CN111079006A

CN111079006A - Message pushing method and device, electronic equipment and medium

Info

Publication number: CN111079006A
Application number: CN201911252578.5A
Authority: CN
Inventors: 刘卓
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-12-09
Filing date: 2019-12-09
Publication date: 2020-04-28
Anticipated expiration: 2039-12-09
Also published as: CN111079006B

Abstract

The invention discloses a message pushing method, a message pushing device, electronic equipment and a medium. The method comprises the following steps: acquiring a candidate message; acquiring state data of a target object; determining a click rate prediction model corresponding to the type of the candidate message, and inputting the state data into the click rate prediction model for click rate prediction; when the predicted click rate meets the requirement of a click rate threshold value, the candidate message is pushed to the client; the click rate prediction model is obtained by performing machine learning training on a plurality of state sample data, the state sample data carries the actual click rate of the same type of message, and the same type of message and the candidate message belong to the same type. And the click rate is predicted by combining the state data of the target object, so that the click rate of the message and the user experience can be improved. Click rate prediction is carried out on the dimensionality of the message type, click rates corresponding to different message types can be determined in a finer-grained manner, and the message pushing accuracy is improved.

Description

Message pushing method and device, electronic equipment and medium

Technical Field

The present invention relates to the field of internet communication technologies, and in particular, to a method and an apparatus for pushing a message, an electronic device, and a medium.

Background

With the rapid development of internet communication technology, networks have become an important way for people to acquire and share information. The server can push various messages to the client, and the user can obtain various messages through the client. For example, the server sends a video message to the client, the client displays the video message to the user, and the user can click on a video link included in the video message and watch the corresponding video by using the video player.

Click-through rate may refer to the ratio of the number of times a certain content on a web page is clicked on to the number of times it is displayed, i.e., clicks/views, which is a percentage. For a pushed message, the number of clicks of the message divided by the number of exposures of the message may be used as its click rate, which may reflect the pushing effect of the message.

In the prior art, a scheme of timing push is often adopted for message push, but the click rate obtained by the timing push is not ideal. Therefore, there is a need to provide a more efficient push scheme for messages.

Disclosure of Invention

In order to solve the problems that the click rate of the message is low when the prior art is applied to pushing the message, the invention provides a message pushing method, a device, electronic equipment and a medium, wherein the message pushing method comprises the following steps:

in one aspect, the present invention provides a message pushing method, where the method includes:

acquiring a candidate message;

acquiring state data of a target object;

determining a click rate prediction model corresponding to the type of the candidate message, and inputting the state data into the click rate prediction model for click rate prediction;

when the predicted click rate meets the requirement of a click rate threshold value, the candidate message is pushed to the client;

the click rate prediction model is obtained by performing machine learning training on a plurality of state sample data, the state sample data carries the actual click rate of the same type of message, and the same type of message and the candidate message belong to the same type.

Another aspect provides a message pushing apparatus, including:

a candidate message acquisition module: for obtaining candidate messages;

a status data acquisition module: the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring state data of a target object;

click rate prediction module: the click rate prediction model is used for determining the click rate prediction model corresponding to the type of the candidate message, and inputting the state data into the click rate prediction model for click rate prediction;

a pushing module: the candidate message is pushed to the client when the predicted click rate meets the requirement of the click rate threshold;

Another aspect provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a set of codes, or a set of instructions, which are loaded and executed by the processor to implement the message pushing method as described above.

Another aspect provides a computer-readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by a processor to implement the message push method as described above.

The message pushing method, the message pushing device, the electronic equipment and the medium have the following technical effects:

when the method is used for pushing the message, a click rate prediction model corresponding to the type of the candidate message is determined, the state data of the target object is input into the click rate prediction model for click rate prediction, and whether the message is pushed or not is determined according to the prediction result. The click rate is predicted by combining the state data of the target object, the current state and the historical state of the user are considered, and the click rate of the message and the user experience can be improved. Click rate prediction is carried out on the dimensionality of the message type, click rates corresponding to different message types can be determined in a finer-grained manner, and the message pushing accuracy is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a schematic diagram of an application environment provided by an embodiment of the invention;

fig. 2 is a schematic flowchart of a message pushing method according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of a model for training click rate prediction according to an embodiment of the present invention;

fig. 4 is a schematic flowchart of a process of pushing the candidate message to the client when the predicted click rate meets the requirement of the click rate threshold according to the embodiment of the present invention;

FIG. 5 is a diagram illustrating an application scenario of click rate prediction according to an embodiment of the present invention;

FIG. 6 is a UI diagram of a client applying the message pushing method provided by the embodiment of the invention;

fig. 7 is a comparison diagram of the effect of a message pushing method compared with a timing pushing method according to an embodiment of the present invention;

fig. 8 is a block diagram of a message pushing apparatus according to an embodiment of the present invention;

fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.

It should be noted that the terms "comprises" and "comprising," and any variations thereof, in the description and claims of the present invention and the above-described drawings, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Referring to fig. 1, fig. 1 is a schematic diagram of an application environment according to an embodiment of the present invention, which may include a client 01 and a server 02, where the client and the server are connected through a network. The server can push the message to the client, and the user can obtain the message through the client. It should be noted that fig. 1 is only an example.

Specifically, the client 01 may include a physical device of a type such as a smart phone, a desktop computer, a tablet computer, a notebook computer, an Augmented Reality (AR)/Virtual Reality (VR) device, a digital assistant, a smart wearable device, and the like, and may also include software running in the physical device, such as a computer program. The operating system running on client 01 may include, but is not limited to, an android system, an IOS system, linux, windows, and the like.

Specifically, the server 02 may include a server operating independently, or a distributed server, or a server cluster composed of a plurality of servers. The server 02 may comprise a network communication unit, a processor and a memory, etc. The server 02 may provide background services for the clients.

In practical applications, the solution provided by the embodiment of the present invention may relate to the related art of artificial intelligence, which will be described in the following specific embodiments. Artificial Intelligence (AI) is a new technical science to study and develop theories, methods, techniques and application systems for simulating, extending and expanding human Intelligence. The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like. With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and applied in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical care, smart customer service, and the like.

Machine Learning (ML) is a multi-domain cross subject, and relates to multiple subjects such as probability theory, statistics, approximation theory, convex analysis and algorithm complexity theory. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

The following describes a specific embodiment of a message pushing method according to the present invention, and fig. 2 is a schematic flow chart of a message pushing method according to an embodiment of the present invention, and this specification provides the method operation steps as described in the embodiment or the flow chart, but may include more or less operation steps based on conventional or non-inventive labor. The order of steps recited in the embodiments is merely one manner of performing the steps in a multitude of orders and does not represent the only order of execution. In practice, the system or server product may be implemented in a sequential or parallel manner (e.g., parallel processor or multi-threaded environment) according to the embodiments or methods shown in the figures. Specifically, as shown in fig. 2, the method may include:

s201: acquiring a candidate message;

in the embodiment of the present invention, the candidate message indicates a corresponding service. The candidate message may be generated based on an update of a message source subscribed by the target object (e.g., a message from a subscription number, a client-owned function (e.g., "dynamic" function, "point-of-view" function)), may be generated based on an update of an organization structure corresponding to the target object (e.g., a message brought by an added contact), may be generated based on an update of the client, and so on, as shown in fig. 6.

The candidate message may carry a type identifier, and the type identifier may be set based on a generation data source (refer to the foregoing description) of the candidate message, such as "subscription number message", "new contact message", "version upgrade message", and the like. The type identifier may be set based on a presentation form of pushing the candidate message to the client, such as a "pop-up window" form, a "red dot" form, and so on. The type identifier may be set in combination with a "form of pushing the candidate message to the client" and a "source of generation data of the candidate message," such as a "public number red dot" form, a "tab red dot" form, a "permission acquisition popup" form, and the like.

Further, the type label may be set in a finer granularity according to the relation between the page and the "red dot", "pop-up window", etc. as the display form, such as the "lock screen pop-up window" (not displayed together with the current page). Of course, the type label may be set according to the position of the display form "red dot", "pop-up window", etc. on the terminal screen (e.g. the middle part of the screen, the top part of the screen), and may be set according to the size and shape of the display form "red dot", "pop-up window", etc.

S202: acquiring state data of a target object;

in the embodiment of the present invention, the target object may indicate a current login account of the client and an identifier of the client. That is, the status data may characterize the status of registered users using the current login account, the status of guests experiencing the service based on the identity of the client. The status data may reflect operational behavior of registered users or guests.

The states here include the current state and the historical state. The current state may indicate a state of the current time point or may indicate a state of a short period of time (which may be referred to as a current time period including the current time point) before the current time point. The history status may indicate a status of a longer time period before the current time point, or may indicate a status of a longer time period before the current time point (including the current time point).

Accordingly, the status data includes current status data and historical status data. Wherein the current state data may be collected in real-time. The historical state data may be pre-collected, where the historical state data may be extracted from storing the pre-collected data.

In a specific embodiment, the server may collect a plurality of behavior data of the target object based on a current time period and a historical time period, where each behavior data carries corresponding attribute information. And then, the server carries out structural processing on the plurality of behavior data according to the corresponding attribute information to obtain the state data. Where the collected behavior data by the server based on the current time period may point to current state data. As for the historical state data, it may be obtained by the server based on the collected behavior data of the historical time period (first historical time period). Of course, the historical state data may also be configured to include collected behavior data based on the current time period by the server.

Behavioral data is data generated based on corresponding event triggers. The corresponding events may include "foreground cut event", "background cut event", "click event", "slide event" (which may include "slide message list event", for example), and the like. Further, "click event" may also include "click tab red dot event", "click friend message red dot event", "click view red dot event", "click contact red dot event", "click dynamic red dot event", and so on. The corresponding attribute information may correspond to a specific event, where the corresponding attribute information may include a common attribute (for example, the "click friend message red dot event" and the "click point red dot event" may correspond to the same common attribute: the click red dot attribute) and a specific attribute (for example, the specific attribute corresponding to the "click friend message red dot event" is the click friend message red dot, and the specific attribute corresponding to the "click point red dot event" is the click point red dot). Of course, the behavior data may characterize content associated with the event, such as "number of message reddots", "whether the message list is at the top", and so on, in addition to characterizing the event type.

And carrying out structured processing on the plurality of behavior data according to the corresponding attribute information. The state data thus obtained is structured data, each of which may indicate a specific event or a class of events. The structured data can be convenient for the server to carry out persistent storage, and the operations of recovering, searching, checking and the like on the state data can be ensured in the future.

In practical application, various behavior data of a target object can be collected by a client, and then the behavior data is organized into structured data with a good structure. And reporting the obtained structured data to a server.

S203: determining a click rate prediction model corresponding to the type of the candidate message, and inputting the state data into the click rate prediction model for click rate prediction;

in the embodiment of the invention, the click rate prediction model is obtained by performing machine learning training on a plurality of state sample data, the state sample data carries the actual click rate of the same type of message, and the same type of message and the candidate message belong to the same type. And performing machine learning training by using a plurality of labeled state sample data to obtain a click rate prediction model, wherein the obtained click rate prediction model has high generalization capability. Click rate prediction models are set in the dimensionality of the message types, and each message type (corresponding to one message pushing scene) can correspond to one click rate prediction model, so that the prediction adaptability of the click rates for pushing different types of messages to different users can be improved, and the reliability and the effectiveness of click rate prediction can be greatly improved.

The click rate output by the click rate prediction model can reflect whether the target object corresponds to the idle operation behavior at present to a certain extent, that is, can reflect whether a registered user or a guest is in an idle state using the client currently. Because if a registered user or guest is not currently using the client (quitting the use of the client, not "background"), the registered user or guest cannot immediately know the situation of a certain message received by the client. If a registered user or guest is currently using the client, but the registered user or guest is using the client to perform a service experience (e.g., chat), the registered user or guest may not view the message even if the registered user or guest knows that the client has received the message.

In a specific embodiment, the inputting the state data into the click-through rate prediction model for click-through rate prediction further includes:

extracting features of at least two dimensions from the state data, the at least two dimensions including a historical state dimension, a current state dimension, liveness, and feedback conditions. Current state features and historical state features may be extracted from the state data based on a historical state dimension and a current state dimension.

Specifically, the current state feature may be extracted from the current state data, and the historical state feature may be extracted from the historical state data. Here, "current state data" and "historical state data" refer to the relevant records in step S202, and are not described in detail.

In practical applications, the server receives the structured data reported by the client, and since each structured data may indicate a specific event or a class of events, the server may calculate the characteristics of the target object (corresponding to the status characteristics of the registered user or the guest) based on the specific event or the class of events.

For the current state feature, it may include:

1) user instant features

The method is used for characterizing the operation behavior of the registered user or the tourist at the current time point. The user instant feature may be derived from data reflecting whether the registered user or guest is currently looking at, the number of red dots of the current message list, etc.

2) User short term (near term) features

For characterizing the operation behavior of the registered user or guest of the current data segment (a short time segment from the current time point to the previous time point, including the current time point)). Various operational behaviors of the registered user or guest in the last several seconds and minutes can be characterized. For example, the number of viewing the QQ space in the last 30 seconds, the number of viewing the point in the last 10 seconds, and the like.

Before performing the user short-term (near-term) feature calculation, data (hereinafter referred to as short-term data) that can reflect the operation behavior of the registered user or guest at the current time period can be determined, and the short-term data can be stored. Of course, the upper limit threshold of the storage capacity can be set to avoid that the short-term data is stored too much, which results in insufficient storage space of the server.

For the historical state features, the method can comprise the following steps:

1) user long-term liveness characteristics

The user long-term characteristics are used for characterizing the operation behaviors of the registered users or tourists in the historical data segment (which can correspond to the state of a longer time segment before the current time point and can correspond to a longer time segment from the current time point (including the current time point) to the previous time segment).

The user long-term activity feature may characterize the activity of various operational behaviors of registered users or guests over the last several days, months, or even years. For example, the liveness of the client (e.g., qq) is used, the liveness of the view space, the liveness of the view article, etc. Liveness can be characterized in days, times, etc. Such as the number of recent active days, the number of recent days of access to a service, etc.

In practical application, the calculation of the long-term activity of the user can be performed by calculating each activity of each day and then combining the activities for multiple days. The corresponding activity is obtained by calculating the average value of the activity in a recent period of time, so that the influence of abnormal operation behaviors of the registered user or the tourist on activity estimation can be greatly reduced. In consideration of the load of the server, the client side can be responsible for computing tasks of various liveness degrees every day, and the server is responsible for merging for multiple days, so that the computing pressure of the server can be shared.

2) User feedback features

For characterizing whether the registered user or the guest clicks on the message after pushing the message to the registered user or the guest. The user feedback feature may be obtained by reflecting data on the number of times that the registered user or the guest knows that the client receives the message (the number of times of exposure of the message), the number of times of clicking on the message, the click rate inside the client, and the like.

Before the user feedback feature calculation is performed, a user feedback list may be established and stored based on the data, and each item of the list records information such as the pushing time of the message, whether the registered user or the guest clicks after the message is pushed, and the like. This facilitates the calculation of the user feedback characteristics from this list. Of course, considering the limited storage space of the server, the upper threshold of the user feedback list may be set, and then several recent entries may be reserved.

3) User static features

The characteristics, such as sex characteristics and age characteristics, of the registered user are stable or even do not change in a short period. The user static features may be made available through the account profile. Of course, when the user static feature calculation is performed, the correctness check and the abnormal value processing may be performed. For example, if the extracted age characteristics of a registered user correspond to 10000 years, an exception label may be made.

Correspondingly, the inputting the state data into the click rate prediction model for click rate prediction includes: and inputting the current state characteristics and the historical state characteristics into the click rate prediction model to predict the click rate.

In another specific embodiment, as shown in fig. 3, the training process of the click-through rate prediction model includes the following steps:

s301: acquiring the state sample data;

the method comprises the steps that a plurality of behavior data from different objects in a preset time period can be collected, and each behavior data carries corresponding attribute information; and according to the corresponding attribute information, carrying out structural processing on the plurality of behavior data to obtain the state sample data. The behavior data used as the state sample data comes from different objects, each of which may correspond to a plurality of behavior data, and the different objects may include the target object described above. Here, for the definition of the behavior data and the content of the structured processing on the behavior data, reference may be made to the relevant description in step S202, and details are not described again. The preset time period here exists in the model training phase.

In practical applications, various behavior data from different objects can be collected by a client and then organized into structured data with a good structure. And reporting the obtained structured data to a server.

S302: inputting the state sample data into a neural network model for click rate prediction training;

inputting the state sample data into a neural network model for click rate prediction training, wherein the method also comprises the following steps: and extracting features of at least two dimensions from the state sample data, wherein the at least two dimensions comprise a history state dimension, a current state dimension, an activity and a feedback condition. The "history state dimension" and the "current state dimension" are mainly in terms of the aforementioned preset time period (existing in the model training phase). The features of the at least two dimensions may include user immediate features, user short-term (near-term) features, user long-term activity features, user feedback features, and user static features. Reference may be made to the preceding description for obtaining these features, which are not described in detail here. In addition, although the state sample data is from different objects, "extracting features of at least two dimensions from the state sample data" is implemented for a certain object, respectively.

Inputting the state sample data into a neural network model for click rate prediction training, wherein the click rate prediction training comprises the following steps: inputting the characteristics of the at least two dimensions into the neural network model for click rate prediction training. The static features of the user carrying the exception labels may be filtered. Since each message type (corresponding to a message pushing scenario) may correspond to a click-through rate prediction model, the features input into the neural network model a (for message type a) may filter out features having a correlation with message type a lower than a preset value (such as the number of times of clicking on message type B).

The neural network model may adopt a DNN (Deep neural networks) model, an XGB (eXtreme Gradient Boosting) model, an LR (Logistic Regression) model, and the like.

In practical application, f (x) ═ y may represent a function to be fitted corresponding to the neural network model, x represents a feature obtained based on the state sample data, and y represents the click rate predicted in the model training stage.

S303: in the training process, adjusting the model parameters of the neural network model until the click rate output by the neural network model is matched with the actual click rate carried by the input state sample data;

for example, training a click-through rate prediction model of the subscription number message may count the actual click-through rate of the message belonging to the subscription number message type in the preset time period (the click-through number of the message/the exposure number of the message, the click-through rate inside the non-client).

A loss value between an intermediate value (click rate as an intermediate result of training) output by the neural network model and a labeled value (actual click rate as a correct answer) of the state sample data can be calculated, and the model parameter is adjusted according to the loss value. Specifically, the neural network model may be trained by a gradient descent method, an initial value of the learning rate is set to 0.0005 to 0.0015, and a value of the learning rate is iteratively adjusted every 1000 to 3000 times. For example, the initial value of the learning rate may be set to 0.001, and the value of the learning rate may be adjusted every 2000 iterations. Of course, the manner of setting the learning rate is not limited to this.

S304: and taking the neural network model corresponding to the adjusted model parameters as the click rate prediction model.

Fig. 5 is a schematic diagram of an application scenario of click rate prediction according to an embodiment of the present invention. In fig. 5, the training data are state sample data, and each state sample data is labeled with the actual click rate of the same type of message; correspondingly, the click rate prediction model trained subsequently can predict the click rate of the candidate message based on the state data of the target object.

In addition, the accuracy and validity of the click rate prediction model can be verified by using ABTest (AB test, a split type intergroup test method).

S204: when the predicted click rate meets the requirement of a click rate threshold value, the candidate message is pushed to the client;

in the embodiment of the present invention, a click-through rate prediction model is set in a dimension of a message type, and each message type (corresponding to one message pushing scenario) corresponds to one click-through rate prediction model, so that a click-through rate threshold corresponding to each message type (corresponding to one message pushing scenario) may be different and may be flexibly set. When setting the click rate threshold, the whole click rate needs to be considered: because if the click rate threshold is set to be too large, the exposure times of the messages are reduced, and the whole click rate is not ideal enough; if the click rate threshold is set to be too small, the data can be pushed to a large number of clients with unsatisfactory predicted click rates, and the total click rate is not ideal enough.

The click-through rate output by the click-through rate prediction model may reflect whether a registered user or guest is currently in an idle state using the client. When the predicted click rate is greater than the click rate threshold, it can be shown that the registered user or the guest is currently in an idle state using the client, and then the candidate message is pushed to the client. When the predicted click rate is less than or equal to the click rate threshold, it may be indicated that the registered user or the guest is not currently in an idle state using the client, and the candidate message is not pushed to the client.

In a specific embodiment, when the predicted click rate meets the requirement of the click rate threshold, pushing the candidate message to the client includes:

s401: when the predicted click rate meets the requirement of the click rate threshold, the candidate message is taken as a target message;

for example, the candidate message belongs to a new contact message type, the click rate threshold corresponding to the message type is 60%, and if the predicted click rate is 70% (> 60%), the candidate message is used as the target message.

S402: determining a sub-time period in which a current time point is located on a reference time period, wherein the reference time period comprises at least two sub-time periods, and each sub-time period corresponds to historical click rate distribution information;

the process for obtaining the historical click rate distribution information comprises the following steps: acquiring a first click condition of each similar message in a historical time period (second historical time); determining the reference time period, and determining the at least two sub-time periods over the reference time period; obtaining a second click condition of each similar message in each sub-time period according to the first click condition and the corresponding relation between the historical time period (second historical time) and the reference time period; and taking the second click condition as the historical click rate distribution information.

The second historical time period may indicate a longer time period before the current time point, or may indicate a longer time period before the current time point (including the current time point). The click case may indicate whether the registered user or guest clicks on the received message. For example, the second historical period may correspond to 1 month (e.g., 30 days), the reference period may correspond to 1 day, and the sub-periods may correspond to 1 hour, 3 hours, etc. For the messages of the same-genus newly-added contact message type, the click condition of each same-genus message in 1 month is obtained, and the same-genus messages can be unevenly distributed in the time dimension of the 1 month. Then, these similar messages are sorted based on the corresponding relationship between the historical time period (second historical time) and the reference time period, and the click rate of each sub-time period is obtained by combining the click condition. The number of the similar messages (equivalent to the number of exposures) corresponding to 30 days is 900, the similar messages corresponding to each sub-time period (morning, noon, afternoon and evening) are respectively extracted for each day of the 30 days, and the click rate corresponding to the morning, the click rate corresponding to the noon, the click rate corresponding to the afternoon and the click rate corresponding to the evening in 1 day can be obtained by combining the specific click condition.

Here, for example, when the current time point is 12, the sub-period is noon.

In addition, each of the messages of the same type described herein may be limited in terms of whether there is a corresponding registered account, whether the registered account is abnormal, and the like, in addition to the limitation of the message type and the second historical time period. Of course, the historical click rate distribution information may include mass historical click rate distribution information and individual historical click rate distribution information. Information is distributed for individual historical click rates. Each of the messages of the same kind may limit specific target objects (such as a specific current login account of the client and an identifier of the client) in terms of whether a corresponding registration account exists, whether the registration account is abnormal, and the like, in addition to the limitations of the message type and the second historical time period.

S403: determining a target time period after the sub-time period;

and when the corresponding sub-time period is the latest time period, acquiring a lower limit threshold of the pushing times, and if the current pushing times do not meet the requirement of the lower limit threshold of the pushing times, directly pushing the target message to the client. The lower threshold of the push times may refer to the minimum number of times of pushing messages (which may be limited to a specific message type or messages of all types) to the client every day, so as to improve the exposure chance of the target message and improve the overall click rate.

When the corresponding sub-period is before the latest sub-period, determining a target period after the sub-period. In connection with the example in step S402, the target time periods are afternoon and evening.

S404: comparing the click rate obtained by prediction with the click rate corresponding to the target time period;

in the first case: the click rate is 80% (greater than 70%) for afternoon and 50% (less than 70%) for evening.

In the second case: the click rate is 50% (less than 70%) for afternoon and 50% (less than 70%) for evening.

S405: and when the predicted click rate is larger than the click rate corresponding to the target time period, pushing the target message to the client.

In the first case, there is a time period (afternoon) when the click rate is greater than the predicted click rate, it is known that there is a better push opportunity than the current time point in the time after the current time point of the day, and specifically, there may be 1 better push opportunity than the current time point in the time after the current time point of the day, so that the target message may not be pushed to the client and the trigger of the better push opportunity may be waited.

In the second case, there is no time period in which the click rate is greater than the predicted click rate, the target message may be pushed to the client. Therefore, better pushing opportunity and integral click quantity can be effectively considered when the message is pushed to the client.

In another specific embodiment, the current time point may correspond to at least two of the candidate messages. Click rate prediction models corresponding to the types of the at least two candidate messages can be respectively determined; respectively inputting the state data into the corresponding click rate prediction models to predict click rates; and obtaining a priority level sequence for message pushing according to the predicted click rates corresponding to the at least two candidate messages.

1) When the at least two candidate messages belong to the same message type, click rate prediction can be performed through the same click rate prediction model, so that a priority level sequence (with a high click rate and a high priority level) of predicted click rates corresponding to the at least two candidate messages can be obtained, and the priority level sequence can indicate the sequence of subsequent message pushing. And then candidate messages corresponding to the click rate meeting the click rate threshold value requirement can be selected and sequentially pushed to the client according to the priority level sequence.

2) When the at least two candidate messages belong to different message types, click rate prediction can be performed through corresponding click rate prediction models, so that a priority level sequence (with a high click rate and a high priority level) of predicted click rates corresponding to the at least two candidate messages can be obtained, and the priority level sequence can indicate a subsequent message pushing sequence. And then candidate messages corresponding to the click rate meeting the requirement of the corresponding click rate threshold value can be selected and sequentially pushed to the client according to the priority level sequence.

In addition, it should be noted that, in the embodiment of the present invention, when the current time point corresponds to at least two candidate messages, the message pushing may be performed by combining the historical click rate distribution information and the priority level sequence. For example, the priority level sequence of the predicted click rates corresponding to the at least two candidate messages may be determined, then the candidate message corresponding to the click rate meeting the requirement of the corresponding click rate threshold is selected as the target message, and then whether a better push opportunity exists after the current time point is judged based on the historical click rate distribution information to determine whether to push the message. Compared with timing pushing, the user experience can be greatly improved, the click effect is effectively improved, and the method can be seen in fig. 7.

As can be seen from the technical solutions provided in the embodiments of the present specification, when message pushing is performed, a click rate prediction model corresponding to the type of the candidate message is determined, state data of the target object is input into the click rate prediction model to perform click rate prediction, and then whether message pushing is performed is determined according to a prediction result. The click rate is predicted by combining the state data of the target object, the current state and the historical state of the user are considered, and the click rate, the click UV (independent visitor) and the user experience of the message can be improved. Click rate prediction is carried out on the dimensionality of the message type, click rates corresponding to different message types can be determined in a finer-grained manner, and the message pushing accuracy is improved.

An embodiment of the present invention further provides a message pushing apparatus, as shown in fig. 8, the apparatus includes:

the candidate message acquisition module 810: for obtaining candidate messages;

the status data acquisition module 820: the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring state data of a target object;

click-through rate prediction module 830: the click rate prediction model is used for determining the click rate prediction model corresponding to the type of the candidate message, and inputting the state data into the click rate prediction model for click rate prediction;

the push module 840: the candidate message is pushed to the client when the predicted click rate meets the requirement of the click rate threshold;

It should be noted that the device and method embodiments in the device embodiment are based on the same inventive concept.

An embodiment of the present invention provides an electronic device, which includes a processor and a memory, where the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the message pushing method provided in the foregoing method embodiment.

Further, fig. 9 shows a schematic hardware structure of an electronic device for implementing the message pushing method provided by the embodiment of the present invention, where the electronic device may participate in forming or including the message pushing apparatus provided by the embodiment of the present invention. As shown in fig. 9, the electronic device 90 may include one or more (shown here as 902a, 902b, … …, 902 n) processors 902 (the processors 902 may include, but are not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 904 for storing data, and a transmission device 906 for communication functions. Besides, the method can also comprise the following steps: a display, an input/output interface (I/O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the I/O interface), a network interface, a power source, and/or a camera. It will be understood by those skilled in the art that the structure shown in fig. 9 is only an illustration and is not intended to limit the structure of the electronic device. For example, the electronic device 90 may also include more or fewer components than shown in FIG. 9, or have a different configuration than shown in FIG. 9.

It should be noted that the one or more processors 902 and/or other data processing circuitry described above may be referred to generally herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part in software, hardware, firmware, or any combination thereof. Further, the data processing circuit may be a single stand-alone processing module, or incorporated in whole or in part into any of the other elements in the electronic device 90 (or mobile device). As referred to in the embodiments of the application, the data processing circuit acts as a processor control (e.g. selection of a variable resistance termination path connected to the interface).

The memory 904 may be used for storing software programs and modules of application software, such as program instructions/data storage devices corresponding to the method described in the embodiment of the present invention, and the processor 902 executes various functional applications and data processing by running the software programs and modules stored in the memory 94, so as to implement one of the message pushing methods described above. The memory 904 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 904 may further include memory located remotely from the processor 902, which may be connected to the electronic device 90 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmitting means 906 is used for receiving or sending data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the electronic device 90. In one example, the transmission device 906 includes a network adapter (NIC) that can be connected to other network devices through a base station so as to communicate with the internet. In one embodiment, the transmitting device 906 can be a Radio Frequency (RF) module, which is used for communicating with the internet in a wireless manner.

The display may be, for example, a touch screen type Liquid Crystal Display (LCD) that may enable a user to interact with a user interface of the electronic device 90 (or mobile device).

Embodiments of the present invention also provide a storage medium, which may be disposed in an electronic device to store at least one instruction, at least one program, a code set, or a set of instructions related to implementing a message pushing method in the method embodiments, where the at least one instruction, the at least one program, the code set, or the set of instructions are loaded and executed by the processor to implement the message pushing method provided in the method embodiments.

Alternatively, in this embodiment, the storage medium may be located in at least one network server of a plurality of network servers of a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.

It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the device and electronic apparatus embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference may be made to some descriptions of the method embodiments for relevant points.

It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A message pushing method, the method comprising:

acquiring a candidate message;

acquiring state data of a target object;

2. The method of claim 1, wherein when the predicted click through rate meets a requirement of a click through rate threshold, pushing the candidate message to a client comprises:

when the predicted click rate meets the requirement of the click rate threshold, the candidate message is taken as a target message;

determining a sub-time period in which a current time point is located on a reference time period, wherein the reference time period comprises at least two sub-time periods, and each sub-time period corresponds to historical click rate distribution information;

determining a target time period after the sub-time period;

comparing the click rate obtained by prediction with the click rate corresponding to the target time period;

and when the predicted click rate is larger than the click rate corresponding to the target time period, pushing the target message to the client.

3. The method of claim 2, wherein the obtaining of the historical click rate distribution information comprises:

acquiring a first click condition of each similar message in a historical time period;

determining the reference time period, and determining the at least two sub-time periods over the reference time period;

obtaining a second click condition of each similar message in each sub-time period according to the first click condition and the corresponding relation between the historical time period and the reference time period;

and taking the second click condition as the historical click rate distribution information.

4. The method according to any one of claims 1-3:

the obtaining the candidate message comprises:

acquiring at least two candidate messages;

the determining a click rate prediction model corresponding to the type of the candidate message and inputting the state data into the click rate prediction model for click rate prediction include:

respectively determining click rate prediction models corresponding to the types of the at least two candidate messages;

respectively inputting the state data into the corresponding click rate prediction models to predict click rates;

when the predicted click rate meets the requirement of the click rate threshold, the candidate message is pushed to the client, and the method comprises the following steps:

and obtaining a priority level sequence for message pushing according to the predicted click rates corresponding to the at least two candidate messages.

5. The method of claim 1, wherein:

the inputting the state data into the click rate prediction model for click rate prediction further comprises:

extracting features of at least two dimensions from the state data, wherein the at least two dimensions comprise a history state dimension, a current state dimension, an activity and a feedback condition;

correspondingly, the inputting the state data into the click rate prediction model for click rate prediction includes:

and inputting the characteristics of the at least two dimensions into the click rate prediction model to predict the click rate.

6. The method of claim 1, wherein the obtaining state data of the target object comprises:

acquiring a plurality of behavior data of the target object based on a current time period and a historical time period, wherein each behavior data carries corresponding attribute information;

and according to the corresponding attribute information, carrying out structural processing on the plurality of behavior data to obtain the state data.

7. The method of claim 1, wherein the training process of the click-through rate prediction model comprises the steps of:

acquiring the state sample data;

inputting the state sample data into a neural network model for click rate prediction training;

in the training process, adjusting the model parameters of the neural network model until the click rate output by the neural network model is matched with the actual click rate carried by the input state sample data;

and taking the neural network model corresponding to the adjusted model parameters as the click rate prediction model.

8. The method of claim 7, wherein:

the acquiring the state sample data includes:

acquiring a plurality of behavior data from different objects within a preset time period, wherein each behavior data carries corresponding attribute information;

according to the corresponding attribute information, performing structural processing on the plurality of behavior data to obtain the state sample data;

inputting the state sample data into a neural network model for click rate prediction training, wherein the method also comprises the following steps:

extracting features of at least two dimensions from the state sample data, wherein the at least two dimensions comprise a history state dimension, a current state dimension, an activity degree and a feedback condition;

inputting the state sample data into a neural network model for click rate prediction training, wherein the click rate prediction training comprises the following steps:

inputting the characteristics of the at least two dimensions into the neural network model for click rate prediction training.

9. A message push apparatus, the apparatus comprising:

a candidate message acquisition module: for obtaining candidate messages;

10. A computer readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by a processor to implement the message pushing method according to any one of claims 1-8.