrix is ​​multiplied by the transpose of the first eigenvector, and a normalized exponential process is performed on the multiplication result to obtain the weight vector of the gated node.

[0050] Optionally, the first input unit is specifically used for:

[0051] Taking the characteristics of the content to be pushed, the time period to be pushed, and the object characteristics as basic characteristics for pushing;

[0052] If the pushed basic feature is a character feature, input the character feature into an embedding layer to obtain a vectorization result corresponding to the character feature;

[0053] If the pushed basic feature is a numerical feature, numerical mapping is performed on the numerical feature to obtain a vectorized result corresponding to the numerical feature;

[0054] If the pushed basic feature is an indefinite length sequence feature, the indefinite length sequence feature is pooled to obtain a vectorized result corresponding to the indefinite length sequence feature.

[0055] Optionally, the first acquiring unit is specifically configured to:

[0056] Obtaining a first label of the target object;

[0057] Get the second tag of the original content in the original content set;

[0058] The to-be-pushed content set is screened out from the original content set based on the matching degree between the first tag and the second tag.

[0059] Optionally, the first acquiring unit is specifically configured to:

[0060] Based on the matching degree, selecting matching content in which the first tag and the second tag match from the original content set;

[0061] Obtaining multiple push indicators for each of the matching contents;

[0062] Based on the plurality of push indicators, the content to be pushed is screened out from the matching content to form the content set to be pushed.

[0063] Optionally, the first determining unit is specifically configured to:

[0064] Acquire the focus task of the target object from among the multiple tasks;

[0065] Acquire the first task completion probability of the focus task of the first combination from the first task completion probabilities of the plurality of tasks of the first combination;

[0066] A first target combination is determined among the plurality of first combinations according to the first task completion probabilities of the focus tasks of the plurality of first combinations.

[0067] Optionally, the first determining unit is specifically configured to:

[0068] If there are multiple tasks of interest, determining a first overall first task completion probability of the first combination based on the first task completion probabilities of the multiple tasks of interest in each of the first combinations;

[0069] A first target combination is determined among the plurality of first combinations based on the first overall first task completion probabilities of the plurality of first combinations.

[0070] According to one aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the push processing method as described above when executing the computer program.

[0071] According to one aspect of the present disclosure, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the push processing method as described above is implemented.

[0072] According to one aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. The computer program is read and executed by a processor of a computer device, so that the computer device executes the push processing method as described above.

[0073] In the disclosed embodiment, multiple contents to be pushed and multiple time periods to be pushed are combined in pairs, and the contents to be pushed and time periods to be pushed in each combination are input into the prediction model together with the object features, and the prediction model predicts the probability of the target object completing the task (such as liking, commenting, etc.) for the contents to be pushed at the time to be pushed, determines the appropriate time period and content to be pushed based on the probability, and pushes the content in the determined time period to be pushed. Since the disclosed embodiment combines possible contents to be pushed and possible time periods to be pushed in pairs, and screens based on the first task completion probability obtained from each combination, the contents to be pushed and time periods to be pushed are not isolated in the screening process, but are interconnected, thereby achieving the optimization of contents to be pushed under time limit and the optimization of time periods to be pushed under content limit, and improving the accuracy of content push.

[0074] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or understood by practicing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The accompanying drawings are used to provide further understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation on the technical solution of the present disclosure.

[0076] Figure 1 is a system architecture diagram of the push processing method of the embodiment of the present disclosure;

[0077] Figure 2A -C is a schematic diagram of an interface in which an embodiment of the present disclosure is applied to a scenario where a video content push message is pushed;

[0078] Figure 3 is a flowchart of a push processing method according to an embodiment of the present disclosure;

[0079] Figure 4 yes Figure 3 A flowchart of obtaining the content set to be pushed in step 310;

[0080] Figure 5 is a schematic diagram of screening out a set of content to be pushed based on the matching degree between the first tag and the second tag;

[0081] Figure 6 yes Figure 4 A flowchart of screening the content set to be pushed in step 430;

[0082] Figure 7 This is a schematic diagram of selecting content to be pushed based on multiple push indicators;

[0083] Figure 8 yes Figure 3 A flow chart of determining the number of time periods to be pushed in step 320;

[0084] Fig. 9 yes Figure 3 A schematic diagram of performing the first combination in step 330;

[0085] Fig.10 yes Figure 3 In step 330, a flowchart of using the task completion probability prediction model to obtain a predicted first task completion probability is obtained;

[0086] Fig.11A is a schematic diagram of step 330;

[0087] Fig. 11B is Fig.11A A schematic diagram of adding scene features of the target object on the basis of

[0088] Fig. 12A -C is a schematic diagram of generating vectorized results for different types of push basic features;

[0089] Fig.13 yes Fig.10 A flowchart of step 1030 in FIG.

[0090] Fig.14 yes Fig.13 A flowchart of step 1310 in FIG.

[0091] Fig.15 yes Figure 3 A flowchart of determining a first target combination among a plurality of first combinations in step 340;

[0092] Fig.16 is a flow chart for determining a first target combination based on a plurality of focus tasks;

[0093] Fig.17 yes Figure 3 A flowchart of pushing a reminder message corresponding to the content to be pushed to the target object in the target time period in step 350;

[0094] Fig.18 yes Fig.17 A schematic diagram of selecting a first combination based on the first task completion probability in step 1720;

[0095] Fig.19 yes Fig.17 A flow chart of selecting the first combination after screening in step 1720;

[0096] Fig. 20 is Fig.17 A flow chart for adding content is added after step 1720;

[0097] Fig.21 It is a schematic diagram of adding the added content to the first combination after screening;

[0098] Fig. 22 yes Fig.17 A schematic diagram of performing a second combination in step 1730;

[0099] Fig.23 is a schematic diagram of the second target combination;

[0100] Fig.24 It is a flow chart for training the task completion probability prediction model;

[0101] Fig.25 is an example diagram of extracting a first sample and a second sample from a sample push event;

[0102] Fig.26 yes Fig.24 A flowchart of the training task completion probability prediction model in step 2430;

[0103] Fig. 27 It is an implementation detail diagram of the push processing method of the embodiment of the present disclosure.

[0104] Fig.28 is a module diagram of a push processing device according to an embodiment of the present disclosure;

[0105] Fig.29 According to the embodiment of the present disclosure Figure 3 The terminal structure diagram of the push processing method shown;

[0106] Fig.30 According to the embodiment of the present disclosure Figure 3 The structure diagram of the server 140 of the push processing method is shown. DETAILED DESCRIPTION

[0107] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0108] Before further describing the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are described. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations:

[0109] Deep neural network: It is a multi-layer unsupervised neural network. Deep neural network uses the output features of the previous layer as the input of the next layer for feature learning. After layer-by-layer feature mapping, the features of the existing spatial samples are mapped to another feature space, so as to learn to have better feature expression for the existing input. Deep neural network has multiple nonlinear mapping feature transformations and can fit highly complex functions.

[0110] Gradient descent algorithm: A method for minimizing the loss function. It calculates the gradient of the loss function at the current position and updates the parameters in the direction of gradient descent to find the optimal solution that minimizes the value of the loss function.

[0111] Normalized exponential (softmax function): compresses a K-dimensional vector containing arbitrary real numbers into another K-dimensional vector so that each element ranges between (0,1) and the sum of all elements is 1.

[0112] System architecture and scenario description of the application of the embodiments of the present disclosure

[0113] Figure 1 The system architecture diagram of the push processing method according to the embodiment of the present disclosure includes: a target terminal 110, the Internet 120, a gateway 130, and a server 140.

[0114] The object terminal 110 is a device used by the object to view the reminder message corresponding to the pushed content. It includes various forms such as desktop computers, laptops, PDAs (personal digital assistants), mobile phones, car terminals, home theater terminals, and dedicated terminals. In addition, it can be a single device or a collection of multiple devices. For example, multiple devices are connected through a local area network and work together using a common display device to form a terminal. The terminal can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0115] The gateway 130 is also called an internetwork connector or a protocol converter. The gateway 130 realizes network interconnection at the transport layer and is a computer system or device that acts as a converter. The gateway 130 is a translator between two systems that use different communication protocols, data formats or languages, or even completely different architectures. At the same time, the gateway 130 can also provide filtering and security functions. The message sent by the object terminal 110 to the server 140 is sent to the corresponding server 140 through the gateway 130. The message sent by the server 140 to the object terminal 110 is also sent to the corresponding object terminal 110 through the gateway 130.

[0116] The server 140 refers to a computer system that can provide content message push services to the target terminal 110. Compared with the target terminal 110, the server 140 has higher requirements in terms of stability, security, performance, etc. The server 140 can be a high-performance computer in a network platform, a cluster of multiple high-performance computers, a part of a high-performance computer (such as a virtual machine), a combination of parts of multiple high-performance computers (such as virtual machines), etc. The server 140 can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0117] The embodiments of the present disclosure can be applied in various scenarios, such as Figure 2A -C shows a scenario of viewing video content push messages in an instant short messaging application, etc.

[0118] like Figure 2A The interface of the instant short message application in the object terminal 110 is shown. The instant short message application can be used to communicate with multiple contacts and can also be used to watch videos. After triggering the "Short Message" option in the function bar below, a short message list is displayed. The short message list includes a message bar with multiple contacts and a message push bar with video recommendations. At 10:00 am, there is no new message push in the message recommendation bar for video recommendations. Until 2:00 pm, the short message interface of the instant short message application is as follows Figure 2B shown.

[0119] exist Figure 2B In the video recommendation message push bar, a new message is pushed to the subject, and the new message shows: "You have received an unread message, recommending you video A that you may be interested in". The subject can watch the recommended video A by clicking the new message in the message push bar or triggering the "Video" option in the function bar below. When the subject triggers the "Video" option, the interface of the instant short message application is as follows: Figure 2C shown.

[0120] exist Figure 2C In the example, the instant short message application opens the video interface and plays the recommended video A. The number of likes, favorites, and comments that video A has received before the current time is also displayed in the playback interface. The subject can also like, favorite, and comment on video A in the interface. The video content pushed to the subject and the push time are determined by server 140. If the content pushed to the subject is not the content that the subject wants to watch at the time of the push, then even if the pushed content is of interest to the subject, the subject will not click to watch the content after receiving the push message. Therefore, when pushing a reminder message corresponding to the content to the subject, it is very important to consider the push timing. The disclosed embodiment can improve the accuracy of content push by considering both the time period to be pushed and the content to be pushed.

[0121] General description of the disclosed embodiments

[0122] According to an embodiment of the present disclosure, a push processing method is provided.

[0123] The push processing method is a process of determining the target content and the target time period for the target object, and pushing the reminder message corresponding to the target content to the object terminal 110 of the target object during the target time period. The target object is the object that wants to view the target content. The disclosed embodiment can be applied to the push of different types of content, and the target content can be pictures, videos, articles, applications and other types. The target time period is the time period for pushing the reminder message corresponding to the target content to the target object. For example, if the target time period is 8:00-9:00 and the target content is content A, then a reminder message of content A is sent to the target object at some time between 8:00-9:00. In order to avoid disturbing the target object, the push frequency is generally set. For example, within a day, two time periods are selected as target time periods.

[0124] The push processing method of the embodiment of the present disclosure is executed on the server 140. After the target content and the target time period are determined, the reminder message corresponding to the target content is pushed to the target terminal 110 via the network and the Internet 120 during the target time period.

[0125] At present, in one method, the target content and the target time period are determined separately through independent models. The push effect of the target content determined by this method to the target object during the target time period may not be good. For example, the target time period is determined to be 6:00-8:00 pm through model A; the target content is determined to be a learning content through model B. However, in the time period of 6:00-8:00, the target object does not want to watch learning content. Therefore, when the reminder message of the target content is pushed to the target object during this time period, the target object is not interested and the push effect is poor. If the reminder message of the target content is pushed to the target object between 10:00-12:00 in the morning, and the target object is studying at the time, it is more likely to watch the target content through the reminder message. The disclosed embodiment jointly determines the target content and the target time period based on the combination of the content to be pushed and the time period to be pushed, so that pushing the target content during the target time period can achieve a better push effect and improve the accuracy of the push.

[0126] like Figure 3 As shown, according to one embodiment of the present disclosure, the push processing method includes:

[0127] Step 310: Obtain a set of content to be pushed, where the set of content to be pushed includes a plurality of content to be pushed;

[0128] Step 320: Obtain a set of time periods to be pushed, where the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle;

[0129] Step 330: Perform a first combination of the features of each content to be pushed in the content set to be pushed and each time period to be pushed in the time period set to be pushed, and input each first combination together with the object features of the target object into a task completion probability prediction model to obtain a predicted first task completion probability;

[0130] Step 340: determining a first target combination from a plurality of first combinations based on the first task completion probability of each first combination;

[0131] Step 350: The time period to be pushed in the first target combination is used as the target time period, and a reminder message corresponding to the content to be pushed is pushed to the target object during the target time period.

[0132] The above steps 310 - 350 are described in detail below.

[0133] Detailed description of step 310

[0134] In step 310, the content to be pushed is the content that the target object may be interested in. The content to be pushed is selected from the set of content to be pushed as the target content.

[0135] In one embodiment, Figure 4 As shown, obtain the content set to be pushed, including:

[0136] Step 410: Obtain a first tag of the target object;

[0137] Step 420: Obtain a second tag of the original content in the original content set;

[0138] Step 430: Filter the content set to be pushed from the original content set based on the matching degree between the first tag and the second tag.

[0139] In step 410, the first tag refers to a content tag that the target object may be interested in.

[0140] In one embodiment, the first tag of the target object may be obtained by obtaining the first tag selected by the target object in advance.

[0141] Note that when obtaining the first tag of the target object, the target object's consent must be obtained in advance. In addition, the collection, use and processing of the first tag will comply with relevant laws, regulations and standards. When obtaining the target object's consent, the target object's separate permission or separate consent can be obtained through a pop-up window or by jumping to a confirmation page.

[0142] When the target object is using the software that pushes the content to be pushed, it is necessary to select the tag of interest to it as the first tag from the many tags provided by the software system. For example, the many tags provided by the software system include: "learning", "pets", "games", "sports", "movies", "basketball", "music", "health", etc. The target object selects "pets", "games", "movies", "music" as its first tag according to its own preferences. When obtaining the first tag of the target object, the first tag selected by the target object can be directly obtained.

[0143] The advantage of directly acquiring the first tag selected by the target object autonomously is that the efficiency of acquiring the first tag is high, and the first tag must be the type of content that the target object is interested in, which improves the accuracy of acquiring the first tag.

[0144] In another embodiment, the first tag of the target object can be obtained through the historical viewing log of the target object in a predetermined time period before the current time point. The historical viewing log refers to the relevant records of the pushed content viewed by the target object in a predetermined time period before the current time period.

[0145] The scheduled time period can be one hour, one day, one week, one month, etc. The scheduled time period can be determined based on the frequency with which the target object views the pushed content. If the target object views the pushed content frequently, then the viewing records of a shorter time period can be used to obtain the types of content that the target object is interested in in the recent period; if the target object views the pushed content less frequently, then the viewing records of a longer time period need to be used to ensure that there is enough data to obtain the first tag of the target object. The scheduled time period can also be determined based on the type of pushed content. For example, if the pushed content is video content, a longer time period can be used as the scheduled time period to obtain the long-term preferences of the target object; if the pushed content is news content, since news is time-sensitive, a shorter time period needs to be used as the scheduled time period.

[0146] Please note that obtaining the target object's historical viewing logs also requires obtaining the target object's consent.

[0147] The content recorded in the historical viewing log may include: viewing time, viewing content, content type, whether it is watched completely, whether there are other operations (such as liking, commenting, and collecting content), etc. For example, Table 1 is an example of the historical viewing log of the target object within one day before the current time point.

[0148]

[0149]

[0150] Table 1

[0151] As shown in Table 1, the target object watched content A on May 1, 2023. The content type of content A is "movie". The target object watched content A completely, but did not perform other operations on content A. "Whether to watch completely" and "Whether to perform other operations" are tasks for the watched content in a viewing record. Tasks can also be "Whether to like" and "Whether to comment".

[0152] According to the historical viewing logs in Table 1, the first label of the target object can be obtained.

[0153] In one embodiment, obtaining the first tag of the target object through historical viewing logs includes: determining the target object's preference for pushed content by counting the number of task completions of viewing logs of each content type in the historical viewing logs; and using the preferred content type as the first tag of the target object.

[0154] Since the historical viewing log records whether the target object has watched the content completely and whether there are other operations. Therefore, the target object's attitude towards a viewing content can be preliminarily judged through the historical viewing log. For example, in the historical viewing log shown in Table 1, the target object watched content D completely and performed operations such as liking or commenting on content D. It can be inferred that the target object is very interested in content D. However, the target object neither watched content C completely nor performed any operations. It can be inferred that the target object is not interested in content C. By counting the number of task completions for viewing content of the same content type, the target object's interest in a content type can be determined.

[0155] For example, Table 2 shows the statistics of the number of task completions for each content type in the historical viewing log in Table 1.

[0156] Content Type Views Other operations Movie 12 8 music 10 6 read 4 1 pet 15 10 …… …… ……

[0157] Table 2

[0158] As shown in Table 2, the target object has watched the content of the content type "movie" 12 times and performed other operations 8 times.

[0159] Note that when obtaining the viewing content of each content type in the historical viewing log, the consent of the target object must be obtained.

[0160] In one embodiment, by counting the number of task completions of each content type in the historical viewing log, the first label of the target object is determined, including: calculating the average number of task completions corresponding to each content type, and determining the first label of the target object based on the average value of the operation attributes of all content types. For example, according to the statistics of the number of task completions of each content type in Table 2, the average number of task completions for movie-type content is (12+8) / 3=10; the average number of task completions for music-type content is (10+6) / 2=8. After determining the average number of task completions for each content type, the content types are ranked in descending order according to the average value, and the first predetermined number of content types in the ranking are selected as the first label of the target object. For example, after calculating the average number of task completions of each content type in Table 2, the content types are ranked according to the average value, and the top three content types in the ranking are selected as the first label of the target object, and the first label of the target object is "pet", "movie", and "music". The advantage of using the average value to determine the first label of the target object is that it fairly reflects the impact of each task on the target object's preference, and improves the fairness of determining the first label.

[0161] In another embodiment, by counting the number of task completions of each content type in the historical viewing log, the first label of the target object is determined, including: setting a weight for each task completion number; and determining the first label of the target object based on the weighted average ranking of the task completions of all content types. The reason for setting weights for different task completion times is that different tasks may have different effects on the interest of the target object. For example, the interest of the target object in the content that has performed other operations (such as likes, comments, etc.) may be higher than the interest in the content that has only been watched completely. Therefore, the influence of different tasks on the interest of the target object can be expressed by setting weights. For example, the weight of "full viewing times" is 0.3, and the weight of "other operations times" is 0.7. The weighted average of the number of task completions of different content types is calculated based on the weights. For example, in Table 2, the weighted average of the number of task completions of movie-type content is 12*0.3+8*0.7=9.2; the weighted average of the number of task completions of music-type content is 10*0.3+6*0.7=7.2. After determining the weighted average of the number of task completions for each content type, the content types are ranked in descending order according to the weighted averages, and the first predetermined number of content types in the ranking are selected as the first labels of the target object. For example, the top three content types in Table 2 are selected as the first labels of the target object, so the first labels of the target object that can be obtained are "pets", "movies", and "music". The advantage of using the weighted average result to determine the first label of the target object is that different weights can be set for different task completion times according to the actual application situation, which improves the flexibility of determining the first label.

[0162] Acquiring the first label of the target object through historical viewing logs can determine the content preference of the target object in the recent period, which is conducive to improving the accuracy of acquiring the first label.

[0163] In step 420, the original content set includes original contents of various content types to meet the push requirements to target objects with different preferences. The second tag indicates the type of the original content.

[0164] In one embodiment, obtaining the second tag of the original content in the original content set includes: obtaining the second tag set for the original content by the publishing object of the original content.

[0165] When the publishing object publishes an original content, a second tag may be set for the original content. For example, when the publishing object publishes an original content, the tags set for the original content are: "music", "guitar".

[0166] Note that when obtaining the second tag of the original content, the consent of the publisher of the original content must be obtained.

[0167] When acquiring the second tag of the original content, the tag set by the publishing object can be directly acquired. The advantage of this embodiment is that the efficiency of acquiring the second tag is improved.

[0168] In another embodiment, obtaining the second label of the original content in the original content set includes: inputting the original content into a label prediction model; and the label prediction model outputting the second label of the original content. For example, inputting an original content into the label prediction model, the second label of the original content is obtained as: "movie", "funny".

[0169] The advantage of outputting the second label through the label prediction model is that the accuracy of obtaining the second label is improved.

[0170] Note that when obtaining original content and inputting the original content into the label prediction model, the consent of the publisher of the original content must be obtained.

[0171] In step 430, the matching degree between the first tag and the second tag represents the matching degree between the original content and the target object preference, and the original content whose matching degree meets the predetermined condition is added to the set of content to be pushed.

[0172] In one embodiment, the matching degree between the first label and the second label is the proportion of the overlapping labels of the second label and the first label in all the second labels. Figure 5 As shown, the first tags of the target object include: "Pet", "Game", "Movie", and "Music". The second tags of the original content A in the original content set are: "Movie", and "Funny", among which one tag belongs to the first tag of the target object. Therefore, the matching degree between the second tag of the original content A and the first tag is 50%. Similarly, the second tags of the original content B all belong to the first tag of the target object, so the matching degree between the second tag of the original content B and the first tag is 100%. Among the three second tags of the original content C, "Pet" and "Movie" belong to the first tag of the target object, so the matching degree between the second tag of the original content C and the first tag is 66.7%. Among the three second tags of the original content D, "Game" belongs to the first tag of the target object, so the matching degree between the second tag of the original content D and the first tag is 33.3%. The original content E has no second tag that matches the first tag, so the matching degree is 0. The advantage of determining the matching degree by the proportion of overlapping tags in all second tags is that the matching degree between the original content and the target object preference can be quickly determined by overlapping tags, which improves the efficiency of obtaining the content set to be pushed.

[0173] The original content whose matching degree meets the predetermined condition is added to the set of content to be pushed. In one embodiment, the predetermined condition is to add the original content whose matching degree with the first tag is greater than or equal to the preset matching degree threshold to the set of content to be pushed. For example, the preset matching degree threshold is 50%. Figure 5 The matching degree of the first label and the second label shown in the figure adds the original content with a matching degree greater than or equal to 50% to the set of content to be pushed. Therefore, the set of content to be pushed includes original content A (matching degree 50%), original content B (matching degree 100%), and original content C (matching degree 66.7%). In this embodiment, the advantage of determining the set of content to be pushed based on the matching degree threshold is that it ensures that the content to be pushed in the set of content to be pushed is content with a high matching degree with the target object, thereby improving the accuracy of determining the pushed content from the content to be pushed.

[0174] In another embodiment, the preset condition may be: ranking the matching degree of the original content; setting a fixed number, and adding the first fixed number of original contents to the set of contents to be pushed according to the order of the matching degree ranking from high to low. For example, if the fixed number is set to 10, the original contents with the top 10 matching degrees with the first tag are selected from the original content set to be added to the set of contents to be pushed. When the matching degree of the first tag of the target object and the second tag of the original content are both low, if the content to be pushed is screened according to the matching degree threshold, the number of contents in the set of contents to be pushed may not meet the requirements of pushing to the target object. Therefore, through this embodiment, it can be ensured that the number of contents to be pushed in the set of contents to be pushed meets the requirements of pushing to the target object. Through the analysis of the reaction of the target object to the content in the set of contents to be pushed, a more accurate first tag can be obtained, so as to improve the matching degree of the target object and the original content in subsequent screening, thereby obtaining content with a higher matching degree. Therefore, in this embodiment, the advantage of determining the set of contents to be pushed based on the matching degree ranking is that it provides data support for obtaining a more accurate first tag of the target object, thereby gradually improving the accuracy of determining the content to be pushed in the set of contents to be pushed.

[0175] In one embodiment, if Figure 6 As shown, based on the matching degree between the first tag and the second tag, the content set to be pushed is screened out from the original content set, including:

[0176] Step 610: based on the matching degree, filter out matching content having a first tag and a second tag matching from the original content set;

[0177] Step 620: Obtain multiple push indicators for each matching content;

[0178] Step 630: Based on multiple push indicators, select the content to be pushed from the matching content to form a set of content to be pushed.

[0179] In step 610, the process of selecting matching content from the original content set based on the matching degree between the first tag and the second tag is the same as the process of selecting the content set to be pushed in step 430 in the above embodiment, and will not be repeated here.

[0180] In step 620, the push index of the matching content refers to an index used to evaluate the push effect of the content. The push effect refers to the feedback of the target object on the reminder message corresponding to the pushed content. For example, whether the target object watches the pushed content through the reminder message, whether the target object completes watching the pushed content, whether the target object likes or comments on the pushed content, etc. The push index is to count the push effect of the content into specific values ​​in order to evaluate the push effect. The push index can be a push conversion rate, a completion rate, a like rate, or a comment rate, etc.

[0181] When obtaining multiple push indicators for each matching content, you must obtain the consent of the recipients of the matching content.

[0182] For example Figure 7 As shown, the obtained matching contents include: matching content A, matching content B, and matching content C. The push indicators of each matching content are obtained respectively. The push indicators of matching content A include: push conversion rate (0.72), completion rate (0.56), like rate (0.68), and comment rate (0.46), etc. Similarly, the push indicators of matching content B and matching content C can be obtained.

[0183] In step 630, the content to be pushed is selected from the matching content according to the multiple push indicators. Since the push indicator is a numerical representation of the push effect of the content, the push effect obtained after the content to be pushed determined based on the push indicator is pushed to the target object may be better. Figure 7 In the example, according to the push indicators of the three matching contents, the content set to be pushed finally obtained by screening includes matching content A and matching content B.

[0184] In one embodiment, based on multiple push indicators, content to be pushed is screened out from the matching content to form a set of content to be pushed, including: determining the push score of the matching content based on multiple push indicator values; and screening out the content to be pushed based on the push scores of all matching content.

[0185] In one embodiment, the push score of the matching content may be determined by calculating the sum of multiple push indicator values. Figure 7In the example, the push score of matching content A is 0.72+0.56+0.68+0.46=2.42; the push score of matching content B is 0.66+0.72+0.77+0.57=2.72; and the push score of matching content C is 0.32+0.45+0.42+0.39=1.58. The advantage of using the sum to calculate the push score is that it equally reflects the impact of different push indicators on the matching content.

[0186] In another embodiment, the push score of the matching content can be determined by calculating the weighted sum of multiple push indicator values. In this embodiment, it is necessary to set a weight for each push indicator. Based on the above example, the weight of the push conversion rate is 0.7, the weight of the completion rate is 0.5, the weight of the like rate is 0.6, and the weight of the comment rate is 0.6. The push score of matching content A is 0.72*0.7+0.56*0.5+0.68*0.6+0.46*0.6=1.468; the push score of matching content B is 0.66*0.7+0.72*0.5+0.77*0.6+0.57*0.6=1.626; the push score of matching content C is 0.32*0.7+0.45*0.5+0.42*0.6+0.39*0.6=0.935. The advantage of using weighted sum to calculate the push score is that the weights of different push indicators can be flexibly set according to the actual application situation, which improves the flexibility of screening the content to be pushed.

[0187] After determining the push scores of all matching contents, the contents to be pushed are filtered according to the push scores and the predetermined conditions. The predetermined conditions for filtering may be:

[0188] Set a push score threshold, and use the matching content whose push score is greater than the push score threshold as the content to be pushed;

[0189] Alternatively, the matching contents are sorted from high to low according to the push scores, and the first predetermined number of matching contents are used as the contents to be pushed.

[0190] The method of filtering the content to be pushed according to the push score is similar to the method of filtering the content to be pushed according to the matching degree between the first tag and the second tag in the aforementioned embodiment, and will not be repeated here.

[0191] The advantage of determining the content to be pushed based on the push scores of each matching content is that the calculation speed is fast, which improves the efficiency of determining the content set to be pushed.

[0192] The advantage of the embodiment of steps 610-630 is that the original content is further screened based on the push index, so that the screened content to be pushed has a better push effect among the matching content, which is conducive to improving the push effect.

[0193] The advantage of the embodiment of steps 410-430 is that the original content is preliminarily screened according to the tag of the target object to ensure that the content to be pushed has a high degree of matching with the target object, saving computing resources and improving the matching degree between the content to be pushed and the target object.

[0194] Detailed description of step 320

[0195] The set of time periods to be pushed is a set of candidate time periods for content push for a target object, including multiple time periods to be pushed. The scheduled push cycle is divided into multiple time periods to be pushed according to a predetermined number. The scheduled push cycle refers to the time period for pushing to the target object. For example, the target object is pushed once or more times every day, or once or more times every week. The predetermined number refers to the number of multiple time periods to be pushed divided in the preset scheduled push cycle.

[0196] In one embodiment, Figure 8 As shown, the number of time periods to be pushed in the time period set to be pushed is determined by:

[0197] Step 810: Obtain prediction accuracy requirements;

[0198] Step 820: Determine the number of time periods to be pushed based on the number of contents to be pushed in the content set to be pushed and the prediction accuracy requirement.

[0199] The prediction accuracy requirement refers to the accuracy requirement for the probability of completion of the first task. The accuracy requirement can be 0.1, 0.01, etc. An accuracy requirement of 0.1 means that the probability of completion of the task is constrained to the tenth place; an accuracy requirement of 0.01 means that the probability of completion of the task is constrained to the hundredth place, and so on. The higher the accuracy requirement, the more accurate the result of the probability of completion of the first task is, the smaller the limited time range may be, and the more time periods to be pushed may be.

[0200] The more content to be pushed in the content set to be pushed, the larger the limited time range may be, and the more time periods to be pushed. This is because in the subsequent steps, the content to be pushed needs to be combined with the time periods to be pushed. If the number of content to be pushed is too large, and the number of time periods to be pushed is too large due to the small time range, then the number of combinations is large and the amount of calculation is too large, which can easily lead to reduced calculation efficiency and memory utilization.

[0201] In one embodiment, based on the number of contents to be pushed in the set of contents to be pushed and the prediction accuracy requirement, determining the number of time periods to be pushed includes:

[0202] Determining a first score based on the number of contents to be pushed;

[0203] determining a second score based on the prediction accuracy requirement;

[0204] determining a total score based on the first score and the second score;

[0205] Based on the total score, the number of time slots to be pushed is determined.

[0206] Based on the number of contents to be pushed, the first score may be determined by using a first comparison table. Table 3 shows an example of a first comparison table.

[0207] The number of contents to be pushed First Score 1-100 100 101-200 90 201-300 80 …… More than 1000 10

[0208] Table 3

[0209] Referring to Table 3, if the number of contents to be pushed is 230, the corresponding first score is 80.

[0210] Based on the prediction accuracy requirement, the second score may be determined by using a second comparison table. Table 4 shows an example of a second comparison table.

[0211] Prediction accuracy requirements Second score 0.0001 100 0.001 80 0.01 60 0.1 40

[0212] Table 4

[0213] Referring to Table 4, if the prediction accuracy requirement is 0.01, the second score corresponding to the prediction accuracy requirement is 60.

[0214] The total score is calculated based on the first score and the second score. In one embodiment, the total score is calculated using an average. For example, if the first score is 80 and the second score is 60, then the total score is (80+60) / 2=70. The advantage of using an average to calculate the total score is that the number of contents to be pushed and the prediction accuracy requirement have the same effect on the calculation of the total score, which improves the fairness of determining the number of time periods to be pushed.

[0215] In another embodiment, the total score is calculated using a weighted average. In this embodiment, first, weights need to be set for the first score and the second score respectively. For example, the weight of the first score is 0.6, the weight of the second score is 0.4, the first score is 80, and the second score is 60, then the total score is 80*0.6+60*0.4=72. The advantage of using a weighted average to calculate the total score is that different weights can be flexibly set for the first score and the second score according to the needs of the actual application, which improves the flexibility of determining the number of time periods to be pushed.

[0216] The number of time periods to be pushed is determined according to the total score. The higher the total score, the more time periods to be pushed are required.

[0217] In one embodiment, the number of time periods to be pushed can be determined by a third comparison table of total scores and the number of time periods to be pushed. Table 5 shows an example of the third comparison table.

[0218] Total score Number of time slots to be pushed 91-100 20 81-90 18 71-80 16 61-70 14 …… 12

[0219] Table 5

[0220] Reference Figure 5 If the total score is 75, then the number of waiting-for-pushing periods is 16, that is, the scheduled push cycle is divided into 16 waiting-for-pushing periods.

[0221] The method of determining the number of time periods to be pushed through the third comparison table has the advantages of being easy to query and having high processing efficiency.

[0222] This embodiment can quickly determine the number of time periods to be pushed according to the scores, thereby improving the efficiency of dividing the scheduled push cycle into multiple time periods to be pushed.

[0223] The advantage of the embodiment of steps 810-820 is that the number of time periods to be pushed is determined according to the number of contents to be pushed and the accuracy requirement for the completion probability of the first task, so that the number of time periods to be pushed can meet the requirements of the task, thereby improving the accuracy of dividing the time periods to be pushed.

[0224] Detailed description of step 330

[0225] In step 330, firstly, the characteristics of each content to be pushed in the content set to be pushed and each time period to be pushed in the time period to be pushed are combined in a first combination. The first combination means combining the characteristics of each content to be pushed with all the time periods to be pushed in the time period to be pushed in sequence. Fig. 9As shown, the contents to be pushed include: content A to be pushed, content B to be pushed, and content C to be pushed; the time periods to be pushed include: 0:00-4:00, 4:00-8:00, 8:00-12:00, 12:00-16:00, 16:00-20:00, and 20:00-24:00. Combine the features of content A to be pushed with the time periods to be pushed in the time period concentration to obtain multiple first combinations: (features of content A to be pushed, 0:00-4:00), (features of content A to be pushed, 4:00-8:00), ..., and (features of content A to be pushed, 20:00-24:00); then combine the features of content B to be pushed with the time periods to be pushed in the time period concentration to obtain multiple first combinations: (features of content B to be pushed, 0:00-4:00), (features of content B to be pushed, 4:00-8:00), ..., and (features of content B to be pushed, 20:00-24:00); then combine the features of content C to be pushed with the time periods to be pushed in the time period concentration to obtain multiple first combinations: (features of content C to be pushed, 0:00-4:00), (features of content C to be pushed, 4:00-8:00), ..., and (features of content C to be pushed, 20:00-24:00).

[0226] After forming a plurality of first combinations, each first feature is input into the task completion prediction probability model together with the object feature of the target object.

[0227] The object feature of the target object refers to the features related to the target object. The object feature can be the features of the target object itself, such as the industry, education level, etc. of the target object. For the same feature, if the feature value is different, the target object may have different preferences for content. For example, target objects in different industries may have different preferences for content. Taking video content as an example, objects in the news industry may pay more attention to videos related to hot events, and objects in the catering industry may like videos related to health and wellness.

[0228] Note that when obtaining the target object's object characteristics, the target object's consent must be obtained in advance. In addition, the collection, use and processing of these object characteristics will comply with relevant laws, regulations and standards. When obtaining the target object's consent, the target object's separate permission or consent can be obtained through pop-ups or by jumping to a confirmation page.

[0229] The object characteristics can also be the object characteristics of the object group (such as family, work group) where the target object is located. This is because, in some cases, the object's preference for content is greatly affected by the object group to which it belongs. For example, the object itself likes to watch music-related video content, but because other members of the object's family group like to watch game-related video content, the object may also be interested in game-related video content. Considering the object characteristics of the object group where the target object is located can improve the accuracy of content push. Similar to the above, when obtaining the object characteristics of the object group (such as family, work group) where the target object is located, the consent of the object group members must be obtained in advance, so it will not be repeated.

[0230] The task completion prediction probability model is used to predict the target object's first task completion probability for the content to be pushed in the first combination of the time period to be pushed. A task refers to an item related to the content to be pushed that is used to measure the target object's interest in the content to be pushed. For example, a task is whether the target object likes the content to be pushed, or whether the target object collects the content to be pushed.

[0231] In one embodiment, the task completion probability prediction model is used to predict the first task completion probability of the target object for multiple tasks in the waiting period. The task completion probability prediction model includes a vectorization layer, a cascade layer, multiple expert sub-models, multiple gating nodes corresponding to multiple tasks, and multiple prediction sub-models.

[0232] like Fig.10 As shown, each first combination is input into the task completion probability prediction model together with the object feature of the target object to obtain the predicted first task completion probability, including:

[0233] Step 1010: vectorize each first combination and the object feature of the target object, and concatenate the vectorization results to obtain a first feature vector;

[0234] Step 1020, inputting the first feature vector into multiple expert sub-models;

[0235] Step 1030: Perform a weighted sum of the outputs of multiple expert sub-models through the gating nodes corresponding to each task to obtain a second feature vector corresponding to the task;

[0236] Step 1040: Input the second feature vector corresponding to the task into the prediction model corresponding to the task to obtain the first task completion probability corresponding to the task.

[0237] The process of steps 1010-1040 is as follows Fig.11A As shown in the figure on the right, there are multiple first combinations of features of the content to be pushed and the time period to be pushed. One of the first combinations is input into the vectorization layer of the task completion probability prediction model together with the object features.

[0238] In another embodiment, each first combination is input into a task completion probability prediction model together with the object characteristics of the target object to obtain a predicted first task completion probability, including: each first combination is input into a task completion probability prediction model together with the object characteristics of the target object and the scene characteristics of the target object to obtain a predicted first task completion probability.

[0239] Scenario characteristics refer to the characteristics of the current scenario in which the target object is viewing the pushed content. The same target object may also view different pushed content in different scenarios. For example, in a work scenario, the target object may want to view push content related to work more; when at home, he may want to view push content related to his own preferences more. Therefore, scenario characteristics will also have an impact on the probability of completing the first task. The probability of completing the first task is the probability that the target object will complete the task in the scenario during the waiting period for the pushed content. Scenario characteristics include: characteristics of the geographical area where the target object is currently located, characteristics of the current input day type (weekday, weekend, or holiday), etc.

[0240] When obtaining the scene features of the target object, the target object's consent must be obtained in advance, which is similar to the above and will not be repeated here.

[0241] The characteristics of the current geographical area refer to the characteristics of the geographical area where the target object is currently located. This geographical area can be an administrative geographical area, such as City A, City B, etc., or it can be a geographical area divided by entities on the map, such as the XX shopping mall area, the XX hospital area, etc. For administrative geographical areas, the push content that the target object is interested in has an impact. For example, if the target object is in City A, the push content can be tourist attractions or food culture related to City A. For geographical areas divided by entities, the push content that the target object is interested in also has an impact. For example, if the target object is in the XX shopping mall area, the push content can be activity information or special projects in the mall; if the target object is in the XX hospital area, the push content can be content related to health and wellness.

[0242] The current input day type feature refers to the feature of the type of day when the target object views the push content. It is divided into weekdays, weekends, and holidays. The type of push content that the target object wants to view may be different on different types of days. For example, on weekdays (Monday to Friday), the target object is often at work, and the push content that the target object wants to view may be work-related; on weekends (Saturday and Sunday), when the target object does not need to work, the push content that the target object wants to view may be related to his or her hobbies; on holidays, the target object may want to view push content related to holiday culture.

[0243] like Fig. 11BAs shown, the object features, scene features, and each first combination are input into the vectorization layer of the task completion probability prediction model for feature vectorization.

[0244] The advantage of inputting the scene features together with the object features and the first combination into the task completion probability prediction model is that the impact of the scene on the content preference of the target object is taken into account, thereby improving the accuracy of predicting the first task completion probability.

[0245] At the vectorization layer, since the object features, the features of the content to be pushed in the first combination, and the features of the time period to be pushed include multiple types of features, it is necessary to vectorize the features into a form that can be learned by machines.

[0246] In one embodiment, vectorizing each first combination with an object feature of a target object includes:

[0247] The characteristics of the content to be pushed, the time period to be pushed, and the characteristics of the object are used as the basic characteristics of the push;

[0248] If the pushed basic feature is a character feature, the character feature is input into the embedding layer to obtain the vectorization result corresponding to the character feature;

[0249] If the pushed basic feature is a numerical feature, the numerical feature is numerically mapped to obtain the vectorized result corresponding to the numerical feature;

[0250] If the pushed basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain a vectorized result corresponding to the variable-length sequence feature.

[0251] Based on the aforementioned embodiment of inputting scene features into the task completion probability prediction model, the push basic features may include scene features in addition to the features of the content to be pushed, the time period to be pushed, and the object features.

[0252] The push basic features may be character features, such as industry features in object features, geographic area features in scene features, and the like.

[0253] If the pushed basic feature is a character feature, the character feature can be input into the embedding layer to obtain the vectorized result corresponding to the character feature.

[0254] Since character features are sequences of characters, each character is considered a discrete symbol that cannot be recognized by machines. The role of the embedding layer is to convert discrete characters into vectors that can be recognized by machines. For example Fig. 12AAs shown, after the character feature "tourism" is input into the embedding layer, the first feature vector is [1,0,1,0,...,1,1,0]; after the character feature "music" is input into the embedding layer, the first feature vector is [0,0,1,0,...,0,1,0]; after the character feature "popular science" is input into the embedding layer, the first feature vector is [1,1,0,1,...,1,0,0].

[0255] The push basic feature may also be a numerical feature, for example, a content push quantity feature in a feature of content to be pushed.

[0256] If the pushed basic feature is a numerical feature, the numerical feature is numerically mapped to obtain the vectorized result corresponding to the numerical feature.

[0257] The value mapping process can be to convert the value into its corresponding binary form. Fig. 12B The numerical mapping conversion shown in the figure. The vectorized result of the numerical feature "33" after numerical conversion is "00100001"; the vectorized result of the numerical feature "620" after numerical conversion is "1001101100"; the vectorized result of the numerical feature "99" after numerical conversion is "01100011".

[0258] The push basic feature can also be an indefinite length sequence feature. An indefinite length sequence feature refers to a string feature consisting of a large but uncertain number of characters. For example, an object custom signature in an object feature.

[0259] If the pushed basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain a vectorized result corresponding to the variable-length sequence feature.

[0260] Pooling is the process of reducing the dimensionality of a high-dimensional vector. The specific process of pooling is to integrate several eigenvalues ​​in a preset neighborhood of a high-dimensional vector generated based on the features of an indefinite length sequence into a new eigenvalue. For example Fig. 12CAs shown in the figure, the variable-length sequence feature is the object's custom signature. The object's custom signature is "The scenery is beautiful,...". According to the variable-length sequence feature, the corresponding high-dimensional feature vector can be obtained, and the high-dimensional feature vector is [0,1,0,0,0,1,1,1,1,1,0,1,0,0,0,0,0,1,1,1,0,1,1,1]. The value of the preset neighborhood is set to 3, that is, every 3 adjacent feature quantities in the high-dimensional feature vector are taken as a neighborhood, and the feature quantities in a neighborhood are integrated into one feature quantity. The integration method can be to take the average value of the feature quantity in the field as the feature quantity of the neighborhood. For example, in the above high-dimensional feature vector, the first neighborhood is taken as [0,1,0], and the average value of the feature quantity in this neighborhood is 0.33, so the feature value of this neighborhood is 0.33. The second neighborhood is [0,0,1], and the average value of the feature quantity in this neighborhood is 0.33, so the feature value of this neighborhood is 0.33. Similarly, the feature quantities of the subsequent neighborhoods are 1, 0.67, 0, 0.33, 0.67, and 1. Therefore, the vectorized result of the indefinite length sequence feature after pooling is [0.33, 0.33, 1, 0.67, 0, 0.33, 0.67, 1].

[0261] Another way to integrate several features in a preset domain into a new feature is to use the maximum value of the feature quantity in the neighborhood as the feature value of a neighborhood. For example, in the above high-dimensional feature vector, the value of the preset neighborhood is also set to 3, and the maximum value in the neighborhood is taken as the feature value of the neighborhood. Then, the first feature vector obtained after the indefinite length sequence feature is pooled is [1,1,1,1,0,1,1,1].

[0262] In this embodiment, the advantage of using different vectorization methods for different types of push basic features is that the features can be vectorized according to the characteristics of different types of push basic features, thereby ensuring the accuracy of the feature vectorization results.

[0263] Each feature gets a corresponding vectorized result after passing through the vector layer. The vectorized results corresponding to all features are input into the cascade layer, and all vectorized results are cascaded into a vector.

[0264] Cascading all vectorized results means concatenating the vectorized results in the order of output to form a vector. For example, the vectorized results include: [1,0,1,1,0,1,1,0], [0,0,1,0,1,0,1,0], [0,0,0,1,1,0,1,1], [0,1,1,1,1,0,0,1,0,1]. After concatenating the vectorized results, the first eigenvector obtained is: [1,0,1,1,0,1,1,0,0,0,1,0,1,0,1,0,1,0,1,1,0,1,1,1,1,0,0,1,0,1].

[0265] like Fig.11A (or Fig. 11B As shown in FIG. 1 , after obtaining the first feature vector output by the cascade layer, the first feature vector is input into multiple expert sub-models. Multiple expert sub-models are used to analyze and calculate each dimension of the first feature vector in order to balance the sharing and mutual exclusion between multiple tasks. Each expert sub-model obtains an output based on the input first feature vector. The outputs of multiple expert sub-models are input into the gating nodes corresponding to the tasks.

[0266] Each task corresponds to a gating node, which is used to integrate the output of the expert sub-model into a high-order vector for the task. In each gating node, the outputs of multiple expert sub-models are weighted and summed to obtain the second eigenvector corresponding to the task.

[0267] In one embodiment, Fig.13 As shown, through the gating nodes corresponding to each task, the outputs of multiple expert sub-models are weighted and summed to obtain the second feature vector corresponding to the task, including:

[0268] Step 1310: Obtain a weight vector of a gated node, where the weight vector indicates weights of multiple expert sub-models;

[0269] Step 1320: Through the gating node, the outputs of multiple expert sub-models are weighted and summed using the weight vector to obtain a second feature vector corresponding to the task.

[0270] The weight vector refers to the weights of the outputs of multiple expert sub-models for outputting the second feature vector corresponding to a task in a gating node corresponding to a task. The weights of the outputs of the same expert sub-model may be different in different gating nodes. This is because different gating nodes correspond to different tasks, and the contributions of the same expert sub-model may be different for different tasks.

[0271] In one embodiment, Fig.14 As shown, the weight vector of the gated node is obtained, including:

[0272] Step 1410: Obtain a first matrix;

[0273] Step 1420: multiply the first matrix by the transpose of the first eigenvector, and perform normalized exponential processing on the multiplication result to obtain a weight vector of the gated node.

[0274] The first matrix is ​​a trainable parameter matrix. The number of rows in the first matrix is ​​the number of expert sub-models, and the number of columns is the dimension of the first eigenvector. The dimension of the first eigenvector is the sum of the dimensions of the vectorized results of all features. For example, if the number of expert sub-models is 4 and the dimension of the first eigenvector is 10, then the number of rows in the first matrix is ​​4 and the number of columns is 10.

[0275] Multiplying the first matrix by the transpose of the first eigenvector is equivalent to multiplying a matrix by a column vector, and the result is a column vector with a dimension equal to the number of expert sub-models.

[0276] For example, the first matrix is It can be seen that the number of expert sub-models is 3, the dimension of the first eigenvector is 4; the first eigenvector is [0,1,1,1], so the result of multiplication is:

[0277] According to the order of the expert sub-models in the task completion probability prediction model, each row in the multiplication result represents the weight of an expert sub-model in the gate node in turn.

[0278] The result of the multiplication is normalized exponentially to obtain a weight vector. The purpose of normalized exponential processing is to limit each value in the result of the multiplication operation to a specific range, so that the values ​​of the first weights of different models are in the same range and comparable. The normalized exponential processing uses the softmax function. The softmax function inputs the result of the multiplication of the first matrix and the first eigenvector, and outputs the weight vector of the gated node. Therefore, the process of calculating the first weight of the first shared model is shown in Formula 1:

[0279] g t (X) = softmax(W t ·X T ) (Formula 1).

[0280] In Formula 1, W t refers to the first matrix in the gating node corresponding to the task, X refers to the first eigenvector, g t (X) refers to the weight vector of the gating node corresponding to the task. Each row of the weight vector represents the weight of an expert sub-model.

[0281] The advantage of obtaining the weight vector through normalized index processing in this embodiment is that the first weights of each model are in the same numerical range, which can more effectively reflect the influence of different expert sub-models on obtaining the second eigenvector, thereby improving the accuracy of obtaining the second eigenvector.

[0282] In step 1320, in the gate node, the outputs of multiple expert sub-models are weighted and summed using the weight vector to obtain the second feature vector corresponding to the task. The process of obtaining the second feature vector is shown in Formula 2:

[0283]

[0284] In formula 2, g t (X) refers to the weight vector of the gated node, i represents the weight corresponding to the i-th expert sub-model in the weight vector, and e i (X) refers to the output of the ith expert sub-model. The output of the expert sub-model is multiplied by the weight, and then K (the number of expert sub-models) multiplication results are summed to obtain the second eigenvector.

[0285] The advantage of the embodiment of steps 1310-1320 is that the outputs of the expert sub-models are weighted and summed based on the weight vector, which reflects the different impacts of different expert sub-models on the task and improves the accuracy of determining the second feature vector corresponding to the task.

[0286] like Fig.11A (or Fig. 11B As shown), after the gated node outputs the second feature vector, the second feature vector is input into the prediction sub-model corresponding to the task. The prediction sub-model will output the target object's first task completion probability for the content to be pushed and the corresponding task during the period to be pushed. For example, the task corresponding to the prediction sub-model is: whether the target object will like the content to be pushed during the period of 1:00-2:00. The second feature vector is input into the prediction sub-model, and the probability of completing the first task output by the prediction sub-model is 0.65, that is, the target object has a 65% probability of liking the content to be pushed during the period of 1:00-2:00.

[0287] The process of predicting the sub-model to obtain the first task completion probability of the corresponding task is shown in Formula 3:

[0288]

[0289] In formula 3, represents the second feature vector of the gating node output corresponding to the task; p t represents the probability of completing the first task obtained by the prediction sub-model; θ t (*) indicates the prediction sub-model corresponding to the task, which can be a deep neural network model.

[0290] The advantage of the embodiment of steps 1010-1040 is that by analyzing the preferences of the target object for the content to be pushed during the push period through multiple tasks, the prediction result is more comprehensive, and the accuracy of determining the first target combination based on the completion probability of the first task is improved.

[0291] Detailed description of step 340

[0292] In step 340 , a first target combination is determined based on the first task completion probability of each first combination.

[0293] The first target combination is a combination of a time period to be pushed and content to be pushed to a target object.

[0294] The number of first target combinations to be determined can be determined according to the preset number of pushes to the target object within the preset push cycle. For example, if the preset push cycle is 1 day and the preset number of pushes is 1, then only one first target combination needs to be determined from all first combinations; similarly, if the preset number of pushes is 3, then 3 first target combinations need to be determined from all first combinations.

[0295] Based on the first task completion probability of each first combination, determine the first target combination, including: sorting the first task completion probability of all first combinations from high to low; obtaining the first N first combinations in the sorting as the first target combination, where N refers to the preset number of pushes.

[0296] In the aforementioned embodiment, the task completion probability prediction model is used to predict the first task completion probability of multiple tasks. Therefore, in one embodiment, based on the first task completion probability of each first combination, determining the first target combination includes: determining the first overall task completion probability of the first combination based on the first task completion probability of the multiple tasks in the first combination; determining the first target combination based on the first overall task completion probability of each first combination.

[0297] In one embodiment, the first overall task completion probability of the first combination can be calculated by calculating the average of the first task completion probabilities of multiple tasks. For example, the task completion probability prediction model predicts the first task completion probabilities of the first combination for three tasks, which are: 0.34, 0.56, 0.74, and 0.59. Therefore, the first overall task completion probability of the first combination is (0.34+0.56+0.74+0.59) / 4=0.56. The advantage of using the average to determine the first overall task completion probability is that the first task completion probability of each task has the same impact on determining the first target combination, which improves the fairness of determining the first target combination in multiple tasks.

[0298] Determining a first target combination based on the first overall task completion probability of each first combination includes: sorting the first overall task completion probabilities of all first combinations; and obtaining the first N first combinations in the sorting as the first target combination.

[0299] In this embodiment, the advantage of determining the first target combination based on the first task completion probability of multiple tasks is that the influence of each task on determining the first target combination is comprehensively considered, thereby improving the comprehensiveness of determining the first target combination.

[0300] In another embodiment, if Fig.15 As shown, based on the first task completion probability of each first combination, determining a first target combination among multiple first combinations includes:

[0301] Step 1510: Obtain the focus task of the target object from among the multiple tasks;

[0302] Step 1520: Obtain a first task completion probability of a focused task of the first combination from the first task completion probabilities of the plurality of tasks of the first combination;

[0303] Step 1530: Determine a first target combination from among the multiple first combinations according to the first task completion probabilities of the focus tasks of the multiple first combinations.

[0304] The target object's attention task refers to the task that can represent the target object's interest in the pushed content among multiple tasks. For example, multiple tasks include: whether to like, whether to comment, and whether to forward. If the target object likes to comment on the content he is interested in when viewing the pushed content, then the target object's attention task is "whether to comment".

[0305] In one embodiment, obtaining the focus tasks of the target object includes: obtaining a task completion table of the target object within a predetermined time period before the current time; and determining the focus tasks of the target object based on the task completion table.

[0306] When obtaining the target object's attention task, the target object's consent must be obtained in advance. This is similar to the previous content, so I will not repeat it.

[0307] The task completion table is used to count the target object's completion of each task within a predetermined time period before the current time. An example of a task completion table is shown in Table 6.

[0308] Task Number of tasks completed (within 1 day) Like 10 Comment 5 Collect 2

[0309] Table 6

[0310] Table 6 shows the statistics of the number of times the target object completed each task within 1 day before the current time point. The number of "Like" is 10 times, the number of "Comment" is 5 times, and the number of "Collect" is 2 times.

[0311] Determining the focus task of the target object based on the task completion table may be to use the task with the highest number of task completions as the focus task of the target object.

[0312] Take the task completion table in Table 6 as an example, where the task "Like" has been completed 10 times, the highest number of times among all tasks. In other words, the target object tends to like the content that he is interested in. Therefore, "Like" is the task that the target object pays attention to.

[0313] In this embodiment, the focus task is determined by counting the number of times the target object completes each task, which ensures that the focus task is the action most frequently performed by the target object on the pushed content, thereby improving the accuracy of obtaining the focus task.

[0314] After obtaining the target object's concerned task, the first task completion probability of the concerned task is obtained from the first task completion probabilities of multiple tasks. For example, the target object's concerned task is task B, the first task completion probability of task B corresponding to the first combination A is 0.56, the first task completion probability of task B corresponding to the first combination B is 0.37, and the first task completion probability of task B corresponding to the first combination C is 0.73.

[0315] The first combinations are sorted in descending order according to the probability of completing the first task of the concerned task, and the first N first combinations are obtained in the sorting as the first target combination. This is because the greater the probability of completing the first task of the concerned task, the more likely the target object is to complete the target task, that is, the target object is more likely to be interested in the content to be pushed during the period to be pushed in the first combination. For example, the preset number of pushes N is 1. Based on the above example, the first combination with the highest probability of completing the first task of task B is taken as the first target combination. Therefore, the first target combination is the first combination C.

[0316] The advantage of the embodiment of steps 1510-1530 is that the first target combination is determined by the first task completion probability of the target object's focus task, which improves the matching degree between the first target combination and the target object and improves the accuracy of obtaining the first target combination.

[0317] In another embodiment, if Fig.16 As shown, according to the first task completion probability of the focus tasks of the multiple first combinations, determining the first target combination from the multiple first combinations includes:

[0318] Step 1610: if there are multiple tasks of interest, determine a first overall task completion probability of the first combination based on the first task completion probabilities of the multiple tasks of interest in each first combination;

[0319] Step 1620: Determine a first target combination among the multiple first combinations based on the first overall task completion probabilities of the multiple first combinations.

[0320] The target object may perform multiple tasks for the content it is interested in. For example, if the target object is interested in a push content, he or she may like it and comment on it.

[0321] If there are multiple tasks of interest, then the first task completion probability of each task of interest corresponding to the first combination is obtained, and then the first overall task completion probability of the first combination is determined according to the first task completion probability of each task of interest.

[0322] For example, the first task completion probabilities of multiple focus tasks of multiple first combinations are shown in Table 7:

[0323] First Group Focus on Task A Focus on Task B The first combination A 0.45 0.56 First Group B 0.39 0.78 The first group C 0.69 0.44

[0324] Table 7

[0325] As shown in Table 7, the target object's attention tasks include attention content A and attention content B.

[0326] Determining the first overall task completion probability may be by calculating the average of the first task completion probabilities of the plurality of focus tasks. For example, as shown in Table 7, the first overall task completion probability of the first combination A is (0.45+0.56) / 2=0.505. The first overall task completion probability of the first combination B is (0.39+0.78) / 2=0.585. The first overall task completion probability of the first combination C is (0.69+0.44) / 2=0.565.

[0327] The advantage of calculating the first overall task completion probability by means of an average is that the first task completion probabilities of different focus tasks have the same influence on determining the first overall task completion probability, thereby improving the fairness of determining the first overall task completion probability.

[0328] The method for determining the first overall task completion probability can also be by calculating the weighted average of the first task completion probabilities of multiple focus tasks. In this embodiment, weights must first be set for different focus tasks. For example, in the above example, the weight set for focus task A is 0.6, and the weight set for focus task B is 0.4. Then, the first overall task completion probability of the first combination A is 0.45*0.6+0.56*0.4=0.494; the first overall task completion probability of the first combination B is 0.39*0.6+0.78*0.4=0.546; the first overall task completion probability of the first combination C is 0.69*0.6+0.44*0.4=0.59.

[0329] The advantage of calculating the first overall task completion probability by weighted average is that the weights of the first task completion probabilities of different focus tasks can be adjusted according to actual application conditions, thereby improving the flexibility of calculating the first overall task completion probability.

[0330] After determining the first overall task completion probabilities of multiple first combinations, sort the multiple first combinations in descending order of the first overall task completion probabilities, and then obtain the first N first combinations in the sorting as the first target combination. For example, the first overall task completion probability of first combination A is 0.49, the first overall task completion probability of first combination B is 0.55, and the first overall task completion probability of first combination C is 0.59. If the preset number of pushes N is 1, then the first combination with the highest first overall task completion probability is used as the first target combination. Therefore, the first target combination is first combination C.

[0331] The advantage of the embodiment of steps 1610-1620 is that the first target combination is determined by acquiring multiple focus tasks of the target object, so that the acquired first target combination does not only depend on the task completion probability of the first task of one task, thereby improving the accuracy of determining the first target combination.

[0332] Detailed description of step 350

[0333] The time period to be pushed in the first target combination is used as the target time period, and a reminder message corresponding to the content to be pushed in the first target combination is pushed to the target object during the target time period.

[0334] For example, the first target combination is (content A to be pushed, 10:00-12:00), then the reminder message corresponding to the content A to be pushed is pushed to the target object in the time period of 10:00-12:00.

[0335] Pushing a reminder message corresponding to the content to be pushed to the target object can be done by sending the reminder message corresponding to the content to be pushed to the target object in the application that displays the content to be pushed, so that the target object can see the reminder message when opening the application; or sending the reminder message corresponding to the content to be pushed to the terminal device used by the target object, so that the target object can see the reminder message even if the application is not opened.

[0336] Note that sending a reminder message to a target object requires the target object's consent in advance. This is similar to the previous content, so I will not repeat it here.

[0337] In one embodiment, Fig.17 As shown, a reminder message corresponding to the content to be pushed is pushed to the target object during the target period, including:

[0338] Step 1710: Obtain a set of sub-periods to be pushed of a target period, where the set of sub-periods to be pushed includes a plurality of sub-periods to be pushed;

[0339] Step 1720: Obtain a first combination with a target period from among the multiple first combinations, and select a first combination after selection based on the first task completion probability from among the first combinations with the target period;

[0340] Step 1730, perform a second combination on each first combination after screening and each sub-period to be pushed in the sub-period set to be pushed, input each second combination together with the object feature of the target object into the task completion probability prediction model to obtain a predicted second task completion probability;

[0341] Step 1740: determining a second target combination from a plurality of second combinations based on the second task completion probability of each second combination;

[0342] Step 1750: The time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content. In the target sub-time period of the target time period, a reminder message corresponding to the target content is pushed to the target object.

[0343] In step 1710, the sub-period set to be pushed is a set of a target period divided into a predetermined number of sub-periods to be pushed. For example, if the target period is 20:00-22:00, the target period is divided into 8 sub-periods to be pushed, and the time interval of each sub-period to be pushed is 15 minutes, namely: 20:00-20:14, 20:15-20:29, 20:30-20:44, 20:45-20:59, 21:00-21:14, 21:15-21:29, 21:30-21:44, 21:44-22:00. The method of determining the predetermined number of sub-periods to be pushed in the sub-period set to be pushed is similar to the method of determining the number of sub-periods to be pushed in the sub-period set to be pushed in the aforementioned embodiment, and will not be repeated here.

[0344] In step 1720, a first combination whose time period to be pushed is a target time period is obtained from among the multiple first combinations. Fig.18 As shown, the target time period is 20:00-24:00, and the first combinations with the target time period obtained from multiple first combinations are: (features of content A to be pushed, 20:00-24:00), (features of content B to be pushed, 20:00-24:00), and (features of content C to be pushed, 20:00-24:00).

[0345] Obtain the first task completion probability of each first combination with a target time period. It should be noted that in the aforementioned embodiment of determining the target combination based on the first task completion probability of multiple first combinations, there are multiple embodiments of determining the target combination based on the first task completion probability of a single task and based on the first overall task completion probability of multiple tasks. Therefore, obtaining the first completion probability of each first combination with a target time period can also be regarded as obtaining the first overall task completion probability of each first combination with a target time period.

[0346] A screened first combination is screened out from a plurality of first combinations with target time periods.

[0347] In one embodiment, the first combination after screening is screened out based on the probability of completion of the first task, including: ranking multiple first combinations with target time periods in descending order based on the probability of completion of the first task, and obtaining the top M first combinations in the ranking as the first combination after screening, wherein M refers to the preset screening number.

[0348] The preset screening number is set in advance, and the preset screening number needs to be less than the number of the first combination. If the preset screening number is greater than the number of the first combination with the target time period, all the first combinations with the target time period are used as the first combination after screening.

[0349] like Fig.18As shown, the first task completion probability of the first combination A (feature of content A to be pushed, 20:00-24:00) is 0.65, the first task completion probability of the first combination B (feature of content B to be pushed, 20:00-24:00) is 0.47, and the first task completion probability of the first combination C (feature of content C to be pushed, 20:00-24:00) is 0.23. The preset screening number is 2, so the first combinations with the top two first task completion probabilities are selected from the above three first combinations, namely, the first combination A and the first combination B.

[0350] The advantage of filtering the first combination after filtering based on the preset filtering number is that the first combination after filtering can be quickly filtered out through the ranking of the completion probability of the first task of the first combination, thereby improving the efficiency of determining the first combination after filtering.

[0351] In another embodiment, Fig.19 As shown, based on the completion probability of the first task, the first combination after screening is screened out, including:

[0352] Step 1910: Determine, in the first combinations with the target time periods, a first number of first combinations whose first task completion probability is greater than a first threshold;

[0353] Step 1920: If the first number is not less than the second threshold, the first number of first combinations are taken as the first combination after screening; otherwise, the first number of first combinations before the second threshold are taken in descending order of the probability of completion of the first task as the first combination after screening.

[0354] In this embodiment, the first threshold is used to screen the first combinations that meet the conditions. The first threshold is used to screen the first number of first combinations with target time periods. The purpose of setting the first threshold is that if there are multiple first combinations with a high probability of completing the first task, all of these first combinations can enter the next round of confirmation to obtain a more accurate push time period and push content.

[0355] The second threshold is used to ensure that the number of first combinations after screening must meet the minimum preset screening number requirement. If the first number is not less than the second threshold, that is, the number of first combinations whose first task completion probability is greater than the first threshold has exceeded the minimum preset screening number, then the first number of first combinations is directly used as the first combination after screening. For example, the first task completion probability of the first combination A is 0.56, the first task completion probability of the first combination B is 0.39, the first task completion probability of the first combination C is 0.76, the first task completion probability of the first combination D is 0.63, and the first task completion probability of the first combination E is 0.47. If the first threshold is 0.5, then the first number of first combinations are: first combination A, first combination C, and first combination D. The first number is 3. If the second threshold is 2, then the first number is greater than the second threshold, and the first combination A, the first combination C, and the first combination D are used as the first combination after screening.

[0356] If the first number is less than the second threshold, it means that the first number of first combinations does not meet the minimum preset screening number requirement, and further screening of the first combinations is required. Sort the first combinations with the target time period in descending order according to the probability of completing the first task, and use the first combinations before the second threshold in the sorting as the first combinations after screening. Based on the above example, if the second threshold is 4, then the first number is less than the second threshold, and the first combinations ranked in the top 4 in the probability of completing the first task in the first combinations AE are selected as the first combinations after screening, namely: first combination C, first combination D, first combination A, and first combination E.

[0357] This embodiment uses the first threshold and the second threshold to jointly screen out the first combination after screening, which not only ensures that the first combination with a higher first completion probability will not be missed, but also ensures that the number of the first combinations after screening meets the minimum preset screening number requirement, thereby improving the accuracy of determining the first combination after screening.

[0358] There may be a time interval between the time when the set of content to be pushed is generated and the time when the first combination after screening is selected. For example, if the scheduled push cycle is 1 day, then the set of content to be pushed is generated at 0:00 every day and the first target combination is determined to determine the target time period and the content to be pushed on this day. However, if you want to further confirm the target sub-time period in the target time period, you can perform the process of further confirming the target sub-time period when the natural time reaches the target time period. For example, the target time period in the first target combination is 12:00-14:00. When the natural time reaches 12:00, it is confirmed which sub-time period of the target time period is the target sub-time period. In the time interval of 0:00-12:00, it is very likely that content will be added, and it is very likely that the target object is particularly interested in these added contents. However, since the added content does not belong to the first combination, the added content cannot be further confirmed whether it is the target content, which easily causes the problem of content omission.

[0359] Therefore, in one embodiment, if Fig. 20 As shown, after step 1720, the push processing method further includes:

[0360] Step 2010: Acquire additional content generated on the content push platform between the generation time of the content set to be pushed and the current time;

[0361] Step 220: Add the combination of the feature of the added content and the target time period to the first combination after screening.

[0362] The method of acquiring the additional content is similar to the method of acquiring the content set to be pushed in the aforementioned embodiment, and will not be described in detail here.

[0363] The current time can be a preset time point, such as 8:00 or 10:00, or a time point entering a target time period, for example, if the target time period is 20:00-22:00, then the current time is 20:00. When the natural time reaches this time point, the additional content is obtained.

[0364] Please note that when obtaining additional content, you also need to obtain the consent of the content publisher.

[0365] Since there is no combination of the time period to be pushed and the added content, the target time period is combined with the added content and added to the first combination after screening. Fig.21As shown, between the generation time of the content set to be pushed and the current time when the first combination after filtering is generated, additional content A, additional content B, and additional content C are added to the content push platform. The added content is obtained at the current time, and each added content is combined with the target time period and added to the first combination after filtering. The original first combination after filtering includes: (features of content A to be pushed, 20:00-24:00), (features of content B to be pushed, 20:00-24:00), (features of added content A, 20:00-24:00), (features of added content B, 20:00-24:00), and (features of added content C, 20:00-24:00).

[0366] The advantage of this embodiment is that the added content is also considered as the content to be pushed, thereby avoiding the problem of the added content being omitted because it does not belong to any first combination, and improving the comprehensiveness and accuracy of the first combination after determination and screening.

[0367] In step 1730, the filtered first combination is combined with each of the sub-periods to be pushed in the sub-period set to be pushed. Fig. 22 As shown, the first combination after screening includes (features of content A to be pushed, 20:00-24:00), (features of content B to be pushed, 20:00-24:00), (features of content A added, 20:00-24:00), (features of content B added, 20:00-24:00), and (features of content C added, 20:00-24:00). The sub-period set to be pushed is a set of sub-periods to be pushed that are divided into 30 minutes in the target period 20:00-24:00, including: 20:00-20:30, 20:30-21:00, ..., 23:30-24:00. The first combination after screening is sequentially combined with the sub-periods to be pushed in the sub-period set to be pushed. After screening, the first combination (features of content A to be pushed, 20:00-24:00) is combined with the sub-periods to be pushed in turn to obtain the second combination: (features of content A to be pushed, 20:00-24:00, 20:00-20:30), (features of content A to be pushed, 20:00-24:00, 20:30-21:00), ..., (features of content A to be pushed, 20:00-24:00, 23:30-24:00). Similarly, the remaining first combinations after screening are combined with the sub-periods to be pushed in turn to obtain the following: Fig. 22 Multiple second combinations are shown.

[0368] Each second combination is input into the task completion probability prediction model together with the object feature of the target object. The task completion probability prediction model obtains the second task completion probability. The second task completion probability is the probability that the target object completes the task for the content to be pushed in the sub-period to be pushed of the period to be pushed.

[0369] Similar to the aforementioned embodiment, in one embodiment, each second combination is input into a task completion probability prediction model together with the object characteristics of the target object to obtain a predicted second task completion probability, including: inputting each second combination into a task completion probability prediction model together with the object characteristics of the target object and the scene characteristics of the target object to obtain a predicted second task completion probability; wherein the second task completion probability is the probability of the target object completing the task for the content to be pushed, in the sub-period to be pushed of the period to be pushed, and in the scenario.

[0370] The process of obtaining the second task completion probability based on the second combination, object features, and scene features is the same as the process of obtaining the first task completion probability based on the first combination, object features, and scene features in the aforementioned embodiment, and will not be repeated here.

[0371] The task completion probability prediction model predicts the second task completion probability in the same manner as the first task completion probability in the aforementioned embodiment, and will not be described in detail here.

[0372] In step 1740, the process of determining the second target combination among multiple second combinations based on the second task completion probability of each second combination is similar to the process of determining the first target combination among multiple first combinations based on the first task completion probability of each first combination in the aforementioned embodiment, and will not be repeated here.

[0373] In step 1750, the time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content. When the natural time reaches the target sub-time period in the target time period, a reminder message corresponding to the target content is pushed to the target object. For example, Fig.23 As shown, the second target combination is (features of content B to be pushed, 20:00-24:00, 20:30-21:00), with "content B to be pushed" as the target content, "20:00-24:00" as the target time period, and "20:30-21:00" as the target sub-time period. When the natural time reaches 20:30-21:00 between 20:00-24:00, the reminder message corresponding to the content B to be pushed is pushed to the target object.

[0374] Please note that pushing reminder messages to the target object requires the target object's consent in advance, which will not be repeated here.

[0375] The advantage of the embodiment of steps 1710-1750 is that the time period to be pushed is further refined to obtain a sub-time period to be pushed with a smaller time range, thereby improving the accuracy of pushing to the target object.

[0376] Training process of task completion probability model in the disclosed embodiment

[0377] In one embodiment, Fig.24 As shown in Figure 1, the task completion probability prediction model is trained in the following way:

[0378] Step 2410: Obtain a sample push event set;

[0379] Step 2420: extract a first sample and a second sample from the sample push event;

[0380] Step 2430: Input the first sample and the second sample extracted from each sample push event into the task completion probability prediction model for training.

[0381] The sample push event set includes multiple sample push events. The sample push event includes sample object features, sample content to be pushed features, sample time period to be pushed, and sample sub-time period to be pushed. Optionally, the sample push event may also include sample scene features. The sample push event also has a training label for the task, and the training label for the task indicates whether the sample object has completed the task for the sample content to be pushed in the sample sub-time period of the sample to be pushed period. For example, the training label in a sample push event is {Task A: 1, Task B; 0}, where a training label of 1 represents that the task is completed, and a training label of 0 represents that the task is not completed. That is to say, in this example, the sample object completed Task A for the sample content to be pushed in the sample push sub-time period of the sample push period, but did not complete Task B.

[0382] The first sample extracted from the sample push event includes sample object features, sample content features to be pushed, and sample time period to be pushed; the second sample includes sample object features, sample content features to be pushed, sample time period to be pushed, and sample sub-time period to be pushed. The reason for extracting the first sample and the second sample is that in the process of determining the target content, target time period, and target sub-time period, the task completion probability prediction model is used to make two predictions, namely, the first task completion probability is predicted based on the object features and the first combination (content to be pushed and time period to be pushed), and the second task completion probability is predicted based on the object features and the second combination (pushing content, time period to be pushed, and sub-time period to be pushed). Therefore, it is necessary to use the first sample and the second sample of the same sample push event to train the task completion probability prediction model respectively.

[0383] The process of steps 2410-2430 is as follows Fig.25As shown, the sample object features, sample content features to be pushed, and sample time period to be pushed of the sample push event A in the sample push event set are extracted as the first sample, and the sample object features, sample content features to be pushed, sample time period to be pushed, and sample sub-time period to be pushed are extracted as the second sample. Both the first sample and the second sample contain the training label of the sample push event, and the first sample and the second sample are respectively input into the task completion probability prediction model for training.

[0384] In one embodiment, Fig.26 As shown, the first sample and the second sample extracted from each sample push event are input into the task completion probability prediction model for training, including:

[0385] Step 2610: for each sample in the first sample and the second sample, for a single task, based on the training label of the single task and the task completion probability predicted by the task completion probability prediction model, determine the error subfunction of the single task;

[0386] Step 2620: Determine the error function of the sample based on the error sub-functions of the multiple tasks;

[0387] Step 2630: Based on the error function of each sample, train a task completion probability prediction model.

[0388] In step 2610, the error sub-function is calculated based on the task completion probability of a single task obtained for the sample based on the task completion probability prediction model. The error sub-function refers to the distance between the task completion probability and the training label corresponding to the task for a single task. The closer the distance, the smaller the error and the better the prediction effect. First, determine the task completion probability of a single task; second, obtain the training label corresponding to the task, if the training label indicates that the task is completed, then the training label is 1, otherwise, the training label is 0; finally, based on the task completion probability of a single task and the training label of the task, determine the error sub-function of the task. The calculation process for determining the error sub-function is shown in Formula 4:

[0389] L t =-q t ·log(p t )-(1-q t )·log(1-p t )(Formula 4).

[0390] In formula 4, q t represents the training label (0 or 1), p t represents the task completion probability of the task. For example, if the training label of task A is 1 and the task completion probability is 0.5, then the error subfunction L of the task is t =-1·log(0.5)=0.3.

[0391] In step 2620, after determining the error subfunction of a single task, the error function of the entire sample is determined based on the error subfunctions of all tasks. The error function refers to the gap between the task completion probability of all tasks and the corresponding training labels. The smaller the gap, the better the prediction result of the model.

[0392] In one embodiment, determining the error function of the sample based on the error sub-functions of the multiple tasks includes: summing the error sub-functions of the multiple tasks to obtain the error function.

[0393] In this embodiment, the calculation process of determining the error function is shown in Formula 5:

[0394]

[0395] In Formula 5, T represents the number of tasks, L t Represents the error subfunction of a single task. For example, the sample includes three tasks, among which the error subfunction of task A is 0.3; the error subfunction of task B is 0.45; and the error subfunction of task C is 0.27. The error function of the sample is 0.3+0.45+0.27=1.02.

[0396] The advantage of determining the error function by summing up the error sub-functions is that the error sub-function corresponding to each task contributes the same amount to determining the error function of the sample, thereby improving the fairness of multiple tasks when training the model.

[0397] In another embodiment, determining the error function of a sample based on the error sub-functions of a plurality of tasks comprises:

[0398] Get the task weights of multiple tasks;

[0399] Based on the task weights, the error sub-functions of multiple tasks are weighted and summed to obtain the error function.

[0400] In this embodiment, the calculation process of determining the error function is shown in Formula 6:

[0401]

[0402] In Formula 6, T represents the number of tasks, W t represents the task weight of a single task, L tRepresents the error subfunction of a single task. The task weights of multiple tasks can be determined according to the actual application scenario. For example, the sample includes three tasks, the task weight of task A is 0.5, the task weight of task B is 0.3, and the task weight of task C is 0.2. The error subfunction of task A is 0.3; the error subfunction of task B is 0.45; and the error subfunction of task C is 0.27. Then, the error function of the sample is 0.5*0.3+0.3*0.45+0.2*0.27=0.339.

[0403] The advantage of weighting the error sub-functions and determining the error function is that different weights can be set for different tasks according to the actual application scenarios, thereby improving the flexibility of the training model.

[0404] In step 2630, after determining the error function of each sample, the task completion probability prediction model is trained based on the error function. The purpose of training is to reduce the error function, and a reduced error function represents a more accurate prediction result. Therefore, the training process is to reduce the error function obtained based on the updated model by updating the parameters in the task completion probability prediction model.

[0405] The task completion probability prediction model can be trained based on the error function using a gradient descent method so that the parameters of the task completion probability prediction model can be updated in the direction of the gradient descent of the error function.

[0406] The advantage of the embodiment of steps 2610-2630 is that the error function of the entire sample is determined by the error sub-function of a single task, so that the prediction result of each task has an impact on the training model, thereby improving the comprehensiveness and accuracy of the training model.

[0407] The advantage of the implementation example of steps 2410-2430 is that the first sample and the second sample are divided according to the sample push event, so that the task completion probability prediction model can be comprehensively trained according to the two forms of samples, which is conducive to improving the model training effect, and further improving the accuracy of the first task completion probability and the second task completion probability obtained based on the task completion probability prediction model.

[0408] Detailed diagram of the implementation of the push processing method of the disclosed embodiment

[0409] Refer to the following Fig. 27 , the implementation details of the push processing method of the embodiment of the present disclosure are explained in detail and by way of example.

[0410] In step 2710, matching content is screened out from the original content set based on the first tag of the target object and the second tag of the original content in the original content set; and content to be pushed is screened out from the matching content based on multiple push indicators to form a content set to be pushed.

[0411] In step 2711, a set of time periods to be pushed is obtained, where the set of time periods to be pushed includes a plurality of time periods to be pushed into which a predetermined push cycle is divided.

[0412] In step 2712, a first combination is performed between the feature of each content to be pushed in the content to be pushed set and each time period to be pushed in the time period to be pushed set.

[0413] In step 2713, each first combination is input into a task completion probability prediction model together with the object features of the target object and the scene features in which the target object is located, so as to obtain a predicted first task completion probability.

[0414] In step 2714, a first target combination is determined among the plurality of first combinations based on the first task completion probability of each first combination.

[0415] In step 2715, the time period to be pushed in the first target combination is taken as the target time period, and a sub-time period set to be pushed of the target time period is obtained, where the sub-time period set to be pushed includes a plurality of sub-time periods to be pushed.

[0416] In step 2716, among the multiple first combinations, a first combination with a target time period is obtained, and among the first combinations with a target time period, a filtered first combination is filtered out based on the first task completion probability.

[0417] In step 2717, additional content generated on the content push platform between the generation time of the content set to be pushed and the current time is obtained; and the combination of the characteristics of the additional content and the target time period is added to the first combination after screening.

[0418] In step 2718, each first combination after screening and each sub-period to be pushed in the sub-period to be pushed are combined into a second combination, and each second combination is input into the task completion probability prediction model together with the object characteristics of the target object and the scene characteristics of the target object to obtain the predicted second task completion probability.

[0419] In step 2719 , a second target combination is determined among the plurality of second combinations based on the second task completion probability of each second combination.

[0420] In step 2720, the time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content. In the target sub-time period of the target time period, a reminder message corresponding to the target content is pushed to the target object.

[0421] Description of the apparatus and device of the present disclosure

[0422] It is to be understood that, although the steps in the above-mentioned flowcharts are sequentially displayed according to the characterization of arrows, these steps are not necessarily executed in sequence according to the order of arrow characterization. Unless there is a clear description in the present embodiment, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0423] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to the characteristics of the target content such as target content attribute information or attribute information sets, the permission or consent of the target content will be obtained first, and the collection, use and processing of these data will comply with the relevant laws, regulations and standards of the relevant countries and regions. In addition, when the embodiment of the present application needs to obtain the attribute information of the target content, it will obtain the separate permission or separate consent of the target content through a pop-up window or jump to a confirmation page. After clearly obtaining the separate permission or separate consent of the target content, the necessary target content-related data used to enable the normal operation of the embodiment of the present application will be obtained.

[0424] Fig.28 The structure diagram of the push processing device 2800 provided in the embodiment of the present disclosure. The push processing device 2800 includes:

[0425] The first acquisition unit 2810 is used to acquire a set of content to be pushed, where the set of content to be pushed includes a plurality of content to be pushed;

[0426] The second acquisition unit 2820 is used to acquire a set of time periods to be pushed, where the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle;

[0427] The first input unit 2830 is used to perform a first combination of the features of each content to be pushed in the content set to be pushed and each time period to be pushed in the time period to be pushed, and input each first combination together with the object features of the target object into the task completion probability prediction model to obtain a predicted first task completion probability, wherein the first task completion probability is a probability that the target object completes the task for the content to be pushed in the time period to be pushed;

[0428] A first determining unit 2840 is configured to determine a first target combination from among the plurality of first combinations based on a first task completion probability of each first combination;

[0429] The pushing unit 2850 is configured to use the time period to be pushed in the first target combination as the target time period, and push a reminder message corresponding to the content to be pushed to the target object during the target time period.

[0430] Optionally, the first input unit 2830 is specifically used to: input each first combination together with the object characteristics of the target object and the scene characteristics of the target object into a task completion probability prediction model to obtain a predicted first task completion probability; wherein the first task completion probability is the probability that the target object completes the task in the scenario during the period to be pushed for the content to be pushed.

[0431] Optionally, the pushing unit 2850 is specifically used for:

[0432] Obtain a sub-period set to be pushed of a target period, where the sub-period set to be pushed includes multiple sub-periods to be pushed;

[0433] Obtaining a first combination with a target period from among the plurality of first combinations, and filtering out a filtered first combination from among the first combinations with the target period based on the first task completion probability;

[0434] Perform a second combination on each first combination after screening and each sub-period to be pushed in the sub-period to be pushed set, input each second combination together with the object characteristics of the target object into the task completion probability prediction model, and obtain a predicted second task completion probability, wherein the second task completion probability is the probability that the target object completes the task for the content to be pushed in the sub-period to be pushed of the period to be pushed;

[0435] determining a second target combination among the plurality of second combinations based on the second task completion probability of each second combination;

[0436] The time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content. In the target sub-time period of the target time period, a reminder message corresponding to the target content is pushed to the target object.

[0437] Optionally, the push unit 2850 is specifically used to: input each second combination together with the object characteristics of the target object and the scene characteristics of the target object into a task completion probability prediction model to obtain a predicted second task completion probability; wherein the second task completion probability is the probability that the target object completes the task for the content to be pushed, in the sub-period to be pushed of the period to be pushed, and in the scene.

[0438] Optionally, the push processing device 2800 further includes:

[0439] A third acquisition unit (not shown), specifically configured to acquire additional content generated on the content push platform between the generation time of the content set to be pushed and the current time;

[0440] The adding unit (not shown) is specifically used to add the combination of the feature of the added content and the target time period to the first combination after screening.

[0441] Optionally, the pushing unit 2850 is specifically used for:

[0442] Determining a first number of first combinations having a first task completion probability greater than a first threshold value among the first combinations with the target time period;

[0443] If the first number is not less than the second threshold, the first number of first combinations are taken as the first combination after screening; otherwise, the first combinations before the second threshold are taken in descending order of the probability of completing the first task as the first combination after screening.

[0444] Optionally, the task completion probability prediction model is trained in the following way:

[0445] Obtain a sample push event set, wherein the sample push event set includes multiple sample push events, the sample push event includes sample object features, sample content features to be pushed, sample push period, and sample push sub-period, the sample push event has a task training label, and the task training label indicates whether the sample object completes the task for the sample content to be pushed in the sample push sub-period of the sample push period;

[0446] Extracting a first sample and a second sample from the sample push event, wherein the first sample includes a sample object feature, a sample content feature to be pushed, and a sample push period, and the second sample includes a sample object feature, a sample content feature to be pushed, a sample push period, and a sample push sub-period;

[0447] The first sample and the second sample extracted from each sample push event are input into the task completion probability prediction model for training.

[0448] Optionally, the first sample and the second sample extracted from each sample push event are input into a task completion probability prediction model for training, including:

[0449] For each sample in the first sample and the second sample, for a single task, based on the training label of the single task and the first task completion probability predicted by the task completion probability prediction model, determine an error subfunction of the single task;

[0450] Determine the error function of the sample based on the error sub-functions of multiple tasks;

[0451] Based on the error function of each sample, a task completion probability prediction model is trained.

[0452] Optionally, the task completion probability prediction model includes a vectorization layer, a cascade layer, a plurality of expert sub-models, a plurality of gating nodes corresponding to a plurality of tasks, and a plurality of prediction sub-models;

[0453] The first input unit 2830 is specifically used for:

[0454] Vectorize each first combination and the object feature of the target object, and concatenate the vectorization results to obtain a first feature vector;

[0455] inputting the first feature vector into a plurality of expert sub-models;

[0456] The outputs of multiple expert sub-models are weighted and summed through the gating nodes corresponding to each task to obtain a second feature vector corresponding to the task;

[0457] The second feature vector corresponding to the task is input into the prediction model corresponding to the task to obtain the first task completion probability corresponding to the task.

[0458] Optionally, the first input unit 2830 is specifically used for:

[0459] Obtain a weight vector of the gate node, where the weight vector indicates weights of multiple expert sub-models;

[0460] Through the gating node, the outputs of multiple expert sub-models are weighted and summed using the weight vector to obtain a second feature vector corresponding to the task.

[0461] Optionally, the first input unit 2830 is specifically used for:

[0462] Obtain a first matrix, wherein the number of rows of the first matrix is ​​the number of expert sub-models, and the number of columns is the dimension of the first eigenvector;

[0463] The first matrix is ​​multiplied by the transpose of the first eigenvector, and the multiplication result is subjected to normalized exponential processing to obtain a weight vector of the gated node.

[0464] Optionally, the first input unit 2830 is specifically used for:

[0465] The characteristics of the content to be pushed, the time period to be pushed, and the characteristics of the object are used as the basic characteristics of the push;

[0466] If the pushed basic feature is a character feature, the character feature is input into the embedding layer to obtain the vectorization result corresponding to the character feature;

[0467] If the pushed basic feature is a numerical feature, the numerical feature is numerically mapped to obtain the vectorized result corresponding to the numerical feature;

[0468] If the pushed basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain a vectorized result corresponding to the variable-length sequence feature.

[0469] Optionally, the first acquiring unit 2810 is specifically configured to:

[0470] Get the first label of the target object;

[0471] Get the second tag of the original content in the original content set;

[0472] Based on the matching degree between the first tag and the second tag, a content set to be pushed is filtered out from the original content set.

[0473] Optionally, the first acquiring unit 2810 is specifically configured to:

[0474] Based on the matching degree, matching contents having a first tag and a second tag matching each other are selected from the original content set;

[0475] Get multiple push metrics for each matching content;

[0476] Based on multiple push indicators, the content to be pushed is filtered out from the matching content to form a content set to be pushed.

[0477] Optionally, the first determining unit 2840 is specifically configured to:

[0478] Among multiple tasks, obtain the attention task of the target object;

[0479] Obtaining a first task completion probability of a focused task of the first combination from the first task completion probabilities of the plurality of tasks of the first combination;

[0480] A first target combination is determined from among the plurality of first combinations according to first task completion probabilities of the focus tasks of the plurality of first combinations.

[0481] Optionally, the first determining unit 2840 is specifically configured to:

[0482] If there are multiple tasks of interest, determine a first overall first task completion probability of the first combination based on the first task completion probabilities of the multiple tasks of interest in each first combination;

[0483] A first target combination is determined among the plurality of first combinations based on the first overall first task completion probabilities of the plurality of first combinations.

[0484] Reference Fig.29 , Fig.29The structural block diagram of the terminal part for implementing the push processing method of the embodiment of the present disclosure includes: a radio frequency (RF) circuit 2910, a memory 2915, an input unit 2930, a display unit 2940, a sensor 2950, ​​an audio circuit 2960, a wireless fidelity (WiFi) module 2970, a processor 2980, and a power supply 2990. Those skilled in the art can understand that Fig.29 The terminal structure shown does not constitute a limitation on the mobile phone or computer, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0485] RF circuit 2910 can be used for receiving and sending signals during information transmission or calls. In particular, after receiving the downlink information from the base station, it is sent to processor 2980 for processing; in addition, the designed uplink data is sent to the base station.

[0486] The memory 2915 may be used to store software programs and modules. The processor 2980 executes various functional applications and data processing of the content terminal by running the software programs and modules stored in the memory 2915 .

[0487] The input unit 2930 may be used to receive input digital or character information and generate key signal input related to the setting and function control of the content terminal. Specifically, the input unit 2930 may include a touch panel 2931 and other input devices 2932.

[0488] The display unit 2940 may be used to display input information or provided information and various menus of the content terminal. The display unit 2940 may include a display panel 2941.

[0489] The audio circuit 2960, the speaker 2961, and the microphone 2962 may provide an audio interface.

[0490] In this embodiment, the processor 2980 included in the terminal can execute the push processing method of the previous embodiment.

[0491] The terminals of the embodiments of the present disclosure include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The embodiments of the present invention can be applied to various scenarios, including but not limited to content recommendation, data screening, etc.

[0492] Fig.30A block diagram of the structure of a portion of a server for implementing the push processing method of an embodiment of the present disclosure. The server may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 3022 (for example, one or more processors) and memory 3032, and one or more storage media 2130 (for example, one or more mass storage devices) storing application programs 3042 or data 3044. Among them, the memory 3032 and the storage medium 3030 can be short-term storage or permanent storage. The program stored in the storage medium 3030 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 3022 can be configured to communicate with the storage medium 3030 and execute a series of instruction operations in the storage medium 3030 on the server.

[0493] The server may also include one or more power supplies 3026, one or more wired or wireless network interfaces 3050, one or more input and output interfaces 3058, and / or one or more operating systems 3041, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0494] The central processor 3022 in the server can be used to execute the push processing method of the embodiment of the present disclosure.

[0495] The embodiment of the present disclosure also provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the push processing methods of the aforementioned embodiments.

[0496] The embodiment of the present disclosure also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device executes and implements the above-mentioned push processing method.

[0497] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar contents, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "comprises" and "comprising" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0498] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated content, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the associated content before and after is in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0499] It should be understood that in the description of the embodiments of the present disclosure, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to not include the number, and above, below, within, etc. are understood to include the number.

[0500] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0501] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0502] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0503] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0504] It should also be understood that the various implementations provided in the embodiments of the present disclosure can be combined arbitrarily to achieve different technical effects.

[0505] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Technical personnel familiar with the art can also make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A push processing method, characterized in that: include: Acquire a set of content to be pushed, where the set of content to be pushed includes a plurality of content to be pushed; Acquire a set of time periods to be pushed, wherein the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle; Performing a first combination of the features of each of the contents to be pushed in the set of contents to be pushed and each of the time periods to be pushed in the set of time periods to be pushed, and inputting each of the first combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted first task completion probability, wherein the first task completion probability is a probability that the target object completes the task for the contents to be pushed in the time period to be pushed; determining a first target combination among a plurality of the first combinations based on the first task completion probability of each of the first combinations; The time period to be pushed in the first target combination is used as a target time period, and a reminder message corresponding to the content to be pushed is pushed to the target object during the target time period.

2. The push processing method according to claim 1, characterized in that: The step of inputting each of the first combinations together with the object feature of the target object into a task completion probability prediction model to obtain a predicted first task completion probability comprises: inputting each of the first combinations together with the object feature of the target object and the scene feature where the target object is located into a task completion probability prediction model to obtain a predicted first task completion probability; The first task completion probability is the probability that the target object completes the task for the content to be pushed in the scenario during the period to be pushed.

3. The push processing method according to claim 1, characterized in that: The step of pushing a reminder message corresponding to the content to be pushed to the target object during the target time period includes: Acquire a sub-period set to be pushed of the target period, wherein the sub-period set to be pushed includes a plurality of sub-periods to be pushed; Acquire the first combination with the target time period from among the plurality of the first combinations, and select a screened first combination from among the first combinations with the target time period based on the first task completion probability; Performing a second combination on each of the first combinations after screening and each of the sub-periods to be pushed in the sub-periods to be pushed, inputting each of the second combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted second task completion probability, wherein the second task completion probability is a probability that the target object completes the task for the content to be pushed in the sub-period to be pushed of the period to be pushed; determining a second target combination among a plurality of the second combinations based on the second task completion probability of each of the second combinations; The time period to be pushed in the second target combination is used as the target time period, the sub-time period to be pushed is used as the target sub-time period, and the content to be pushed is used as the target content, and in the target sub-time period of the target time period, a reminder message corresponding to the target content is pushed to the target object.

4. The push processing method according to claim 3, characterized in that: The step of inputting each of the second combinations together with the object feature of the target object into a task completion probability prediction model to obtain a predicted second task completion probability comprises: inputting each of the second combinations together with the object feature of the target object and the scene feature of the target object into a task completion probability prediction model to obtain a predicted second task completion probability; The second task completion probability is the probability that the target object completes the task for the content to be pushed, in the sub-period to be pushed of the period to be pushed, and in the scenario.

5. The push processing method according to claim 3, characterized in that: In the first combination with the target time period, after filtering out the filtered first combination based on the first task completion probability, the push processing method further includes: Acquire additional content generated on the content push platform between the generation time of the content set to be pushed and the current time; The combination of the feature of the added content and the target time period is added to the first combination after screening.

6. The push processing method according to claim 3, characterized in that: The step of screening out a first screening combination based on the first task completion probability includes: Determining, among the first combinations with the target time period, a first number of the first combinations having a probability of completing the first task greater than a first threshold; If the first number is not less than the second threshold, the first number of the first combinations is taken as the first combination after screening; otherwise, the first combinations before the second threshold are taken in descending order of the probability of completing the first task as the first combination after screening.

7. The push processing method according to claim 3, characterized in that: The task completion probability prediction model is trained in the following way: Acquire a sample push event set, wherein the sample push event set includes a plurality of sample push events, the sample push event includes a sample object feature, a sample content feature to be pushed, a sample to be pushed period, and a sample to be pushed sub-period, the sample push event has a task training label, and the task training label indicates whether the sample object completes the task for the sample content to be pushed and the sample to be pushed sub-period in the sample to be pushed period; Extracting a first sample and a second sample from the sample push event, wherein the first sample includes the sample object feature, the sample content feature to be pushed, and the sample to be pushed period, and the second sample includes the sample object feature, the sample content feature to be pushed, the sample to be pushed period, and the sample to be pushed sub-period; The first sample and the second sample extracted from each of the sample push events are input into the task completion probability prediction model for training.

8. The push processing method according to claim 7, characterized in that: The step of inputting the first sample and the second sample extracted from each sample push event into the task completion probability prediction model for training includes: For each of the first sample and the second sample, for a single task, based on the training label of the single task and the first task completion probability predicted by the task completion probability prediction model, determine an error subfunction of the single task; Determining an error function of the sample based on a plurality of error sub-functions of the tasks; Based on the error function of each of the samples, the task completion probability prediction model is trained.

9. The push processing method according to claim 1, characterized in that: The task completion probability prediction model includes a vectorization layer, a cascade layer, multiple expert sub-models, multiple gating nodes corresponding to the multiple tasks, and multiple prediction sub-models; The step of inputting each of the first combinations together with the object feature of the target object into a task completion probability prediction model to obtain a predicted first task completion probability includes: Vectorize each of the first combinations and the object feature of the target object, and concatenate the vectorization results to obtain a first feature vector; Inputting the first feature vector into a plurality of the expert sub-models; Through the gating nodes corresponding to each of the tasks, the outputs of the plurality of expert sub-models are weighted and summed to obtain a second feature vector corresponding to the task; The second feature vector corresponding to the task is input into the prediction model corresponding to the task to obtain the first task completion probability corresponding to the task.

10. The push processing method according to claim 9, characterized in that: The step of performing weighted summation on the outputs of the plurality of expert sub-models through the gating nodes corresponding to each of the tasks to obtain a second feature vector corresponding to the task includes: Obtaining a weight vector of the gating node, wherein the weight vector indicates weights of a plurality of the expert sub-models; The outputs of the plurality of expert sub-models are weighted and summed using the weight vector through the gating node to obtain a second feature vector corresponding to the task.

11. The push processing method according to claim 10, characterized in that: The obtaining of the weight vector of the gated node comprises: Obtain a first matrix, wherein the number of rows of the first matrix is ​​the number of the expert sub-models, and the number of columns is the dimension of the first eigenvector; The first matrix is ​​multiplied by the transpose of the first eigenvector, and a normalized exponential process is performed on the multiplication result to obtain the weight vector of the gated node.

12. The push processing method according to claim 9, characterized in that: The vectorizing each of the first combinations and the object features of the target object comprises: Taking the characteristics of the content to be pushed, the time period to be pushed, and the object characteristics as basic characteristics for pushing; If the pushed basic feature is a character feature, input the character feature into an embedding layer to obtain a vectorization result corresponding to the character feature; If the pushed basic feature is a numerical feature, numerical mapping is performed on the numerical feature to obtain a vectorized result corresponding to the numerical feature; If the pushed basic feature is an indefinite length sequence feature, the indefinite length sequence feature is pooled to obtain a vectorized result corresponding to the indefinite length sequence feature.

13. The push processing method according to claim 1, characterized in that: The step of obtaining the content set to be pushed includes: Obtaining a first label of the target object; Get the second tag of the original content in the original content set; The to-be-pushed content set is screened out from the original content set based on the matching degree between the first tag and the second tag.

14. The push processing method according to claim 13, characterized in that: The filtering out the to-be-pushed content set from the original content set based on the matching degree between the first tag and the second tag includes: Based on the matching degree, selecting matching content in which the first tag and the second tag match from the original content set; Obtaining multiple push indicators for each of the matching contents; Based on the plurality of push indicators, the content to be pushed is screened out from the matching content to form the content set to be pushed.

15. The push processing method according to claim 1, characterized in that: The determining a first target combination from a plurality of the first combinations based on the first task completion probability of each of the first combinations comprises: Acquire the focus task of the target object from among the multiple tasks; Acquire the first task completion probability of the focus task of the first combination from the first task completion probabilities of the plurality of tasks of the first combination; A first target combination is determined among the plurality of first combinations according to the first task completion probabilities of the focus tasks of the plurality of first combinations.

16. The push processing method according to claim 15, characterized in that: The determining a first target combination from a plurality of the first combinations according to the first task completion probabilities of the focus tasks of the plurality of the first combinations comprises: If there are multiple tasks of interest, determining a first overall first task completion probability of the first combination based on the first task completion probabilities of the multiple tasks of interest in each of the first combinations; A first target combination is determined among the plurality of first combinations based on the first overall first task completion probabilities of the plurality of first combinations.

17. A push processing device, characterized in that: include: A first acquisition unit is used to acquire a set of content to be pushed, wherein the set of content to be pushed includes a plurality of content to be pushed; A second acquisition unit is used to acquire a set of time periods to be pushed, wherein the set of time periods to be pushed includes a plurality of time periods to be pushed divided into a predetermined push cycle; A first input unit is used to perform a first combination of the features of each of the to-be-pushed content in the to-be-pushed content set and each of the to-be-pushed time periods in the to-be-pushed time period set, and input each of the first combinations together with the object features of the target object into a task completion probability prediction model to obtain a predicted first task completion probability, wherein the first task completion probability is a probability that the target object completes the task for the to-be-pushed content in the to-be-pushed time period; a first determining unit, configured to determine a first target combination among a plurality of the first combinations based on the first task completion probability of each of the first combinations; The pushing unit is used to use the time period to be pushed in the first target combination as a target time period, and push a reminder message corresponding to the content to be pushed to the target object during the target time period.

18. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the push processing method according to any one of claims 1 to 16 is implemented.

19. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the push processing method according to any one of claims 1 to 16 is implemented.

20. A computer program product, comprising a computer program, wherein the computer program is read and executed by a processor of a computer device, so that the computer device executes the push processing method according to any one of claims 1 to 16.

Citation Information

Patent Citations

  • Message pushing method and device, computer equipment and storage medium

    CN111460294A

  • Push time prediction method and device, equipment, storage medium and program product

    CN113382087A

  • Media resource recommendation method, and multi-target fusion model training method and device

    CN113516522A