CN117527743A - Message pushing method, training method, device, electronic equipment and storage medium - Google Patents

Abstract

The disclosure provides a message pushing method, a model training method, a message pushing device, a model training device, electronic equipment, a storage medium and a program product, relates to the technical field of data processing, and particularly relates to the technical field of content pushing. The specific implementation scheme is as follows: judging whether the push time of the push list belongs to the active time of the user according to a first association relation between the user characteristics and the active time; responding to the active time of the pushing time belonging to the user, and determining click frequency corresponding to the pushing time according to the user characteristics; determining the pushing quantity according to the click frequency; and pushing the messages in the push list according to the pushing quantity.

Description

Message pushing method, training method, device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of data processing technologies, and in particular, to the field of content pushing technologies, and in particular, to a message pushing method, a model training method, a message pushing device, a model training device, an electronic device, a storage medium, and a program product.

Background

Currently, many applications are provided with message push functionality. Specifically, the application may push some application-related messages to the user terminal, so as to attract the attention of the user, improve the frequency of using the application by the user, and enhance the user viscosity.

Disclosure of Invention

The disclosure provides a message pushing method, a model training method, a message pushing device, a model training device, electronic equipment, a storage medium and a program product, and the message pushing method, the model training device, the electronic equipment, the storage medium and the program product can be used for pushing messages by combining the active time and the click frequency of a user.

According to an aspect of the present disclosure, there is provided a message pushing method, including: judging whether the push time of the push list belongs to the active time of the user according to a first association relation between the user characteristics and the active time; responding to the active time of the pushing time belonging to the user, and determining click frequency corresponding to the pushing time according to the user characteristics; determining the pushing quantity according to the click frequency; and pushing the messages in the push list according to the pushing quantity.

According to another aspect of the present disclosure, there is provided a model training method including: extracting historical characteristic data from historical behavior data, wherein the historical characteristic data comprises historical user characteristics, historical push quantity characteristics and historical click time characteristics; training a content push model by using the historical user characteristics, the historical push quantity characteristics and the historical click time characteristics; the trained content push model at least characterizes a first association relationship and a second association relationship, the first association relationship at least indicates an association relationship between user features and active time, and the second association relationship indicates an association relationship between user features, time features and click frequency.

According to another aspect of the present disclosure, there is provided a message pushing apparatus, including: the first judging module is configured to judge whether the push time of the push list belongs to the active time of the user according to a first association relation between the user characteristics and the active time; the second judging module is configured to respond to the active time of the user, and determine click frequency corresponding to the push time according to the user characteristics; the quantity determining module is configured to determine the pushing quantity according to the click frequency; and the pushing module is configured to push the messages in the pushing list according to the pushing quantity.

According to another aspect of the present disclosure, there is provided a model training apparatus including: the extraction module is configured to extract historical feature data from the historical behavior data, wherein the historical feature data comprises historical user features, historical push quantity features and historical click time features; the training module is configured to train the content pushing model by using the historical user characteristics, the historical pushing quantity characteristics and the historical click time characteristics; the trained content push model at least characterizes a first association relationship and a second association relationship, the first association relationship at least indicates an association relationship between user features and active time, and the second association relationship indicates an association relationship between user features, time features and click frequency.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as mentioned above.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the above-mentioned method.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the above mentioned method.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 illustrates a schematic block diagram of an exemplary system architecture to which the message pushing method or model training method of the present disclosure may be applied;

Fig. 2 is a flow diagram of a message pushing method according to a first embodiment of the present disclosure;

fig. 3 is a flow diagram of a message pushing method according to a second embodiment of the present disclosure;

fig. 4 is a flow diagram of a message pushing method according to a third embodiment of the present disclosure;

FIG. 5 is a flow diagram of a model training method according to a fourth embodiment of the present disclosure;

fig. 6 is a schematic block diagram of a message pushing device according to a fifth embodiment of the present disclosure;

FIG. 7 is a schematic block diagram of a model training apparatus according to a sixth embodiment of the present disclosure;

FIG. 8 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

Fig. 1 shows a schematic block diagram of an exemplary system architecture to which the message pushing method or model training method of the present disclosure may be applied.

As shown in fig. 1, a system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various client applications, such as video-type applications, live applications, instant messaging tools, mailbox clients, social platform software, etc., may be installed on the terminal devices 101, 102, 103.

The terminal devices 101, 102, 103 may be hardware or software. When the terminal devices 101, 102, 103 are hardware, they may be various electronic devices with display screens, including but not limited to smartphones, tablets, electronic book readers, laptop and desktop computers, and the like. When the terminal devices 101, 102, 103 are software, they can be installed in the above-listed electronic devices. Which may be implemented as multiple software or software modules (e.g., multiple software or software modules for providing distributed services) or as a single software or software module. The present invention is not particularly limited herein.

The server 105 may be a server providing various services, such as a background server providing support for the terminal devices 101, 102, 103. For example, the background server may process to provide a message push service to the terminal devices based on data uploaded by the respective terminal devices.

It should be noted that, the message pushing method or the model training method provided by the embodiments of the present disclosure may be executed by the server 105 or the terminal devices 101, 102, 103, and accordingly, the message pushing device or the model training device may be disposed in the server 105 or the terminal devices 101, 102, 103.

It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

With continued reference to fig. 2, fig. 2 is a flow diagram of a message pushing method according to a first embodiment of the present disclosure. The execution body of the message pushing method 200 may be a server, cloud computing resource, or the like that provides support for applications. As shown in fig. 2, the message pushing method 200 may include the following steps:

step 201, according to a first association relationship between the user characteristics and the active time, it is determined whether the push time of the push list belongs to the active time of the user.

In this embodiment, the execution body may acquire user personal information from the terminal device of the user, and determine the user characteristics according to the user personal information. The user personal information may include, among other things, user identification (id), region, age, gender, etc., without limitation. Based on the user personal information, the execution subject extracts the user characteristics through characteristic engineering or other technologies. The user characteristics may be represented in the form of vectors, etc., without limitation.

It should be noted that, the region, sex, etc. included in the personal information of the user in this embodiment are obtained after the personal information is provided by the user (i.e. the user's own consent is signed), including but not limited to before the user uses the function, notifying the user to read the user protocol (notification), and signing the protocol (authorization) including the information about the authorized user.

It should be appreciated that the executing body may extract the user features during each push determination without departing from the teachings of the present disclosure, or may store the user features and user identifications for subsequent use after the user features are first extracted, which is not limited by the present disclosure.

In this embodiment, after the execution body obtains the push list, it may be determined, according to the user feature and the first association relationship, whether the push time of the push list belongs to the active time of the user. Wherein the first association indicates an association between the user characteristic and the active time.

It should be appreciated that the push time of the push list may be the time the executing body obtains the push list, may be the time the push list is generated, may be the time the push list is carried in a data packet containing the push list, and is not limited herein.

As one example, the first association may be learned from historical behavior data of a plurality of historical users. For example, the execution subject may extract historical user features and historical click time features from historical behavior data of a plurality of users, and learn a first association relationship based on the extracted historical user features and historical click time features. Wherein the historical click time feature may be determined based on the time the historical user clicked on the pushed message, without limitation. In this example, the execution subject mines deep associations between user features and active times from historical behavioral data so that active times of other users can be estimated later based on the deep associations.

As one example, the first association is represented in a model form. For example, a first decision model is built in the execution body, the first decision model being a pre-trained network model for characterizing the association between different user characteristics and active time. The execution subject may input the user characteristic into the first determination model to determine whether the push time belongs to the active time.

For example, the first decision model determines, according to the input user characteristics, the push time and the first association relation mined in the learning stage, a probability that the push time is an active time, if the probability is greater than a probability threshold, an indication identifier indicating that the push time belongs to the active time may be output, and if the probability is less than or equal to the probability threshold, an indication identifier indicating that the push time does not belong to the active time may be output.

It should be appreciated that, without departing from the teachings of the present disclosure, the first determination model may also be configured to output an active time of the user based on the input user characteristics, and the execution subject may determine whether the push time belongs to the active time, where the first determination model may embody an association between the user characteristics and the active time, and the present disclosure does not limit the input and output of the first determination model.

It should be appreciated that the first decision model may be a deep learning model, such as a deep neural network (Deep Neural Networks, DNN), long Short-term memory network (LSTM), or other model, without departing from the teachings of the present disclosure, which is not limited in this regard.

It should be appreciated that the first association may also be obtained and expressed in other forms without departing from the teachings of the present disclosure, e.g., the first association may be embodied in a list form, e.g., the executing body may determine and store, for each user, their corresponding activity time separately in a pre-built activity schedule that embodies the first association. The present disclosure is not limited in this regard.

In the above embodiment, the execution body determines whether to push or not according to the active time of the user in the pushing process based on the push list, so that the click probability after the message is pushed can be improved, and the dislike degree of the user on the pushed message can be reduced.

Step 202, in response to the push time belonging to the active time of the user, determining the click frequency corresponding to the push time according to the user characteristics.

In this embodiment, after determining that the push time belongs to the active time of the user, the executing body determines the click frequency of the user in the time period corresponding to the push time, and uses the click frequency as the click frequency corresponding to the push time.

As an example, the execution subject may obtain an association relationship (i.e., a second association relationship) between the user feature, the time feature, and the click frequency, and determine the click frequency corresponding to the push time according to the user feature, the push time, and the second association relationship.

Alternatively, the second association is represented in a model form. The execution main body inputs the user characteristics and the time characteristics corresponding to the pushing time into a second judgment model, and determines the click frequency corresponding to the pushing time, wherein the second judgment model is used for representing a second association relationship among the user characteristics, the time characteristics and the click frequency, and is learned from historical behavior data of a plurality of historical users. For example, the learning process of the second association relationship may include: dividing a time of day into a plurality of time periods; extracting historical user characteristics, historical push quantity characteristics, historical click time characteristics and the like from the historical behavior data of each historical user; determining the historical click times in each time period based on the historical click time characteristics, and determining the historical click frequency according to the historical click times in each time period and the historical pushing quantity characteristics of each time period; and learning a second association relation according to the determined historical user characteristics, the time characteristics of each time period and the historical click frequency of each time period to obtain a second judging model. In this example, the execution body mines the deep association relationship among the user features, the time features and the click frequency from the historical behavior data of the historical user of the application, so that the click frequency of other users at each time can be estimated based on the deep association relationship in the follow-up process, and the number of pushing messages can be controlled more accurately.

It should be appreciated that the second decision model may be a deep learning model, e.g., DNN, LSTM, etc., as well as other models, without departing from the teachings of the present disclosure, which is not limited in this disclosure.

It should be appreciated that the second association may be obtained and expressed in other manners without departing from the teachings of the present disclosure, e.g., the second association may be embodied in a list form, e.g., the executing entity may determine, for each user, how often it clicks for each time period, and store it in a pre-built active schedule that embodies the second association. The present disclosure is not limited in this regard.

Step 203, determining the pushing quantity according to the click frequency.

In this embodiment, since the click frequency can embody the frequency of the messages pushed by the users in the time period corresponding to the push time, the execution body determines the push number based on the click frequency, so that the push number can be better ensured to be in the tolerance range of the users, and the use experience of the users is improved.

As one example, the execution body may determine a push threshold for the user, and determine the number of pushes based on the push threshold and the click frequency.

For example, the execution body may calculate the product of the push threshold and the click frequency, and take the product as the push number.

For another example, the execution body may calculate a product of the push threshold and the click frequency as the reference push number, and determine the push number according to the reference push number, the counted push total amount, and the maximum push amount. For example, the execution body may determine a maximum push amount of each user, calculate a difference between the maximum push amount and the push total amount, determine the difference as the push amount if the difference is smaller than the reference push amount, and determine the reference push amount as the push amount if the difference is greater than or equal to the reference push amount. In the example, the maximum pushing amount is set for each user, so that physical examination of the user is improved, and the situation that the user is disliked due to excessive total pushing amount in the same day, and then application pushing is unloaded or closed is reduced.

It should be appreciated that the maximum push amounts for different users may be the same without departing from the teachings of the present disclosure, e.g., a maximum push amount is pre-given as the maximum push amount for all users. The maximum pushing amounts of the different users may also be different, for example, given training data including historical user characteristics and labels of the respective historical user characteristics, the labels indicating the maximum pushing amounts, based on which training data a model for determining the maximum pushing amounts of the different user characteristics is trained, which is used by the executing body to determine the maximum pushing amounts of the current user. The present disclosure is not limited in this regard.

It should be appreciated that the maximum push of the user may be the maximum push per day or the maximum push per time period of each day without departing from the teachings of the present disclosure, which is not limited in this disclosure.

It should be appreciated that the push thresholds for different users may be the same without departing from the teachings of the present disclosure, e.g., a push threshold may be pre-given as the push threshold for all users, or the push thresholds for different users may be different, e.g., pre-given training data comprising historical user characteristics and labels for each of the historical user characteristics indicating push thresholds, training a model for determining the push thresholds for different user characteristics based on the training data, the executing subject using the model to determine the recommendation threshold for the current user. The present disclosure is not limited in this regard.

It should be appreciated that the push threshold of the user may be a daily push threshold or a push threshold for each time period of the day without departing from the teachings of the present disclosure, which is not limited in this disclosure.

It should be appreciated that the executing body may also determine the push number according to the click frequency in combination with other parameters without departing from the teachings of the present disclosure, for example, determine the push number according to the click frequency and the number of messages in the push list, e.g., taking the product of the click frequency and the number of messages in the push list as the push number, which is not limited by the present disclosure.

Step 204, pushing the message in the push list according to the pushing quantity.

In this embodiment, after determining the pushing number, the executing body may push the message in the push list based on the pushing number. For example, if the determined push number n=0, the execution body may not push the message. If N is not equal to 0 and the number of messages M in the push list is greater than N, the execution body may randomly select N messages from the push list to push, or select N messages with the highest scores in the push list to push, where the scores of the messages may be user interest values, and the like, and the method is not limited herein. If N is not equal to 0 and M is not equal to N, the execution body can push all messages in the push list to the user.

According to some embodiments of the present disclosure, the execution body may determine whether to push a message to a user based on whether the push time belongs to an active time of the user, thereby implementing time control of the push message, determining the number of messages pushed to the user based on a click frequency of the user, and implementing number control of the push messages. The message pushing is controlled from two aspects of pushing time and the number of pushing messages, so that the situation that negative operations such as unloading application or closing pushing functions are caused due to the fact that the pushing control is inaccurate to cause burden to a user can be reduced.

Fig. 3 is a flow diagram of a message pushing method according to a second embodiment of the present disclosure. As shown in fig. 3, the message pushing method 300 may include the following steps:

step 301, judging whether the push time belongs to the active time of the user for the content feature according to the user feature, the content feature of the message in the push list and the first association relation.

In this embodiment, the first association relationship further indicates an association relationship among the user feature, the content feature of the message and the active time, and considers whether to execute the subsequent pushing operation in combination with the active time of the user on the messages of different types, so that the message pushing is more accurate. For example, the time of interest of the user for different types of messages may be different, e.g., the user may like to watch entertainment class messages in the morning to keep happy mood open for a new day, like to click on news class messages in the evening to learn what happens on the day. In the embodiment, the preference degree of the user to various messages in different time periods is fully considered, whether the messages in the push list are pushed to the user in the push time is determined, the probability of being clicked after the messages are pushed is improved, and the use experience of the user is improved.

It should be appreciated that the content features may be text features of the pushed message or video features corresponding to the pushed message without departing from the teachings of the present disclosure, which is not limited in this disclosure.

As one example, the first association is learned from historical behavior data of a plurality of historical users. For example, the execution body may extract a history user feature, a content feature of a message of a history click, and a history click time feature from history behavior data of a plurality of users, and learn a first association relationship based on the extracted features. In this example, the execution subject mines deep associations between user features, content features, and temporal features from historical behavioral data, so that the activity time of other users for messages of different content features can be estimated subsequently based on the deep associations.

It should be appreciated that the first association may also be learned from historical behavioral data of the user without departing from the teachings of the present disclosure, or the first association may be preset, which is not limiting of the present disclosure.

As one example, the first association is represented in a model form. For example, the execution subject has built therein a first determination model, which is a pre-trained network model for characterizing the association relationship between the user characteristics, the content characteristics of the message, and the active time. The execution body may input the user characteristic and the content characteristic of the message in the push list into the first determination model to determine whether the push time belongs to an active time of the user for the content characteristic of the message in the push list. For example, the first decision model outputs the user's active time for the content feature based on the input user feature and the content feature. And the execution body judges whether the push time belongs to the active time of the user for the content characteristics of the messages in the push list according to the output active time and the push time.

It should be appreciated that the execution body may also input the temporal feature corresponding to the push time into the first decision model without departing from the teachings of the present disclosure. The first judging model determines the probability that the push time is the active time of the user for the content feature according to the input user feature, the content feature, the time feature of the push time and the first association relation mined in the learning stage, if the probability is larger than a probability threshold, an indication mark indicating that the push time belongs to the active time of the user for the content feature can be output, and if the probability is smaller than or equal to the probability threshold, an indication mark indicating that the push time does not belong to the active time of the user for the content feature can be output. The execution main body judges whether the pushing time belongs to the active time according to the indication mark, and the first judgment model shows the association relation among the user characteristics, the content characteristics of the message and the active time.

It should be appreciated that the first decision model may be a deep learning model, e.g., DNN, LSTM, or other model, without departing from the teachings of the present disclosure, which is not limited in this disclosure.

It should be appreciated that the first association may also be represented in other forms without departing from the teachings of the present disclosure, e.g., the first association may be represented in a list form, e.g., the execution subject may determine and store respective times of activity for the respective users for different content features in a pre-built activity schedule that represents the first association. The present disclosure is not limited in this regard.

Step 302, in response to the push time belonging to the active time of the user, determining the click frequency corresponding to the push time according to the user characteristics.

Step 303, determining the pushing quantity according to the click frequency.

Step 304, pushing the message in the push list according to the pushing quantity.

In this embodiment, steps 302 to 304 are substantially the same as steps 202 to 204 in fig. 2, and are not repeated here.

According to some embodiments of the present disclosure, an executing body may determine whether to push a message to a user based on whether the push time belongs to an active time of the user, thereby implementing time control of the push message, determining a number of messages pushed to the user based on a click frequency of the user, implementing number control of the push message, and reducing a burden on the user due to inaccurate push control, thereby causing a situation of unloading negative operations such as application or closing a push function. In addition, the execution main body fully considers the preference degree of the user to various messages in different time periods, determines whether to push the messages in the push list to the user in push time, improves the probability of being clicked after the message is pushed, and improves the use experience of the user.

Fig. 4 is a flow chart of a message pushing method according to a third embodiment of the present disclosure. As shown in fig. 4, the message pushing method 400 may include the following steps:

step 401, determining whether the push time of the push list belongs to the active time of the user according to the first association relationship.

In this embodiment, if the execution subject determines that the push time of the push list does not belong to the active time of the user, step 402 is executed, and if it determines that the push time of the push list belongs to the active time of the user, step 403 is executed. The process of determining whether the push time of the push list belongs to the active time of the user by the executing body may refer to the related content of the first embodiment or the second embodiment, which is not described herein.

Step 402, updating the push time according to the active time of the user.

In this embodiment, the executing body may determine an active time corresponding to the push time based on the first association relationship mentioned above, and update the push time based on the active time. For example, one of a start time, an intermediate time, and an end time of the active time is determined as the push time, and taking the example that the first association relationship indicates the association relationship between the user feature and the active time, the execution subject may input the user feature into the above-mentioned first determination model which takes the user feature as input and takes the active time as output, and the first determination model determines the active time corresponding to the user feature based on the input user feature and the learned first association relationship. And the execution main body determines that the push time belongs to the active time if the push time is determined to be in a time period of a certain active time based on the output of the first judging model, and determines the active time which is closest to the push time and is later than the push time as the active time corresponding to the push time if the push time is determined not to be in a time period of any active time.

It should be appreciated that the executing entity may also determine the active time corresponding to the push time by other means without departing from the teachings of the present disclosure, which is not limited in this disclosure.

Step 403, determining click frequency corresponding to the pushing time according to the user characteristics.

Step 404, determining the pushing quantity according to the click frequency.

Step 405, pushing the message in the push list according to the pushing number.

In this embodiment, steps 403 to 405 are substantially the same as steps 202 to 204 in fig. 2, and will not be described here.

It should be appreciated that the executing body may initiate the time monitoring task and perform steps 403 and 404 after updating the active time without departing from the teachings of the present disclosure; after the time monitoring task indicates that the current time has reached the updated push time, step 405 is performed. The execution body may also start the time monitoring task after updating the active time, and execute steps 403 to 405 after the time monitoring task indicates that the current time reaches the updated push time, which is not limited herein.

In some embodiments of the present disclosure, the time monitoring task may be implemented based on a timer. For example, based on the active time and the time of updating the push time, the timing time value of the timer is determined, so that the executing body determines whether the current time reaches the updated push time by monitoring the timer, thereby achieving the effect of delaying push.

As one example, the process of executing the subject start time monitoring task may include: judging whether a time monitoring task matched with the updated push time exists or not according to the updated push time and the monitoring time period of each existing time monitoring task; if yes, monitoring whether the current time reaches the updated push time or not through the matched time monitoring task; if it is determined that the current time does not exist, a new time monitoring task is created based on the updated push time to monitor whether the current time of task monitoring reaches the updated push time. For example, the time monitoring task may include a monitoring period and an information list for storing a user identification of the user and an identification of the push list. After determining the updated push time, if it is determined that the monitoring time period of an existing time monitoring task includes the updated push time, the executing body may add the user identifier of the user and the list identifier of the push list to the information list of the time monitoring task; if it is determined that the monitoring time periods of the existing time monitoring tasks do not include updated push time, a new time monitoring task can be created, wherein the monitoring time periods of the time monitoring tasks include time periods corresponding to the updated push time, and a user identifier of the user and a list identifier of the push list are recorded in the information list. After each time monitoring task monitors that the current time reaches the monitoring time period, each piece of information in the information list of the time monitoring task is acquired, user characteristics are determined based on user identifiers of each piece of information, a recommendation list is determined based on list identifiers corresponding to each user identifier, and steps 403 to 405 are executed. In this example, the execution body may create a time monitoring task for multiple push lists together, which may reduce the number of tasks and reduce resource consumption.

It should be understood that, without departing from the teachings of the present disclosure, the executing body may close the pushing task of the pushing list after determining that the pushing time does not belong to the active time of the user, that is, if the determination result in step 401 is no, the flow ends. The present disclosure does not limit the operation measures in the case where the push time does not belong to the active time of the user.

According to some embodiments of the present disclosure, an executing body may determine whether to push a message to a user based on whether the push time belongs to an active time of the user, thereby implementing time control of the push message, determining a number of messages pushed to the user based on a click frequency of the user, implementing number control of the push message, and reducing a burden on the user due to inaccurate push control, thereby causing a situation of unloading negative operations such as application or closing a push function. In addition, the effect of delaying pushing can be achieved by updating the pushing time, the probability of pushing the pushing list is improved, and the clicking probability of a user is improved.

With continued reference to fig. 5, fig. 5 is a flow diagram of a model training method according to a fourth embodiment of the present disclosure. The execution subject may be, for example, a server, etc., as shown in fig. 5, the model training method 500 may include the steps of:

Step 501, extracting historical feature data from historical behavior data.

In this embodiment, the history feature data includes a history user feature, a history push number feature, and a history click time feature. The historical behavior data may be a public data set about clicking behavior of the user on the pushed message, or may be behavior data uploaded by the client.

It should be noted that, the user information/behavior data in this embodiment is obtained after the user is authorized to provide (i.e., the user is authorized to agree), including but not limited to, before the user uses the message pushing function, the user is notified to read the user protocol (notification), and the protocol (authorization) including the authorization related user information/behavior data is signed.

Alternatively, the execution body may continuously acquire the behavior data of the client and store the behavior data in the historical behavior database. After the data in the historical behavior database reaches a certain amount, the execution subject can perform feature extraction on the historical behavior data again based on offline feature engineering or other technologies, and perform model training again to update the content push model.

As one example, the historical user characteristics may indicate information of a user identification, region, age, gender, etc. of the historical user, without limitation herein.

As one example, since the time granularity is as small as a second level, which is not significant to the judgment of the user's active time, in this embodiment, the time granularity may be classified into a minute level or an hour level, for example, a 10 minute granularity, or a 1 hour granularity in determining the historical click time feature. In other words, the execution body may determine the historical click time feature corresponding to the behavior log based on the time and the division granularity in the behavior log indicating the user performing the message click operation in the historical behavior data. For example, if the granularity of division is 10 minutes, the behavior log indicates that the user clicks on the message at 10:03:30, and the historical click time feature corresponding to the behavior log may be a time feature corresponding to a preset time period of 10:00:00-10:10:00.

As one example, the historical push quantity feature may be determined from a total number of messages received for each time period divided in advance. For example, a day may be divided into the following time periods: and counting the total number of messages received in the time periods respectively from 6 points to 9 points, from 9 points to 12 points, from 12 points to 18 points, from 18 points to 23 points and from 23 points to 6 points so as to determine the historical pushing quantity characteristics corresponding to each time period.

It should be appreciated that the executing body may also extract other features from the historical behavior data without departing from the teachings of the present disclosure, e.g., the executing body may also extract historical content features of historical clicked messages from the historical behavior data in order to subsequently learn the associations between user features, content features, and active times. As another example, the executive body may extract historical viewing time characteristics of the message from the historical behavior data in order to learn the first association in conjunction with the historical viewing time characteristics of the message, as this disclosure is not limited in this regard.

Step 502, training a content push model using the historical user characteristics, the historical push quantity characteristics, and the historical click time characteristics.

In this embodiment, the trained content push model at least characterizes a first association relationship and a second association relationship, where the first association relationship at least indicates an association relationship between the user feature and the active time, and the second association relationship indicates an association relationship between the user feature, the time feature and the click frequency. The content push model is applicable to the message push method mentioned above, so that the execution subject of the message push method pushes the messages in the push list based on the first association relationship and the second association relationship characterized by the content push model. The pushing process may refer to the related description above, and will not be described herein.

In some embodiments of the present disclosure, the process of executing the subject training content push model may include: training a first learning network by using the historical user characteristics and the historical click time characteristics to obtain a first judging model in the content pushing model, wherein the first judging model is used for representing a first association relation; and training a second learning network by using the historical user characteristics, the historical pushing quantity characteristics and the historical click time characteristics to obtain a second judging model in the content pushing model, wherein the second judging model is used for representing a second association relation. In this example, the first association and the second association are respectively learned using two learning networks, and the learning time of each learning network can be reduced.

It should be appreciated that the first and second associations may also be characterized by a model, i.e., a historical user feature, a historical push quantity feature, and a historical click time feature, based on which a learning network is trained to obtain a content push model without departing from the teachings of the present disclosure.

The training process of the first learning network is exemplarily described below.

As one example, the execution body may construct training data of the first learning network based on the historical user characteristics and the historical click time characteristics to learn an association of the user characteristics and the active time.

For example, a first type of sample may be included in the constructed training data, and sample data for the first type of sample may include historical user characteristics and historical click time characteristics, and a tag for the sample data may be, for example, an active identifier. Optionally, the training data further includes a second type of sample, sample data of the second type of sample may include a historical user feature and a historical non-clicked time feature, and a tag of the sample data may be, for example, an inactive flag. For example, the active flag is 1, and the inactive flag is 0. The historical non-clicked time feature may be determined based on a behavior log indicating that the user performs the message clicking operation, for example, for each day of log, a time period corresponding to a time point in the behavior log indicating that the user performs the message clicking operation in one day is removed, and a time feature corresponding to the remaining time period is taken as the historical non-clicked time feature. The first determination model trained based on the training data can determine whether the user is active at the push time according to the input user characteristics and the push time, i.e. whether the push time belongs to the active time of the user.

It should be understood that the historical non-click time characteristics may also be determined by other means without departing from the teachings of the present disclosure, which is not limited in this regard.

For another example, sample data in the constructed training data may include historical user characteristics and labels of the sample data may include historical click time characteristics. The first decision model trained based on the training data may output an activity time of the user based on the input user characteristics.

Optionally, the history feature data further includes history content features of the history clicked message. The training of the first learning network to obtain a first decision model in the content push model using the historical user characteristics and the historical click time characteristics by the executing entity may include: training a first learning network by using the historical user characteristics, the historical click time characteristics and the historical content characteristics to obtain a first judging model, wherein the first association relationship represented by the first judging model also indicates the association relationship among the user characteristics, the content characteristics of the message and the active time. For example, the execution body may construct training data for the first learning network based on the historical user characteristics, the historical content characteristics, and the historical click time characteristics, and train the first learning network based on the training data.

For example, sample data in the training data may include historical user characteristics, historical content characteristics, and historical click time characteristics, and the label of the sample data may be, for example, an active identification. The first determination model trained based on the training data may determine, according to the input user characteristics, content characteristics and push time, whether the user's reaction to the content characteristics at the push time is active, i.e. whether the push time belongs to the user's active time for the content characteristics.

As another example, sample data in the training data may include historical user features and historical content features, and the tags of the sample data may be, for example, historical click time features. The first determination model trained based on the training data may determine an active time of the user for the content feature based on the input user feature and the content feature, so as to subsequently determine whether the push time belongs to the active time of the user for the content feature.

Because the execution body extracts the content characteristics of the message clicked by the user in the historical behavior data, the first learning network can learn the association relationship among the user characteristics, the content characteristics of the message and the active time, and the obtained first judgment model can characterize the association relationship for subsequent use.

Optionally, the historical feature data further includes a historical viewing time feature of the message. The training of the first learning network to obtain a first decision model in the content push model using the historical user characteristics and the historical click time characteristics by the executing entity may include: first training data of the first learning network is constructed based on the historical user features and the historical click time features, second training data of the first learning network is constructed based on the historical user features and the historical view time features, and the first learning network is trained based on the first training data and the second training data.

The training process of the second learning network is exemplarily described below.

As one example, the historical push quantity feature extracted from the historical behavior data by the execution body may be determined based on the total number of messages received over different time periods. The execution body may determine the number of historical clicks in the above-described time period based on the historical click time characteristics, and determine the historical click frequency characteristics of each time period based on the number of historical clicks in each time period and the historical push number characteristics of each time period. The execution body constructs training data of the second learning network according to the determined historical user characteristics, the time characteristics of each time period and the historical click frequency characteristics of each time period, each sample data in the training data can comprise the historical user characteristics and the time characteristics corresponding to each time period, and the labels of the sample data can be the historical click frequency characteristics, for example. And learning the association relation among the user characteristics, the time characteristics and the click frequency of the training data reaction by using a second learning network, so that the second judgment model can judge the click frequency of the user in the time period to which the push time belongs according to the input user characteristics and the push time.

Alternatively, the execution subject monitors the operation of the historical user to close the push function or the historical user to uninstall the application in the process of extracting the features in the historical behavior data, so as to be used as a negative feedback sample for determining the tolerance degree of the user to push. For example, after detecting that the historical behavior data includes a behavior log indicating that the pushing function is closed by the historical user or the application is uninstalled by the historical user, the execution subject may generate historical feature data according to a recording time of the behavior log, where the historical feature data indicates that a historical click frequency feature of a historical user feature corresponding to the historical user in a time period corresponding to the recording time is a specified value. Wherein the specified value may be, for example, a value of 0 or less than 0, such that the finally learned second association indicates a lower click frequency for the period of time. In this example, the execution subject may construct a negative feedback sample based on the user closing the push function or unloading the application, increasing the amount of data used for learning, so that the learned second association relationship more accurately reflects the user's tolerance to push.

It should be appreciated that the first learning network and the second learning network may be DNNs, LSTM, etc., without departing from the teachings of the present disclosure, which is not limited in this disclosure.

According to some embodiments of the present disclosure, an executing body may learn an association relationship between user characteristics and active times, and an association relationship between user characteristics, active times, and click frequencies, based on historical behavior data, in order to construct a content push model. The execution body performs message pushing based on the constructed content pushing model, so that the time and the number of messages can be pushed more accurately.

The above steps of the various methods are divided, for clarity of description, and may be combined into one step or split into multiple steps when implemented, so long as they include the same logic relationship, and they are all within the protection scope of the present disclosure; it is within the scope of this disclosure to add insignificant modifications to the algorithm or flow or introduce insignificant designs, but not to alter the core design of its algorithm and flow.

In the technical scheme of the disclosure, the acquisition, storage, application and the like of the related user personal information all conform to the regulations of related laws and regulations, and the public sequence is not violated.

Fig. 6 is a schematic block diagram of a message pushing device according to a fifth embodiment of the present disclosure. As shown in fig. 6, the message pushing device 600 may include: a first decision module 601, a second decision module 602, a quantity determination module 603 and a push module 604. The first determining module 601 is configured to determine, according to a first association between the user characteristic and the active time, whether the push time of the push list belongs to the active time of the user. The second determining module 602 is configured to determine, according to the user characteristics, the click frequency corresponding to the push time in response to the push time belonging to the active time of the user. The number determination module 603 is configured to determine the number of pushes according to the click frequency. The pushing module 604 is configured to push messages in the push list according to the number of pushes.

In some embodiments of the present disclosure, the first association further indicates an association between the user characteristic, the content characteristic of the message, and the active time, the first decision module 601 is further configured to: and judging whether the push time belongs to the active time of the user for the content features according to the user features, the content features of the messages in the push list and the first association relation.

In some embodiments of the present disclosure, the first decision module 601 is further configured to: and inputting the user characteristics into a first judging model to judge whether the push time belongs to the active time of the user, wherein the first judging model is used for representing a first association relationship and is learned from historical behavior data of a plurality of historical users.

In some embodiments of the present disclosure, the second determination module is further configured to: and inputting the user characteristics and the time characteristics corresponding to the pushing time into a second judging model, and determining the click frequency corresponding to the pushing time, wherein the second judging model is used for representing a second association relationship among the user characteristics, the time characteristics and the click frequency and is obtained by learning historical behavior data of a plurality of historical users.

In some embodiments of the present disclosure, the quantity determination module 603 includes: a threshold determination submodule configured to determine a push threshold of the user; the quantity determination submodule is configured to determine the pushing quantity according to the pushing threshold value and the click frequency.

In some embodiments of the present disclosure, the quantity determination submodule includes: a product unit configured to calculate a product of the push threshold and the click frequency as a reference push number; the quantity determining unit is configured to determine the pushing quantity according to the reference pushing quantity, the counted pushing total quantity and the maximum pushing quantity.

In some embodiments of the present disclosure, the apparatus further comprises: and the updating module is configured to respond to the active time of the user, update the push time according to the active time of the user and execute the step of determining the click frequency corresponding to the push time according to the user characteristics.

Fig. 7 is a schematic block diagram of a model training apparatus according to a sixth embodiment of the present disclosure. As shown in fig. 7, the model training apparatus 700 may include: an extraction module 701 and a training module 702. Wherein the extraction module 701 is configured to extract historical feature data from the historical behavior data, the historical feature data including a historical user feature, a historical push quantity feature, and a historical click time feature. The training module 702 is configured to train the content push model using the historical user characteristics, the historical push quantity characteristics, and the historical click time characteristics. The trained content push model at least characterizes a first association relationship and a second association relationship, the first association relationship at least indicates an association relationship between user features and active time, and the second association relationship indicates an association relationship between user features, time features and click frequency.

In some embodiments of the present disclosure, training module 702 includes: the first training sub-module is configured to train the first learning network by using the historical user characteristics and the historical click time characteristics to obtain a first judging model in the content pushing model, wherein the first judging model is used for representing a first association relation; the second training sub-module is configured to train a second learning network by using the historical user characteristics, the historical pushing quantity characteristics and the historical click time characteristics to obtain a second judging model in the content pushing model, wherein the second judging model is used for representing a second association relation.

In some embodiments of the present disclosure, the historical feature data further includes historical content features of the message of the historical click; the first training submodule is further configured to: training a first learning network by using the historical user characteristics, the historical click time characteristics and the historical content characteristics to obtain a first judging model, wherein the first association relationship represented by the first judging model also indicates the association relationship among the user characteristics, the content characteristics of the message and the active time.

In some embodiments of the present disclosure, the extraction module 701 is further configured to: and responding to the historical behavior data, wherein the historical behavior data comprises a specified type of behavior log, generating historical characteristic data according to the recording time of the behavior log, the specified type of behavior log indicates that a historical user closes a pushing function or unloads an application, and the generated historical characteristic data indicates that the historical click frequency characteristic of the historical user characteristic corresponding to the historical user in a time period corresponding to the recording time is a specified numerical value.

It is to be noted that this embodiment is an implementation of the apparatus corresponding to the above-described method embodiment, and this embodiment may be implemented in cooperation with the above-described method embodiment. The related technical details mentioned in the above method embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the related technical details mentioned in the present embodiment can also be applied in the above-described method embodiments.

It should be noted that, each module involved in this embodiment is a logic module, and in practical application, one logic unit may be one physical unit, or may be a part of one physical unit, or may be implemented by a combination of multiple physical units. Furthermore, in order to highlight the innovative part of the present disclosure, elements that are not so close to solving the technical problem presented by the present disclosure are not introduced in the present embodiment, but it does not indicate that other elements are not present in the present embodiment.

According to a seventh embodiment of the present disclosure, the present disclosure also provides an electronic device comprising at least one processor and a memory. Wherein the memory is communicatively coupled to the at least one processor and stores instructions executable by the at least one processor to enable the at least one processor to perform the methods recited in the embodiments.

According to an eighth embodiment of the present disclosure, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as mentioned in the above embodiments.

According to a ninth embodiment of the present disclosure, the present disclosure also provides a computer program product comprising a computer program which, when executed by a processor, implements the method according to the above-mentioned embodiments.

FIG. 8 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 8, the device 800 includes a processor 801 that may perform various suitable actions and processes in accordance with a computer program stored in a Read Only Memory (ROM) 802 or a computer program loaded from a memory 808 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The processor 801, the ROM 802, and the RAM 803 are connected to each other by a bus 804. An input/output (I/O) interface 805 is also connected to the bus 804.

Various components in device 800 are connected to I/O interface 805, including: an input unit 806 such as a keyboard, mouse, etc.; an output unit 807 such as various types of displays, speakers, and the like; memory 808, such as a magnetic disk, optical disk, etc.; and a communication unit 809, such as a network card, modem, wireless communication transceiver, or the like. The communication unit 809 allows the device 800 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

The processor 801 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of processor 801 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, digital Signal Processors (DSPs), and any suitable processor, controller, microcontroller, etc. The processor 801 performs the various methods and processes described above, such as a message pushing method or a model training method. For example, in some embodiments, the message pushing method or model training method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as memory 808. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 800 via ROM 802 and/or communication unit 809. When the computer program is loaded into RAM 803 and executed by processor 801, one or more steps of the message pushing method or model training method described above may be performed. Alternatively, in other embodiments, the processor 801 may be configured to perform the message pushing method or the model training method in any other suitable manner (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel or sequentially or in a different order, provided that the desired results of the technical solutions of the present disclosure are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims (25)

1. A message pushing method, comprising:

judging whether the push time of the push list belongs to the active time of the user according to a first association relation between the user characteristics and the active time;

responding to the push time belonging to the active time of the user, and determining click frequency corresponding to the push time according to the user characteristics;

determining the pushing quantity according to the click frequency;

And pushing the messages in the pushing list according to the pushing quantity.

2. The method of claim 1, wherein the first association further indicates an association between a user characteristic, a content characteristic of a message, and an active time;

judging whether the push time of the push list belongs to the active time of the user according to the first association relation between the user characteristics and the active time, comprising:

and judging whether the push time belongs to the active time of the user for the content characteristics according to the user characteristics, the content characteristics of the messages in the push list and the first association relation.

3. The method of claim 1, wherein the determining whether the push time of the push list belongs to the active time of the user according to the first association between the user characteristic and the active time comprises:

and inputting the user characteristics into a first judging model to judge whether the push time belongs to the active time of the user, wherein the first judging model is used for representing the first association relationship and is learned from historical behavior data of a plurality of historical users.

4. A method according to any one of claims 1 to 3, wherein said determining, according to the user characteristics, the click frequency corresponding to the push time comprises:

And inputting the user characteristics and the time characteristics corresponding to the pushing time into a second judging model, and determining the click frequency corresponding to the pushing time, wherein the second judging model is used for representing a second association relationship among the user characteristics, the time characteristics and the click frequency and is obtained by learning historical behavior data of a plurality of historical users.

5. The method according to any one of claims 1 to 4, wherein the determining the number of pushes according to the click frequency comprises:

determining a push threshold of the user;

and determining the pushing quantity according to the pushing threshold and the click frequency.

6. The method of claim 5, wherein the determining the number of pushes based on the push threshold and the click frequency comprises:

calculating the product of the push threshold and the click frequency as a reference push quantity;

and determining the pushing quantity according to the reference pushing quantity, the counted pushing total quantity and the maximum pushing quantity.

7. The method of any one of claims 1 to 6, further comprising:

and responding to the push time which does not belong to the active time of the user, updating the push time according to the active time of the user, and executing the step of determining the click frequency corresponding to the push time according to the user characteristics.

8. A model training method, comprising:

extracting historical characteristic data from historical behavior data, wherein the historical characteristic data comprises historical user characteristics, historical push quantity characteristics and historical click time characteristics;

training a content push model using the historical user characteristics, the historical push quantity characteristics, and the historical click time characteristics;

the trained content push model at least characterizes a first association relationship and a second association relationship, the first association relationship at least indicates an association relationship between user features and active time, and the second association relationship indicates an association relationship among user features, time features and click frequency.

9. The method of claim 8, wherein the training a content push model using the historical user feature, the historical push quantity feature, and the historical click time feature comprises:

training a first learning network by using the historical user characteristics and the historical click time characteristics to obtain a first judging model in the content pushing model, wherein the first judging model is used for representing the first association relation;

and training a second learning network by using the historical user characteristics, the historical pushing quantity characteristics and the historical click time characteristics to obtain a second judging model in the content pushing model, wherein the second judging model is used for representing the second association relation.

10. The method of claim 9, wherein the historical feature data further comprises historical content features of messages of historical clicks;

training a first learning network by using the historical user characteristics and the historical click time characteristics to obtain a first judging model in the content pushing model, wherein the first judging model comprises the following steps:

training the first learning network by using the historical user characteristics, the historical click time characteristics and the historical content characteristics to obtain the first judging model, wherein the first association relationship represented by the first judging model also indicates the association relationship among the user characteristics, the content characteristics of the message and the active time.

11. The method of claim 8, wherein the extracting historical feature data from historical behavioral data comprises:

and responding to the historical behavior data, wherein the historical behavior data comprises a behavior log of a specified type, the historical characteristic data is generated according to the recording time of the behavior log, the specified type of behavior log indicates that a historical user closes a pushing function or the historical user uninstalls an application, and the generated historical characteristic data indicates that the historical click frequency characteristic of the historical user characteristic corresponding to the historical user in the time period corresponding to the recording time is a specified numerical value.

12. A message pushing device, comprising:

the first judging module is configured to judge whether the push time of the push list belongs to the active time of the user according to a first association relation between the user characteristics and the active time;

the second judging module is configured to respond to the push time belonging to the active time of the user, and determine the click frequency corresponding to the push time according to the user characteristics;

the quantity determining module is configured to determine the pushing quantity according to the click frequency;

and the pushing module is configured to push the messages in the pushing list according to the pushing quantity.

13. The apparatus of claim 12, wherein the first association further indicates an association between a user characteristic, a content characteristic of a message, and an active time, the first determination module being further configured to:

and judging whether the push time belongs to the active time of the user for the content characteristics according to the user characteristics, the content characteristics of the messages in the push list and the first association relation.

14. The apparatus of claim 12, wherein the first determination module is further configured to:

And inputting the user characteristics into a first judging model to judge whether the push time belongs to the active time of the user, wherein the first judging model is used for representing the first association relationship and is learned from historical behavior data of a plurality of historical users.

15. The apparatus of any of claims 13-14, wherein the second determination module is further configured to:

and inputting the user characteristics and the time characteristics corresponding to the pushing time into a second judging model, and determining the click frequency corresponding to the pushing time, wherein the second judging model is used for representing a second association relationship among the user characteristics, the time characteristics and the click frequency and is obtained by learning historical behavior data of a plurality of historical users.

16. The apparatus of any of claims 12 to 15, wherein the number determination module comprises:

a threshold determination submodule configured to determine a push threshold of the user;

and the quantity determination submodule is configured to determine the pushing quantity according to the pushing threshold value and the click frequency.

17. The apparatus of claim 16, wherein the quantity determination submodule comprises:

A product unit configured to calculate a product of the push threshold and the click frequency as a reference push number;

the quantity determining unit is configured to determine the pushing quantity according to the reference pushing quantity, the counted pushing total quantity and the maximum pushing quantity.

18. The apparatus of any of claims 12 to 17, further comprising:

and the updating module is configured to respond to the fact that the push time does not belong to the active time of the user, update the push time according to the active time of the user, and execute the step of determining the click frequency corresponding to the push time according to the user characteristics.

19. A model training apparatus comprising:

an extraction module configured to extract historical feature data from historical behavior data, the historical feature data comprising historical user features, historical push quantity features, and historical click time features;

a training module configured to train a content push model using the historical user characteristics, the historical push quantity characteristics, and the historical click time characteristics;

the trained content push model at least characterizes a first association relationship and a second association relationship, the first association relationship at least indicates an association relationship between user features and active time, and the second association relationship indicates an association relationship among user features, time features and click frequency.

20. The apparatus of claim 19, wherein the training module comprises:

the first training submodule is configured to train a first learning network by using the historical user characteristics and the historical click time characteristics to obtain a first judging model in the content pushing model, wherein the first judging model is used for representing the first association relation;

and the second training sub-module is configured to train a second learning network by using the historical user characteristics, the historical pushing quantity characteristics and the historical click time characteristics to obtain a second judging model in the content pushing model, wherein the second judging model is used for representing the second association relation.

21. The apparatus of claim 20, wherein the historical feature data further comprises historical content features of messages of historical clicks; the first training submodule is further configured to:

training the first learning network by using the historical user characteristics, the historical click time characteristics and the historical content characteristics to obtain the first judging model, wherein the first association relationship represented by the first judging model also indicates the association relationship among the user characteristics, the content characteristics of the message and the active time.

22. The apparatus of claim 19, wherein the extraction module is further configured to:

and responding to the historical behavior data, wherein the historical behavior data comprises a behavior log of a specified type, the historical characteristic data is generated according to the recording time of the behavior log, the specified type of behavior log indicates that a historical user closes a pushing function or the historical user uninstalls an application, and the generated historical characteristic data indicates that the historical click frequency characteristic of the historical user characteristic corresponding to the historical user in the time period corresponding to the recording time is a specified numerical value.

23. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

24. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-11.

25. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any of claims 1-11.