CN113902132B

CN113902132B - Training method of negative feedback behavior prediction model, message pushing method and equipment

Info

Publication number: CN113902132B
Application number: CN202111499434.7A
Authority: CN
Inventors: 刘睿智; 章昊; 孙式松
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2021-12-09
Filing date: 2021-12-09
Publication date: 2022-05-24
Anticipated expiration: 2041-12-09
Also published as: CN113902132A

Abstract

The disclosure relates to a training method of a negative feedback behavior prediction model, a message pushing method and equipment, wherein the method comprises the following steps: monitoring feedback information corresponding to the push message; determining a target candidate push message record closest to the current time from candidate push message records corresponding to a target sample terminal; the target sample terminal is a sample terminal which is continuously preset with a plurality of push messages and does not monitor corresponding feedback information; determining a positive sample message record for generating a positive sample from the target candidate push message records; determining a negative sample message record for generating a negative sample from the remaining candidate recommendation message records; training a preset machine learning model according to the positive sample and the negative sample to obtain a negative feedback behavior prediction model; the negative feedback behavior refers to a behavior of closing the message push notification after receiving the push message. The method and the device improve the accuracy of the training sample aiming at the negative feedback behavior and improve the accuracy of the model after training for predicting the negative feedback behavior.

Description

Negative feedback behavior prediction model training method, message pushing method and equipment

Technical Field

The disclosure relates to the technical field of computers, in particular to a negative feedback behavior prediction model training method, a message pushing method and equipment.

Background

At present, after a terminal receives a push message corresponding to an application program, the terminal may select to close the message push notification of the application program, and once the message push notification is closed, the application program cannot transmit information of the application program to the terminal through a message push system corresponding to the application program, and cannot pull a user account corresponding to the terminal; in addition, the message push notification of the terminal closing the application program also indicates that the experience effect of the application program is poor. Therefore, it is very important to accurately predict the negative feedback behavior of the terminal for the message to be pushed, where the negative feedback behavior refers to the behavior of turning off the message push notification.

In the related art, when a negative feedback behavior prediction model of a recommendation system is trained, a training sample is determined by directly acquiring a negative feedback signal, and the training sample is determined by directly clicking a button which is not interested. However, in a message pushing scenario, the terminal closing message pushing notification is operated on the terminal operating system, as shown in fig. 1, the message pushing notification may be closed by directly operating a system switch on the received pushing message, or may be closed in the setting of the operating system, so that a background cannot acquire a signal of the terminal closing message pushing notification in real time, and further cannot accurately determine a training sample for a negative feedback behavior, thereby reducing a training effect of a negative feedback behavior prediction model, resulting in inaccurate negative feedback behavior prediction in the message pushing scenario and a high closing rate of the message pushing notification.

Disclosure of Invention

The invention provides a negative feedback behavior prediction model training method and device and a message pushing method and device, which are used for at least solving the problems that the training effect of a negative feedback behavior prediction model is poor due to the fact that a training sample aiming at a negative feedback behavior cannot be accurately determined in the related technology, and therefore the negative feedback behavior prediction in a message pushing scene is inaccurate and the closing rate of a message pushing notice is high. The technical scheme of the disclosure is as follows:

according to a first aspect of the embodiments of the present disclosure, there is provided a negative feedback behavior prediction model training method, including:

monitoring feedback information corresponding to the push message; the feedback information is sent after the sample terminal receives and displays the push message;

determining a target candidate push message record closest to the current time from candidate push message records corresponding to a target sample terminal; the target sample terminal is a sample terminal in which a preset number of push messages are continuously monitored without corresponding feedback information, and the candidate push message record refers to a push message record corresponding to the push message in which the feedback information is monitored;

determining a positive sample message record from the target candidate push message records, and generating a positive sample according to the positive sample message record;

determining negative sample message records from the remaining candidate recommendation message records, and generating negative samples according to the negative sample message records; the remaining candidate recommended message records are candidate recommended message records except the target candidate pushed message record;

training a preset machine learning model according to the positive sample and the negative sample to obtain a negative feedback behavior prediction model; the negative feedback behavior refers to a behavior of closing the message push notification after receiving the push message.

In an exemplary embodiment, the determining a positive sample message record from the target candidate push message records includes:

determining the target candidate push message record meeting the preset condition as a positive sample message record; the preset condition comprises at least one of the following conditions:

the number of candidate push message records in a first historical time period of a sample terminal corresponding to the target candidate push message record exceeds a first preset number threshold, wherein the first historical time period is a first preset time period before the current time;

sample terminals corresponding to the historical positive sample message records do not contain the sample terminals corresponding to the target candidate push message records;

and the number of communication connection errors of the sample terminal corresponding to the target candidate push message record does not exceed a preset number threshold.

In an exemplary embodiment, the generating a positive sample from the positive sample message record includes:

determining a first negative feedback behavior related characteristic according to the positive sample message record;

generating the positive sample according to the first negative feedback behavior related characteristic;

wherein the first negative feedback behavior-related feature comprises at least one of: the push message type and the push time of the push message in the positive sample message record, a historical push message list of the sample terminal corresponding to the positive sample message record, and the number of push messages of the sample terminal corresponding to the positive sample message record in a preset time period;

the generating negative examples from the negative example message records comprises:

determining a second negative feedback behavior related characteristic according to the negative sample message record;

generating the negative sample according to the second negative feedback behavior related characteristic;

wherein the second negative feedback behavior-related characteristic comprises at least one of: the push message type and the push time of the push message in the negative sample message record, the historical push message list of the corresponding sample terminal of the negative sample message record, and the push message quantity of the corresponding sample terminal in the preset time period of the negative sample message record.

In an exemplary embodiment, the method further comprises:

responding to an event of sending the push message to the sample terminal, and increasing the count of a counter corresponding to the sample terminal according to a preset numerical value increment;

and when the feedback information corresponding to the push message is monitored, resetting the count of a counter of the sample terminal corresponding to the push message, and generating a candidate push message record of the sample terminal according to the push message corresponding to the feedback information.

In an exemplary embodiment, the method further comprises:

obtaining the current count of a counter corresponding to the sample terminal;

and determining the sample terminal as the target sample terminal when the current count exceeds the preset number.

According to a second aspect of the embodiments of the present disclosure, there is provided a message pushing method, including:

acquiring a message to be pushed aiming at a message receiving terminal;

determining negative feedback behavior related characteristics corresponding to the message receiving terminal according to the message to be pushed; the negative feedback behavior is a behavior of closing the message push notification after receiving the push message;

inputting the relevant characteristics of the negative feedback behavior corresponding to the message receiving terminal into a negative feedback behavior prediction model to perform negative feedback behavior prediction to obtain a negative feedback behavior index; the negative feedback behavior index represents the probability that the message receiving terminal closes the message push notification after receiving the message to be pushed;

and sending the message to be pushed to the message receiving terminal according to the comparison condition of the negative feedback behavior index and a preset index threshold.

In an exemplary embodiment, the sending the message to be pushed to the message receiving terminal according to the comparison between the negative feedback behavior index and a preset index threshold includes:

sending the message to be pushed to the message receiving terminal under the condition that the negative feedback behavior index does not exceed the preset index threshold;

wherein the negative feedback behavior prediction model is obtained by training according to the training method of the negative feedback behavior prediction model in the first aspect.

In an exemplary embodiment, the method further comprises:

determining the number of push messages sent to the message receiving terminal within a second historical time period under the condition that the negative feedback behavior index exceeds the preset index threshold; the second historical time period is a second preset time period before the current time;

and when the number does not exceed a second preset number threshold value, sending the message to be pushed to the message receiving terminal.

In an exemplary embodiment, before sending the message to be pushed to the message receiving terminal, the method further includes:

determining the sending time of a push message sent to the message receiving terminal last time;

determining a time difference between the current time and the transmission time;

and under the condition that the time difference value is not smaller than a preset time difference threshold value, executing the step of sending the message to be pushed to the message receiving terminal.

In an exemplary embodiment, the determining, according to the message to be pushed, a negative feedback behavior related characteristic corresponding to the message receiving terminal includes:

determining a third negative feedback behavior related characteristic according to the message to be pushed;

determining negative feedback behavior related characteristics corresponding to the message receiving terminal according to the third negative feedback behavior related characteristics;

wherein the third negative feedback behavior-related feature comprises at least one of: the message pushing method comprises the steps of pushing a message type corresponding to the message to be pushed, the current time, a historical pushing message list corresponding to the message receiving terminal and the quantity of pushing messages of the message receiving terminal in a preset time period.

According to a third aspect of the embodiments of the present disclosure, there is provided a negative feedback behavior prediction model training apparatus, including:

the monitoring unit is configured to monitor feedback information corresponding to the push message; the feedback information is sent after the sample terminal receives and displays the push message;

a target candidate push message record determining unit configured to perform determining a target candidate push message record closest to the current time from candidate push message records corresponding to the target sample terminal; the target sample terminal is a sample terminal in which a preset number of push messages are continuously monitored without corresponding feedback information, and the candidate push message record refers to a push message record corresponding to the push message in which the feedback information is monitored;

a positive sample determining unit configured to perform determining a positive sample message record from the target candidate push message records, and generate a positive sample according to the positive sample message record;

a negative sample determining unit configured to perform determining a negative sample message record from remaining candidate recommendation message records and generate a negative sample according to the negative sample message record; the remaining candidate recommended message records are candidate recommended message records except the target candidate pushed message record;

the training unit is configured to train a preset machine learning model according to the positive sample and the negative sample to obtain a negative feedback behavior prediction model; the negative feedback behavior refers to a behavior of closing the message push notification after receiving the push message.

In an exemplary embodiment, the positive sample determining unit is specifically configured to determine the target candidate push message record satisfying a preset condition as a positive sample message record; the preset condition comprises at least one of the following conditions:

In an exemplary embodiment, the positive sample determination unit includes:

a first feature determination unit configured to perform determining a first negative feedback behavior-related feature from the positive sample message record;

a positive sample generation subunit configured to perform generating the positive sample according to the first negative feedback behavior-related feature;

the negative sample generation unit includes:

a second feature determination unit configured to perform determining a second negative feedback behavior related feature from the negative example message record;

a negative sample generation subunit configured to perform generation of the negative sample in accordance with the second negative feedback behavior related feature;

wherein the second negative feedback behavior-related characteristic comprises at least one of: the push message type and the push time of the push message in the negative sample message record, the historical push message list of the sample terminal corresponding to the negative sample message record, and the push message quantity of the sample terminal corresponding to the negative sample message record in the preset time period.

In an exemplary embodiment, the apparatus further comprises:

a push message recording unit configured to perform incrementing a count of a counter corresponding to the sample terminal by a preset value increment in response to an event of sending the push message to the sample terminal;

the counting resetting unit is configured to reset the counting of the counter of the sample terminal corresponding to the push message when the feedback information corresponding to the push message is monitored, and generate a candidate push message record of the sample terminal according to the push message corresponding to the feedback information.

In an exemplary embodiment, the apparatus further comprises:

a current count acquisition unit configured to perform acquisition of a current count of a counter corresponding to the sample terminal;

a target sample termination determination unit configured to perform determining that the sample termination is the target sample termination if the current count exceeds the preset number.

According to a fourth aspect of the embodiments of the present disclosure, there is provided a message pushing apparatus, including:

a message to be pushed acquisition unit configured to perform acquisition of a message to be pushed for a message receiving terminal;

a negative feedback behavior related characteristic determining unit configured to determine a negative feedback behavior related characteristic corresponding to the message receiving terminal according to the message to be pushed; the negative feedback behavior is a behavior of closing the message push notification after receiving the push message;

the negative feedback behavior index prediction unit is configured to input the relevant characteristics of the negative feedback behavior corresponding to the message receiving terminal into a negative feedback behavior prediction model for negative feedback behavior prediction to obtain a negative feedback behavior index; the negative feedback behavior index represents the probability that the message receiving terminal closes the message push notification after receiving the message to be pushed;

a message to be pushed sending unit configured to execute sending of the message to be pushed to the message receiving terminal according to a comparison condition of the negative feedback behavior index and a preset index threshold;

wherein the negative feedback behavior prediction model is obtained by training according to the training method of the negative feedback behavior prediction model of the first aspect.

In an exemplary embodiment, the to-be-pushed message sending unit includes:

a first sending subunit, configured to send the message to be pushed to the message receiving terminal if the negative feedback behavior index does not exceed the preset index threshold.

In an exemplary embodiment, the to-be-pushed message sending unit further includes:

a push message quantity determination unit configured to perform determining a quantity of push messages sent to the message receiving terminal within a second historical period of time if the negative feedback behavior index exceeds the preset index threshold; the second historical time period is a second preset time period before the current time;

and the second sending subunit is configured to send the message to be pushed to the message receiving terminal when the number does not exceed a second preset number threshold.

a transmission time determination unit configured to perform determination of a transmission time of a push message transmitted to the message reception terminal most recently;

a time difference determination unit configured to perform determining a time difference value between the current time and the transmission time;

an execution unit configured to execute the step of sending the message to be pushed to the message receiving terminal if the time difference value is not smaller than a preset time difference threshold value.

In an exemplary embodiment, the negative feedback behavior related feature determining unit includes:

a third characteristic determining unit configured to perform determination of a third negative feedback behavior related characteristic according to the message to be pushed;

a negative feedback behavior related characteristic determining subunit configured to perform determining a negative feedback behavior related characteristic corresponding to the message receiving terminal according to the third negative feedback behavior related characteristic;

wherein the third negative feedback behavior-related feature comprises at least one of: the type of the push message corresponding to the message to be pushed, the current time, a historical push message list corresponding to the message receiving terminal, and the number of push messages of the message receiving terminal in a preset time period.

According to a fifth aspect of embodiments of the present disclosure, there is provided an electronic apparatus including:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the negative feedback behavior prediction model training method of the first aspect or the message pushing method of the second aspect.

According to a sixth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, wherein instructions of the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the negative feedback behavior prediction model training method of the first aspect or the message pushing method of the second aspect.

According to a seventh aspect of embodiments of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the negative feedback behavior prediction model training method of the first aspect or the message pushing method of the second aspect.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

in the scheme, because the candidate push message record is the push message record corresponding to the push message monitoring the feedback information, the target sample terminal is the sample terminal which does not monitor the corresponding feedback information and has a preset number of continuous push messages, the problem that a terminal closing message push notification signal cannot be obtained in real time under a message push scene is solved, and the training sample aiming at the negative feedback behavior can be accurately determined, the training effect of the negative feedback behavior prediction model is improved, the accuracy of negative feedback behavior prediction in a message push scene is ensured, and the closing rate of message push notification is reduced.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.

Fig. 1 is a schematic diagram of closing message push notification in the related art;

FIG. 2a is a schematic diagram of an application environment, shown in accordance with an exemplary embodiment;

fig. 2b is a diagram illustrating a terminal reporting a status of a message push notification switch according to an exemplary embodiment;

FIG. 3 is a flow diagram illustrating a method of negative feedback behavior prediction model training in accordance with an exemplary embodiment;

FIG. 4 illustrates an example of training a negative feedback behavior prediction model in accordance with an exemplary embodiment;

FIG. 5 is a flow diagram illustrating a method of pushing a message in accordance with an exemplary embodiment;

FIG. 6 is a flow chart illustrating another method of message pushing in accordance with an exemplary embodiment;

FIG. 7 is a flow chart illustrating another method of message pushing in accordance with an exemplary embodiment;

FIG. 8 is a block diagram illustrating a negative feedback behavior prediction model training apparatus in accordance with an exemplary embodiment;

FIG. 9 is a block diagram illustrating a message push device in accordance with an exemplary embodiment;

FIG. 10 is a block diagram illustrating an electronic device in accordance with an example embodiment.

Detailed Description

In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

Referring to fig. 2a, a schematic diagram of an application environment according to an exemplary embodiment is shown, where the application environment includes a terminal 210 and a server 220, and the terminal 210 and the server 220 can communicate through a network connection, and the network can be a wired network or a wireless network.

The terminal 210 may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, etc. The terminal 210 may have client software such as an Application (App) installed therein, and the Application may be a stand-alone Application or a sub-program in the Application. Illustratively, the applications may include video-type applications, information-type applications, live-broadcast applications, and the like. The user of the terminal 210 may log in to the application through pre-registered user information, which may include an account number and a password.

The server 220 may be a server for providing a background service for an application program in the terminal 210, and specifically, the service provided by the server 220 may be a message push service. The server 220 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, middleware service, a domain name service, a security service, a CDN, a big data and artificial intelligence platform, and the like.

In a specific application scenario, when pushing a message to the terminal 210, the server 220 may invoke a trained negative feedback behavior prediction model to predict a negative feedback behavior index of the terminal 210 for the message to be pushed, where the negative feedback behavior refers to a behavior of the terminal 210 closing a message push notification after receiving the message to be pushed, and the negative feedback behavior index refers to a probability of the terminal 210 closing the message push notification after receiving the message to be pushed, so as to determine whether to push the message to be pushed to the terminal 210 according to the negative feedback behavior index.

In the related art, when a negative feedback behavior prediction model of a recommendation system is trained, a training sample is determined by directly acquiring a negative feedback signal, and the training sample is determined by directly clicking a button which is not interested. However, in the message pushing scenario, the terminal closes the message pushing notification and operates on the terminal operating system, and the background cannot acquire a signal of the terminal closing the message pushing notification in real time, so that the training sample in the message pushing scenario cannot be accurately determined. Specifically, as shown in fig. 2b, the terminal reports the on-off state of the message push notification only when the APP of the application program in the terminal is active, if the terminal closes the message push notification in the period of inactivity of the APP, because the APP is in the period of inactivity, the terminal cannot report the closed state of the message push notification to the background server in real time, and only when the APP is active again, the terminal cannot report the closed state of the message push notification to the background server, and at this time, the background cannot determine at what time the terminal is specifically and after which push message is received, and then the message push notification is closed, so that a sample for training the negative feedback behavior prediction model cannot be accurately obtained, the training effect of the negative feedback behavior prediction model is reduced, leading to inaccurate negative feedback behavior prediction in the message push scene and a high closing rate of the message push notification, the negative feedback behavior refers to a behavior of closing the message push notification after receiving the push message.

In view of this, the embodiment of the present disclosure provides a training method for a negative feedback behavior prediction model, which solves the problem that a terminal shutdown message push notification signal cannot be obtained in real time in a message push scenario, so that a training sample for a negative feedback behavior can be accurately determined, a training effect of the negative feedback behavior prediction model is improved, accuracy of negative feedback behavior prediction in the message push scenario is ensured, and reduction of a shutdown rate of message push notification is facilitated.

FIG. 3 is a flowchart illustrating a method for training a negative feedback behavior prediction model according to an exemplary embodiment, as shown in FIG. 3, for example, when the method is applied to the server in FIG. 2a, the method includes the following steps:

in step S301, feedback information corresponding to the push message is monitored.

The feedback information is sent after the sample terminal receives and displays the push message, and the display of the push message by the sample terminal means that the push message is displayed in a message form.

In a specific implementation, the server may monitor any sample terminal of the plurality of sample terminals to monitor feedback information corresponding to the push message sent to any sample terminal.

In practical application, after the push message is sent to the sample terminal by the server, when the sample terminal receives the push message and displays the push message, the push message is fed back to the server as a receive signal, and the receive signal is feedback information corresponding to the push message. If the sample terminal closes the message push notification, the sample terminal does not feed back an arrival (receive) signal to the server because the push message received by the sample terminal cannot be displayed.

In step S303, a target candidate push message record closest to the current time is determined from the candidate push message records corresponding to the target sample terminal.

The target sample terminal is a sample terminal in which a preset number of push messages are continuously sent without monitoring corresponding feedback information, the target sample terminal may be one sample terminal or a plurality of sample terminals, and the preset number may be set according to actual experience, for example, may be 5.

The candidate push message record refers to a push message record corresponding to the push message in which the feedback information is monitored. Specifically, when monitoring feedback information returned by the sample terminal for the push message, the server indicates that the push message corresponding to the feedback information has been received by the sample terminal and is normally displayed, and at this time, the server may generate a candidate message push record corresponding to the sample terminal, where the candidate message push record may include, but is not limited to, a push message type and a push time of the normally displayed push message.

In an exemplary embodiment, in order to accurately generate a candidate push message record corresponding to a sample terminal, the server may configure a corresponding counter for each sample terminal, so that when monitoring feedback information corresponding to a push message, the server may increase, according to a preset value increment, a count of the counter corresponding to the sample terminal in response to each event of sending the push message to the sample terminal; and when the feedback information corresponding to the push message is monitored, resetting the count of a counter of the sample terminal corresponding to the push message, and generating a candidate push message record of the sample terminal according to the push message corresponding to the feedback information. The preset value increment may be 1, that is, each time the server sends a push message to the sample terminal, the count of the counter corresponding to the sample terminal is incremented by 1, and when the feedback information corresponding to the push message is monitored, the count of the counter corresponding to the sample terminal corresponding to the push message is reset to zero.

In an exemplary embodiment, in order to accurately and quickly find a target sample terminal from a plurality of sample terminals to improve the training efficiency of the model, the server may obtain a current count of a counter corresponding to the sample terminal, and determine the sample terminal as the target sample terminal when the current count exceeds the preset number, and further may perform step S303 to determine a target candidate push message record closest to the current time from candidate push message records corresponding to the target sample terminal.

In step S305, a positive sample message record is determined from the target candidate push message records, and a positive sample is generated according to the positive sample message record.

Specifically, if the server sends push messages to a certain sample terminal all the time, and the sample terminal also has a feedback arrival signal (i.e., feedback information) to the server, then when a preset number of consecutive push messages starting from a certain push message and the sample terminal does not feed back the arrival signal any more, it may be determined that the sample terminal has closed the message push notification after the last successful push message (i.e., has monitored the feedback information), and the last successful push message (i.e., has monitored the feedback information) may be determined to be the push message that results in the negative feedback behavior of closing the message push notification, and a record of the push message corresponding to the last successful push message may be used as a record of the positive sample message for the negative feedback behavior.

In one particular embodiment, the positive sample message record may be one or more randomly selected ones of the target candidate push message records.

Considering the problem of noise interference in practical application, for example, the sample terminal may have turned off the message push notification, but the sample terminal may mistakenly return a piece of feedback information due to an error in processing of the sample terminal; or, although the server always sends the push message to the sample terminal, the server does not receive the feedback information returned by the sample terminal; or, the sample terminal already has a corresponding positive sample message record; or the sample terminal closes the message push notification as early as before the application program is online; or because there is an error (for example, the token detection fails) in the communication connection between the sample terminal and the server, although the sample terminal does not close the message push notification, the push message sent by the server to the sample terminal cannot receive the corresponding feedback information, and so on.

To avoid the influence of noise interference on the accuracy of the training sample determination, in an exemplary embodiment, in step S305, when determining a positive sample message record from the target candidate push message records, the target candidate push message record meeting a preset condition may be determined as the positive sample message record, where the preset condition includes at least one of:

Specifically, the first preset time period may be set according to actual needs, and may be 48 hours, for example. The preset number threshold may be set according to actual needs, and may be, for example, 10 times. The historical positive sample message record refers to a positive sample message record determined before the current time, and in practical applications, the positive sample message record determined in a preset time period (for example, three months, five months, and the like) before the current time.

According to the embodiment of the invention, the noise interference is filtered through the preset conditions, so that the accuracy of the determination of the training sample is ensured, the high-quality training sample can be obtained, and the prediction accuracy of the negative feedback behavior prediction model is further improved.

In an exemplary embodiment, the step S305 when generating the positive sample according to the positive sample message record may include:

wherein the first negative feedback behavior-related feature comprises at least one of: the push message type and the push time of the push message in the positive sample message record, the historical push message list of the sample terminal corresponding to the positive sample message record, and the push message quantity of the sample terminal corresponding to the positive sample message record in a preset time period. The preset time period may be a recent time period, such as 2 hours, 4 hours, or the number of push messages sent during the day.

It can be understood that, in order to improve the accuracy of model training, the positive sample may include some other features besides the first negative feedback behavior related feature, for example, a sample terminal feature, a content feature, and a cross feature corresponding to the positive sample message record, where the sample terminal feature may include a corresponding user account representation (e.g., user account identifier, age information, gender information, region information, etc.), historical behavior information (e.g., historical click rate for push message, content list of historical clicks, push message list of historical clicks, etc.); the content characteristics may include content type (e.g., video, live, etc.), content identification, identification of the content publisher, etc.; the cross feature may include a cross feature of a content publisher dimension and a cross feature of a content type dimension, where the cross feature of the content publisher dimension may be determined according to a clicked content publisher list and a publisher corresponding to the current push message, and in a specific implementation, the cross feature of the content publisher dimension may be respectively converted into vector representations, and then a product between vectors is calculated to obtain the cross feature of the content publisher dimension.

In the above embodiment, the first negative feedback behavior related characteristic is a characteristic closely related to the negative feedback behavior, so that the positive sample generated based on the positive sample message record can more accurately represent the negative feedback behavior, which is beneficial to improving the training effect and improving the accuracy of the model after training for predicting the negative feedback behavior.

In step S307, a negative sample message record is determined from the remaining candidate recommendation message records, and a negative sample is generated according to the negative sample message record.

Wherein the remaining candidate recommended message records refer to candidate recommended message records other than the target candidate pushed message record.

In an exemplary embodiment, when generating a negative example according to the negative example message record, the method may include:

wherein the second negative feedback behavior related characteristic comprises at least one of: the push message type and the push time of the push message in the negative sample message record, the historical push message list of the sample terminal corresponding to the negative sample message record, and the push message quantity of the sample terminal corresponding to the negative sample message record in the preset time period. The preset time period may be a recent time period, such as 2 hours, 4 hours, or the number of push messages sent during the current day.

It can be understood that, in order to improve the accuracy of model training, the negative sample may include some other features besides the second negative feedback behavior related feature, for example, a sample terminal feature, a content feature, and a cross feature corresponding to the negative sample message record, which may be referred to in the foregoing description about the sample terminal feature, the content feature, and the cross feature, and is not described herein again.

In the above embodiment, the features in the negative sample correspond to the positive sample, which is beneficial to improving the training effect of the model and improving the accuracy of the model after training for negative feedback behavior prediction.

In step S309, a preset machine learning model is trained according to the positive sample and the negative sample, so as to obtain a negative feedback behavior prediction model.

The negative feedback behavior is a behavior of closing the message push notification after receiving the push message, the negative feedback behavior prediction model can be used for predicting a negative feedback behavior index, and the negative feedback behavior index represents the probability of closing the message push notification after the message receiving terminal receives the message to be pushed.

The preset machine learning model may be a classified machine learning model, or may be a Deep learning model, such as Deep Neural Networks (DNNs) model shown in fig. 4. The aim of model training is to correctly distinguish positive samples from negative samples in training samples, and by inputting a large number of positive sample characteristics and negative sample characteristics into a preset machine learning model, the model can update model parameters through back propagation, so that the model can distinguish the positive samples from the negative samples. In a specific implementation, the capability of the model to distinguish between positive and negative samples can be judged by the size of a loss function, which may be a logarithmic loss function as shown in the following formula:

wherein, y_iFor inputting the true class of sample features, p_iProbability, y, of a predicted input sample feature belonging to class 1_i= 1 as positive sample feature, y_i= 0 is expressed as a negative sample characteristic, and N is the input sample amount.

Specifically, the greater the prediction probability of the model for the positive sample is, the smaller the prediction probability for the negative sample is, the smaller the loss function of the model is, the better the training effect of the representative model is, and the more the model can distinguish the positive sample from the negative sample, so that the preset training end condition may be that the loss value of the loss function reaches the minimum value. Of course, the preset training end condition may also be that the iteration number reaches a preset iteration number threshold, and the preset iteration number threshold may be set according to actual needs, for example, 100 times.

It can be understood that the training method of the negative feedback behavior prediction model according to the embodiment of the present disclosure may be an offline training negative feedback behavior prediction model, or may be an online updating training method performed on the negative feedback behavior prediction model deployed on the online, and the negative feedback behavior prediction model after the online updating training may also be used for online real-time prediction.

In an exemplary embodiment, after training to obtain the negative feedback behavior prediction model, the method may further include:

acquiring a message to be pushed aiming at a message receiving terminal;

determining negative feedback behavior related characteristics corresponding to the message receiving terminal according to the message to be pushed;

According to the embodiment of the invention, the preset machine learning model is trained by using the accurately determined training sample aiming at the negative feedback behavior to obtain the negative feedback behavior prediction model, so that the prediction accuracy of the negative feedback behavior prediction model is ensured, and the accuracy of the negative feedback behavior prediction in a message pushing scene is further improved.

Referring to fig. 5, it shows a flowchart of a message pushing method provided in an embodiment of the present disclosure, and as shown in fig. 5, the method includes:

in step S501, a message to be pushed for a message receiving terminal is acquired.

In step S503, according to the message to be pushed, determining a negative feedback behavior related characteristic corresponding to the message receiving terminal.

In an exemplary embodiment, determining, according to the message to be pushed, a negative feedback behavior related characteristic corresponding to the message receiving terminal may include:

wherein the third negative feedback behavior-related characteristic comprises at least one of: the message pushing method comprises the steps of pushing a message type corresponding to the message to be pushed, the current time, a historical pushing message list corresponding to the message receiving terminal and the quantity of pushing messages of the message receiving terminal in a preset time period.

In practical applications, in order to improve the accuracy of the prediction, the negative feedback behavior related characteristic corresponding to the message receiving terminal may include, in addition to the third negative feedback behavior related characteristic, other characteristics, such as a message receiving terminal characteristic, a content characteristic, and a cross characteristic, where the message receiving terminal characteristic may include a corresponding user account representation (e.g., a user account identifier, age information, gender information, region information, etc.), and historical behavior information (e.g., a historical click rate for historical push messages, a content list for historical clicks, a push message list for historical clicks, etc.). The content characteristics may include a content type (e.g., video, live, etc.) of the content for which the message is to be pushed, a content identification, an identification of a content publisher, and so on; the cross feature may include a cross feature of a content publisher dimension and a cross feature of a content type dimension, where the cross feature of the content publisher dimension may be determined according to a content publisher list clicked by a message receiving terminal and a publisher corresponding to a message to be pushed, and in a specific implementation, the cross feature of the content publisher dimension may be respectively converted into vector representations, and then a product between vectors is calculated to obtain the cross feature of the content publisher dimension.

In the above embodiment, since the third negative feedback behavior-related feature is a feature closely related to the negative feedback behavior, the message to be pushed, which is sent to the message terminal, is fully expressed in the negative feedback behavior, which is beneficial to improving the accuracy of the subsequent negative feedback behavior prediction.

In step S505, the relevant characteristics of the negative feedback behavior corresponding to the message receiving terminal are input to a negative feedback behavior prediction model to perform negative feedback behavior prediction, so as to obtain a negative feedback behavior index.

And the negative feedback behavior index represents the probability of closing the message push notification after the message receiving terminal receives the message to be pushed. The negative feedback behavior prediction model is obtained by training according to the training method of the negative feedback behavior prediction model in the embodiment of the present disclosure, and the training method can refer to the flowchart shown in fig. 3 in the embodiment of the present disclosure, which is not described herein again.

In step S507, the message to be pushed is sent to the message receiving terminal according to the comparison between the negative feedback behavior index and a preset index threshold.

The preset index threshold value can be set according to actual experience, and the smaller the preset index threshold value is, the more accurately the message to be pushed is issued.

When the message to be pushed is ready to be sent, the negative feedback behavior prediction model is called to accurately predict the negative feedback behavior index of the message receiving terminal aiming at the message to be pushed, and then the message to be pushed is sent to the message receiving terminal according to the comparison condition of the negative feedback behavior index and the preset index threshold, so that the closing rate of message pushing notification is favorably reduced, the message to be pushed can reach more terminals, and the daily active user quantity of message pushing pull-up is increased; meanwhile, the possibility that the message receiving terminal receives the push message which does not meet the self requirement is reduced, and the user experience is improved.

In an exemplary embodiment, when the step S507 sends the message to be pushed to the message receiving terminal according to a comparison between the negative feedback behavior index and a preset index threshold, as shown in fig. 6, the step S507 may include:

in step S601, it is determined whether the negative feedback behavior index exceeds a preset index threshold.

Specifically, if the determined result is that the negative feedback behavior index does not exceed the preset index threshold, it indicates that the negative feedback behavior probability of the message receiving terminal for the message to be pushed is relatively small, and the message receiving terminal does not close the message push notification after receiving the message to be pushed, at this time, step S603 may be executed; on the contrary, if the negative feedback behavior index exceeds the preset index threshold as a result of the determination, it indicates that the negative feedback behavior probability of the message receiving terminal for the message to be pushed is high, and the message receiving terminal is likely to close the message pushing notification after receiving the message to be pushed, and at this time, steps S605 to S607 may be executed.

In step S603, the message to be pushed is sent to the message receiving terminal.

In the above embodiment, the message to be pushed is sent to the target terminal only when the negative feedback behavior index does not exceed the preset index threshold, which is beneficial to reducing the closing rate of the message pushing notification, so that the pushing message can reach more terminals, and meanwhile, the possibility that the terminal receives the pushing message which does not meet the self requirement is also reduced, and the user experience is improved.

In step S605, the number of push messages sent to the message receiving terminal in the second history period is determined.

The second historical time period is a second preset time period before the current time, and the second preset time period may be set according to actual experience, for example, may be 3 hours.

In step S607, it is determined whether the number of push messages sent to the message receiving terminal in the second historical time period exceeds a second preset number threshold.

The second preset number threshold may be set according to practical experience, and may be 4 or 5, for example. Specifically, if the result of the determination is that the number of push messages sent to the message receiving terminal in the second historical time period does not exceed the second preset number threshold, step 609 may be executed to: sending the message to be pushed to a message receiving terminal; on the contrary, if the number of the push messages sent to the message receiving terminal in the second historical time period exceeds the second preset number threshold as a result of the determination, step S611 may be executed: the message to be pushed is not sent to the message receiving terminal.

According to the method and the device, under the condition that the negative feedback behavior index exceeds the preset index threshold value, the frequency control logic based on the number of the push messages sent to the message receiving terminal in the historical time period is used for controlling the messages to be pushed to be sent to the message receiving terminal, then only the messages to be pushed meeting the frequency control logic are sent to the message receiving terminal, the messages to be pushed which do not meet the frequency control logic are filtered, and therefore the closing rate of message push notification is reduced while the push messages can reach more terminals.

In order to improve the accuracy of the delivery control of the message to be pushed and ensure that the shutdown rate of the message push notification is reduced, in an exemplary embodiment, as shown in fig. 7, before step S609, the method may further include:

in step S701, a transmission time of a push message that was transmitted to the message receiving terminal most recently is determined.

In step S703, a time difference between the current time and the transmission time is determined.

In step S705, in a case that the time difference is not less than a preset time difference threshold, the step S609 is executed to send the message to be pushed to the message receiving terminal.

The preset time difference threshold may be set according to actual needs, and may be, for example, 10 minutes.

In the embodiment, the frequency control logic is further refined based on the difference between the sending time of the push message sent last time and the current time, so that multiple push messages are prevented from being sent to the message receiving terminal in a short time, the possibility that the terminal closes the message push notification is further reduced, the closing rate of the message push notification is further reduced, and the user experience is further improved.

FIG. 8 is a block diagram illustrating a negative feedback behavior prediction model training apparatus in accordance with an exemplary embodiment. Referring to fig. 8, the training apparatus 800 of negative feedback behavior prediction model comprises a listening unit 810, a target candidate push message record determining unit 820, a positive sample determining unit 830, a negative sample determining unit 840 and a training unit 850, wherein:

a monitoring unit 810 configured to perform monitoring feedback information corresponding to the push message; the feedback information is sent after the sample terminal receives and displays the push message;

a target candidate push message record determining unit 820 configured to determine a target candidate push message record closest to the current time from candidate push message records corresponding to the target sample terminal; the target sample terminal is a sample terminal in which a preset number of push messages are continuously monitored without corresponding feedback information, and the candidate push message record refers to a push message record corresponding to the push message in which the feedback information is monitored;

a positive sample determining unit 830 configured to perform determining a positive sample message record from the target candidate push message records, and generating a positive sample according to the positive sample message record;

a negative example determining unit 840 configured to perform determining a negative example message record from the remaining candidate recommendation message records and generating a negative example according to the negative example message record; the remaining candidate recommended message records are candidate recommended message records except the target candidate pushed message record;

a training unit 850 configured to perform training of a preset machine learning model according to the positive samples and the negative samples, resulting in a negative feedback behavior prediction model; the negative feedback behavior refers to a behavior of closing the message push notification after receiving the push message.

In an exemplary embodiment, the positive sample determining unit 830 is specifically configured to determine the target candidate push message record satisfying a preset condition as a positive sample message record; the preset condition comprises at least one of the following conditions:

In an exemplary embodiment, the positive sample determining unit 830 includes:

wherein the first negative feedback behavior-related feature comprises at least one of: the push message type and the push time of the push message in the positive sample message record, a historical push message list of the corresponding sample terminal of the positive sample message record, and the push message quantity of the corresponding sample terminal in a preset time period of the positive sample message record;

the negative sample generation unit 840 includes:

a negative sample generation subunit configured to perform generating the negative sample according to the second negative feedback behavior related characteristic;

In an exemplary embodiment, the apparatus further comprises:

and the counting resetting unit is configured to reset the counting of the counter of the sample terminal corresponding to the push message when the feedback information corresponding to the push message is monitored, and generate a candidate push message record of the sample terminal according to the push message corresponding to the feedback information.

In an exemplary embodiment, the apparatus further comprises:

Fig. 9 is a block diagram illustrating a message pushing apparatus according to an example embodiment. Referring to fig. 9, the message pushing apparatus 900 includes a to-be-pushed message obtaining unit 910, a negative feedback behavior related feature determining unit 920, a negative feedback behavior index predicting unit 930, and a to-be-pushed message sending unit 940, where:

a to-be-pushed message acquiring unit 910 configured to perform acquiring a to-be-pushed message for a message receiving terminal;

a negative feedback behavior related characteristic determining unit 920, configured to determine a negative feedback behavior related characteristic corresponding to the message receiving terminal according to the message to be pushed; the negative feedback behavior is a behavior of closing the message push notification after receiving the push message;

a negative feedback behavior index prediction unit 930 configured to perform negative feedback behavior prediction by inputting the relevant characteristics of the negative feedback behavior corresponding to the message receiving terminal to a negative feedback behavior prediction model, so as to obtain a negative feedback behavior index; the negative feedback behavior index represents the probability that the message receiving terminal closes the message push notification after receiving the message to be pushed;

a to-be-pushed message sending unit 940, configured to execute sending the to-be-pushed message to the message receiving terminal according to a comparison condition of the negative feedback behavior index and a preset index threshold;

the negative feedback behavior prediction model is obtained by training according to the training method of the negative feedback behavior prediction model provided by the embodiment of the disclosure.

In an exemplary embodiment, the to-be-pushed message sending unit 940 includes:

In an exemplary embodiment, the to-be-pushed message sending unit 940 further includes:

In an exemplary embodiment, the negative feedback behavior related feature determining unit 920 includes:

With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

In one exemplary embodiment, there is also provided an electronic device, comprising a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the negative feedback behavior prediction model training method or the message pushing method provided in any of the above embodiments when executing the instructions stored in the memory.

The electronic device may be a terminal, a server, or a similar computing device, taking the electronic device as a server as an example, fig. 10 is a block diagram of an electronic device shown according to an exemplary embodiment, and as shown in fig. 10, the server 1000 may generate a relatively large difference due to different configurations or performances, and may include one or more Central Processing Units (CPUs) 1010 (the processors 1010 may include but are not limited to Processing devices such as a microprocessor MCU or a programmable logic device FPGA), a memory 1030 for storing data, and one or more storage media 1020 (e.g., one or more mass storage devices) for storing application programs 1023 or data 1022. Memory 1030 and storage media 1020 may be, among other things, transient or persistent storage. The program stored in the storage medium 1020 may include one or more modules, each of which may include a series of instruction operations for a server. Still further, the central processor 1010 may be configured to communicate with the storage medium 1020 and execute a series of instruction operations in the storage medium 1020 on the server 1000. The server 1000 may also include one or more power supplies 1060, one or more wired or wireless network interfaces 1050, one or more input-output interfaces 1040, and/or one or more operating systems 1021, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, and so forth.

Input-output interface 1040 may be used to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the server 1000. In one example, i/o Interface 1040 includes a Network adapter (NIC) that may be coupled to other Network devices via a base station to communicate with the internet. In one example, the input/output interface 1040 may be a Radio Frequency (RF) module, which is used for communicating with the internet in a wireless manner.

It will be understood by those skilled in the art that the structure shown in fig. 10 is merely illustrative and is not intended to limit the structure of the electronic device. For example, server 1000 may also include more or fewer components than shown in FIG. 10, or have a different configuration than shown in FIG. 10.

In an exemplary embodiment, a computer-readable storage medium comprising instructions, such as the memory 1030 comprising instructions, executable by the processor 1010 of the apparatus 1000 to perform the method described above is also provided. Alternatively, the computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

In an exemplary embodiment, there is also provided a computer program product, including a computer program, which when executed by a processor implements the negative feedback behavior prediction model training method or the message pushing method provided in any of the above embodiments.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A negative feedback behavior prediction model training method is characterized by comprising the following steps:

monitoring feedback information corresponding to the push message; the feedback information is an arrival signal, and the arrival signal is returned when the sample terminal receives the push message and displays the push message;

2. The negative feedback behavior prediction model training method of claim 1, wherein the determining a positive sample message record from the target candidate push message records comprises:

3. The method of negative feedback behavior prediction model training of claim 1, wherein the generating positive samples from the positive sample message records comprises:

wherein the first negative feedback behavior related characteristic comprises at least one of: the push message type and the push time of the push message in the positive sample message record, a historical push message list of the sample terminal corresponding to the positive sample message record, and the number of push messages of the sample terminal corresponding to the positive sample message record in a preset time period;

4. The negative feedback behavior prediction model training method according to any one of claims 1 to 3, further comprising:

5. The negative feedback behavior prediction model training method of claim 4, further comprising:

obtaining the current count of a counter corresponding to the sample terminal;

6. A message pushing method, comprising:

acquiring a message to be pushed aiming at a message receiving terminal;

sending the message to be pushed to the message receiving terminal according to the comparison condition of the negative feedback behavior index and a preset index threshold;

the negative feedback behavior prediction model is obtained by training according to the training method of the negative feedback behavior prediction model of any one of claims 1-5.

7. The message pushing method according to claim 6, wherein the sending the message to be pushed to the message receiving terminal according to the comparison between the negative feedback behavior index and a preset index threshold comprises:

and sending the message to be pushed to the message receiving terminal under the condition that the negative feedback behavior index does not exceed the preset index threshold.

8. The message pushing method according to claim 7, wherein the method further comprises:

9. The message pushing method according to claim 7 or 8, wherein before sending the message to be pushed to the message receiving terminal, the method further comprises:

determining a transmission time of a push message transmitted to the message receiving terminal most recently;

10. The message pushing method according to claim 6, wherein the determining, according to the message to be pushed, the negative feedback behavior related characteristic corresponding to the message receiving terminal comprises:

11. A negative feedback behavior prediction model training apparatus, comprising:

the monitoring unit is configured to monitor feedback information corresponding to the push message; the feedback information is an arrival signal, and the arrival signal is returned when the sample terminal receives the push message and displays the push message;

a positive sample determining unit configured to perform determining a positive sample message record from the target candidate push message records and generate a positive sample according to the positive sample message record;

12. The negative feedback behavior prediction model training device according to claim 11, wherein the positive sample determining unit is specifically configured to perform determining the target candidate push message record satisfying a preset condition as a positive sample message record; the preset condition comprises at least one of the following conditions:

13. The negative feedback behavior prediction model training device according to claim 11, wherein the positive sample determination unit comprises:

a first characteristic determination unit configured to perform a determination of a first negative feedback behavior related characteristic from the positive sample message record;

the negative sample generation unit includes:

14. The negative feedback behavior prediction model training apparatus according to any one of claims 11 to 13, further comprising:

15. The negative feedback behavior prediction model training apparatus of claim 14, further comprising:

16. A message push apparatus, comprising:

the message sending unit to be pushed is configured to execute the comparison condition according to the negative feedback behavior index and a preset index threshold value and send the message to be pushed to the message receiving terminal;

17. The message pushing device as claimed in claim 16, wherein the message sending unit to be pushed comprises:

18. The message pushing device as claimed in claim 17, wherein the message sending unit to be pushed further comprises:

a push message quantity determining unit configured to perform determining a quantity of push messages sent to the message receiving terminal within a second historical period of time if the negative feedback behavior index exceeds the preset index threshold; the second historical time period is a second preset time period before the current time;

19. The message pushing apparatus according to claim 17 or 18, wherein the message sending unit to be pushed further comprises:

an execution unit configured to execute the step of sending the message to be pushed to the message receiving terminal if the time difference value is not less than a preset time difference threshold value.

20. The message pushing apparatus according to claim 16, wherein the negative feedback behavior related characteristic determining unit comprises:

21. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the negative feedback behavior prediction model training method of any one of claims 1 to 5 or the message pushing method of any one of claims 6 to 10.

22. A computer-readable storage medium, wherein instructions in the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the negative feedback behavior prediction model training method of any of claims 1 to 5, or the message pushing method of any of claims 6 to 10.