CN105915438B - Information push method, apparatus and system - Google Patents
Information push method, apparatus and system Download PDFInfo
- Publication number
- CN105915438B CN105915438B CN201610235874.4A CN201610235874A CN105915438B CN 105915438 B CN105915438 B CN 105915438B CN 201610235874 A CN201610235874 A CN 201610235874A CN 105915438 B CN105915438 B CN 105915438B
- Authority
- CN
- China
- Prior art keywords
- message
- push
- pushed
- user
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/21—Monitoring or handling of messages
- H04L51/214—Monitoring or handling of messages using selective forwarding
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/58—Message adaptation for wireless communication
Abstract
The invention discloses a kind of information push methods, apparatus and system.Wherein method includes: to obtain message to be pushed, and chooses the push attributive character of message;The data that the case where having pushed user feeds back, which are pushed to, according to message obtains training sample data;Using the training sample data training pattern, model parameter is generated;The push attributive character parameter of the message is determined according to the association attributes of the association attributes of the message and user to be pushed, the push attributive character parameter of the message and the model parameter are substituted into model, obtain whether to user's PUSH message to be pushed probability, whether the message to user to be pushed is pushed according to the determine the probability.Present invention consideration has pushed user to the click data of message, which is reflection user to the whether interested key factor of message, and message can be accurately pushed to interested user in the way of this push, greatly improve pushing efficiency.
Description
Technical field
The present invention relates to Internet technical fields, and in particular to a kind of information push method, apparatus and system.
Background technique
Message push is referred to sending from server end to terminal and be connected, and transmits certain information.Push function is in mobile phone
The scene applied in exploitation is more and more, such as the new information that the news client on mobile phone is pushed with regard to real-time reception, facilitates use
Read newest news information in family.
In message push technology, it is particularly important for how screening purpose user.In the prior art, server end exists
Before intending to push a message, user can be generally screened by way of static matching.Specifically, server end is true first
Surely the feature for needing static matching, for pushing red packet message, server end chooses geographical location, product version, user's category
The features such as property;Qualified user is searched according to these features, red packet message is then pushed to these users.This push
The influence factor that mode considers is only some preset static natures, relatively simple, tends not to accurately to be pushed to pair
The interested user there of this message, pushing efficiency be not high.
Summary of the invention
In view of the above problems, it proposes on the present invention overcomes the above problem or at least be partially solved in order to provide one kind
State information push method, the apparatus and system of problem.
According to an aspect of the invention, there is provided a kind of information push method, comprising:
Message to be pushed is obtained, the push attributive character of the message is chosen;
The data that the case where having pushed user feeds back, which are pushed to, according to message obtains training sample data, the training sample
Data have pushed user to the click data of the message described in including at least;
Using the training sample data training pattern, model parameter is generated;
Determine that the push attribute of the message is special according to the association attributes of the association attributes of the message and user to be pushed
Parameter is levied, the push attributive character parameter of the message and the model parameter are substituted into model, whether obtained to wait push
Whether the probability of user's PUSH message to user to be pushed pushes the message according to the determine the probability.
According to another aspect of the present invention, a kind of message driving means is provided, comprising:
Message processing module chooses the push attributive character of the message suitable for obtaining message to be pushed;
Sample extraction module obtains training sample suitable for being pushed to the data that the case where having pushed user feeds back according to message
Data, the training sample data have pushed user to the click data of the message including at least described;
Model training module is suitable for utilizing the training sample data training pattern, generates model parameter;
Message pushing module, suitable for according to the association attributes of the message and the determination of the association attributes of user to be pushed
The push attributive character parameter of message substitutes into the push attributive character parameter of the message and the model parameter in model,
Obtain whether to user's PUSH message to be pushed probability, according to the determine the probability whether to user to be pushed push described in disappear
Breath.
According to another aspect of the present invention, a kind of message push system is provided, comprising: above-mentioned message driving means,
Further include: message management platform and message task scheduler;
The message management platform is suitable for: customizing message according to message template, sends a message to message task scheduler;
The message task scheduler is suitable for: scheduling message forms push task, is sent to message pushing module and is pushed away
It send.
Information push method, the apparatus and system provided according to the present invention, after the push attributive character for choosing message,
These Feature Selections push purpose user is not matched not instead of simply, and foundation has pushed click data conduct of the user to message
Training sample data training pattern is determined according to updated model parameter is dynamically adjusted to which with dynamic adjustment model parameter
A little purpose users continue PUSH message.Present invention consideration has pushed user to the click data of message, which is reflection
Message can be accurately pushed to interested the whether interested key factor of message by user in the way of this push
User greatly improves pushing efficiency.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, the followings are specific embodiments of the present invention.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 shows the flow chart of information push method according to an embodiment of the invention;
Fig. 2 shows the flow charts of information push method in accordance with another embodiment of the present invention;
Fig. 3 shows the functional block diagram of message driving means according to an embodiment of the invention;
Fig. 4 shows the functional block diagram of message push system according to an embodiment of the invention.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
Fig. 1 shows the flow chart of information push method according to an embodiment of the invention.As shown in Figure 1, this method
Include the following steps:
Step S101 obtains message to be pushed, and chooses the push attributive character of the message.
In a specific embodiment of the invention, there is server end special message push system to be used for user's end
Hold PUSH message.The message push system includes message management platform, and message task scheduler and use are proposed by the present invention
The device of information push method PUSH message.Firstly, message management platform is responsible for customizing message according to message template.Disappeared with red packet
For breath, it is related with message itself that message management platform customizes type of message, message pattern, official documents and correspondence content of red packet message etc.
Attribute.After customizing message, message task scheduler schedules message is as push task.If some message is scheduled as pushing
Task filters out the purpose user for pushing the message using method provided by the invention then as message to be pushed.
It in this step, gets after the message pushed, first chooses the push attributive character of the message.In the present invention, push away
Attributive character is sent to designate the factor whether message is pushed that influences.These factors are not only related with the attribute of message itself, also
It is related with the attribute of purpose user and some other association attributes.Specifically, the push attributive character of message may include following
Information it is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, user
Figure Characteristics and product version.Wherein, type of message and message pattern are the attributes of message itself.User's Figure Characteristics are
The attribute of purpose user may include the characteristic informations such as user's gender, age, hobby.Regional information, temporal information, same type
The clicking rate of message, product version these be to belong to other association attributes corresponding with message.By taking weather forecast as an example, the message
It is closely bound up with regional information and temporal information, it also should be used as push attributive character.Push attributive character listed above is only
Example, in actual application, push attributive character can according to PUSH message actual content and Group-oriented carry out spirit
Setting living, the invention is not limited in this regard.
Step S102 is pushed to the data that the case where having pushed user feeds back according to message and obtains training sample data,
Middle training sample data include at least and have pushed user to the click data of message.
The present invention unlike the prior art, the present invention no longer with static nature match purpose user, but mainly according to
According to having pushed user to the click data of message as training sample data training pattern, with dynamic adjustment model parameter, according to
Dynamic, which adjusts updated model parameter and determines to which purpose user, continues PUSH message.
The initial stage of completion has just been customized in message, server end can preassign a collection of user user as a purpose,
The data of feedback are collected after to these user's PUSH messages as initial training sample data, which at least wraps
Containing these users to the click data of message.
Step S103 generates model parameter using training sample data training pattern.
Model used by the embodiment of the present invention is according to constructed by the push attributive character parameter and model coefficient of message
Assessment push probability model, wherein model coefficient be influence push attributive character parameter weight, by having pushed user
Training pattern is come to the click data of message, carries out optimization and handles the model coefficient optimized, and obtain by model system
The determining push probability threshold value of number.Training process in this way, it is desirable to which obtained model parameter is the subsequent push purpose of screening
The index of user.
Step S104 determines that the push attribute of message is special according to the association attributes of the association attributes of message and user to be pushed
Parameter is levied, the push attributive character parameter and model parameter of message are substituted into model, obtains whether pushing to user to be pushed
The probability of message, according to determine the probability whether to user's PUSH message to be pushed.
After training obtains model parameter, this step goes out to push purpose user according to trained model discrimination.Due to
Selected push attributive character is related with the association attributes of user, therefore, for particular message and specific user, it is first determined
Push attributive character parameter.For playing message safely, if the push attributive character chosen includes: type of message, message sample
Formula, regional information, temporal information, the clicking rate of same type message, user's Figure Characteristics and product version, it is determined that each push away
Send attributive character parameter can specifically: type of message is security classes, and message pattern is exaggeration pattern, and regional information is Beijing, when
Between information be * day * month * of * * *, the clicking rate of same type message is for 30% (especially by the number for inquiring preservation clicking rate data
Obtained according to library), user's Figure Characteristics are the Figure Characteristics of the specific user, and product version is mobile phone bodyguard.Determine pushing away for message
After sending attributive character parameter, the push attributive character parameter and model parameter of message are substituted into model, whether obtained to wait push away
The probability for sending user's PUSH message, according to determine the probability whether to user's PUSH message to be pushed.
The information push method provided according to the present invention is not simple after the push attributive character for choosing message
These Feature Selections push purpose user is matched, but foundation has pushed user to the click data of message as number of training
It is determining to which purpose user according to updated model parameter is dynamically adjusted with dynamic adjustment model parameter according to training pattern
Continue PUSH message.Present invention consideration has pushed user to the click data of message, which is reflection user to message
Whether interested key factor, message can be accurately pushed to interested user in the way of this push, significantly
Improve pushing efficiency.
According to above description it is found that the feelings that the technology scene that the present invention is applicable in, which belongs to, is response variable for two classification
Condition, that is, investigate whether to specific user push particular message.In the following embodiments, the present invention is with two sorted logics
(logistic) it for regression model, describes in detail to the concrete scheme of information push method.
Fig. 2 shows the flow charts of information push method in accordance with another embodiment of the present invention.As shown in Fig. 2, the party
Method includes the following steps:
Step S201 obtains message to be pushed, and chooses the push attributive character of the message.
In a specific embodiment of the invention, there is server end special message push system to be used for user's end
Hold PUSH message.The message push system includes message management platform, and message task scheduler and use are proposed by the present invention
The device of information push method PUSH message.Firstly, message management platform is responsible for customizing message according to message template.Disappeared with red packet
For breath, it is related with message itself that message management platform customizes type of message, message pattern, official documents and correspondence content of red packet message etc.
Attribute.After customizing message, message task scheduler schedules message is as push task.If some message is scheduled as pushing
Task filters out the purpose user for pushing the message using method provided in this embodiment then as message to be pushed.
It in this step, gets after the message pushed, first chooses the push attributive character of the message.In the present invention, push away
Attributive character is sent to designate the factor whether message is pushed that influences.These factors are not only related with the attribute of message itself, also
It is related with the attribute of purpose user and some other association attributes.Specifically, the push attributive character of message may include following
Information it is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, user
Figure Characteristics and product version.Wherein, type of message and message pattern are the attributes of message itself.User's Figure Characteristics are
The attribute of purpose user may include the characteristic informations such as user's gender, age, hobby.Regional information, temporal information, same type
The clicking rate of message, product version these be to belong to other association attributes corresponding with message.By taking weather forecast as an example, the message
It is closely bound up with regional information and temporal information, it also should be used as push attributive character.Push attributive character listed above is only
Example, in actual application, push attributive character can according to PUSH message actual content and Group-oriented carry out spirit
Setting living, the invention is not limited in this regard.
Step S202 according to the association attributes of message and has pushed user's to before having pushed user's PUSH message
Association attributes determine the push attributive character parameter of message.
The initial stage of completion has just been customized in message, server end can preassign a collection of user user as a purpose,
It selects to these user's PUSH messages.Preassigned user, which can be, to be randomly selected, and is also possible to as prior art
It is chosen by static nature matching.It is related to these users' according to the association attributes of message after determining purpose user
Attribute determines the push attributive character parameter of message.For playing message safely, if the push attributive character chosen includes: disappearing
Cease type, message pattern, regional information, temporal information, the clicking rate of same type message, user's Figure Characteristics and product version
This, it is determined that each push attributive character parameter can specifically: type of message is security classes, and message pattern is exaggeration pattern, ground
Domain information is Beijing, and temporal information is * days * month * of * * *, and the clicking rate of same type message is 30% (especially by inquiry preservation
The database of clicking rate data obtains), user's Figure Characteristics are the Figure Characteristics of the specific user, and product version is defended for mobile phone
Scholar.
In two sorted logic regression models, push attributive character parameter is indicated with x, x is n-dimensional vector parameter, x=(x1,
x2,…,xn), wherein x1,x2,…,xnThe parameter value of attributive character is pushed for n.By taking type of message as an example, be divided into security classes,
Red packet class, commercial paper etc., the corresponding specific value of each classification, security classes 1, red packet class is 2, and commercial paper is 3 etc.,
The specific value is exactly the parameter value for pushing attributive character.And so on, other push attributive character have also respectively corresponded parameter
Value, can determine according to the actual situation, finally obtain the parameter value of each push attributive character, then obtain n-dimensional vector x.
Step S203, to user's PUSH message has been pushed, the data fed back are as training sample data, the training sample
Notebook data includes at least and has pushed user to the click data of message.
To after having pushed user's PUSH message, the data of feedback are collected, the data of the feedback mainly include to have pushed
User is to the click data of message, if some, which has pushed user, clicks the message, the click data of feedback be clicked, if
Some has pushed user and has not clicked the message, and the click data of feedback is not click on.In addition to this, the data of collection can also wrap
Clicking rate containing the message.
Step S204, will be according to the push category of the association attributes of message and the message for the association attributes determination for having pushed user
Property characteristic parameter and training sample data substitute into model, carry out optimization processing according to the optimization algorithm of model, generate model
Parameter.
The formula of two sorted logic regression models used by the present embodiment is as follows:
P (y=1 | x;θ)=hθ(x)
P (y=0 | x;θ)=1-hθ(x) (2)
Wherein, x is the push attributive character parameter of message, and θ is model coefficient, hθIt (x) is push probability, y=1 expression pushes away
It send, y=0 expression does not push.
The value that P is determined according to the range of purpose user of push and the click data of feedback substitutes into x and P above-mentioned
In formula, optimization processing is carried out according to batch gradient descent algorithm, with specific reference to following formula:
Above-mentioned gradient descent algorithm is repeated, until convergence, wherein α is constant.It is final to determine the θ optimized, and according to θ
Obtained push probability threshold value.
Step S205 determines that the push attribute of message is special according to the association attributes of the association attributes of message and user to be pushed
Levy parameter.
After training obtains model parameter, so that it may according to the subsequent purpose user of the model discrimination.Due to selected
It is related with the association attributes of user to push attributive character, therefore, for particular message and specific user, it is first determined push attribute
Characteristic parameter determines x.The determination process of x can be found in the description of previous step, repeat no more.
It can be stored in database by the push attributive character parameter that this step determines, in case subsequent training pattern uses.
I.e. after message has been pushed a period of time, in above-mentioned steps S202, it is no longer necessary to according to the association attributes of message and
The association attributes of push user redefine the push attributive character parameter of message, but push can be directly taken out from database
Attributive character parameter.Certainly, for user to be pushed, also need this step according to the association attributes of message and the phase of user to be pushed
Close the push attributive character parameter that attribute determines message.
Step S206, by the push category of the message determined according to the association attributes of message and the association attributes of user to be pushed
Property characteristic parameter and model parameter substitute into model in, obtain whether to user's PUSH message to be pushed probability.
The step S205 x determined the and step S204 model parameter generated is substituted into θ to substitute into model, that is, substitutes into above-mentioned public affairs
In formula (1), h is obtainedθ(x)。
Probability is compared with push probability threshold value, determines whether use to be pushed according to comparison result by step S207
Family PUSH message.
By hθ(x) it is compared with predetermined push probability threshold value, if hθ(x) it is greater than or equal to push probability threshold
Value, it is determined that user's PUSH message to be pushed;If hθ(x) it is less than push probability threshold value, it is determined that not to user to be pushed
PUSH message.
After step S207 is determined to user's PUSH message to be pushed, method further include: pushed to user to be pushed
After message, the data that the case where pushing messages to user to be pushed is fed back are obtained, are stored in database for use as subsequent
Training sample data.That is, method provided by the invention is the process of a dynamic optimization, according to PUSH message feedback
Data dynamic adjustment model parameter improves the precision and validity of push so that model parameter keeps optimizing in real time.
Fig. 3 shows the functional block diagram of message driving means according to an embodiment of the invention.As shown in figure 3, the dress
Setting includes: message processing module 301, sample extraction module 302, model training module 303 and message pushing module 304.
Message processing module 301 is suitable for obtaining message to be pushed, and chooses the push attributive character of message.
Push attributive character designates the factor whether message is pushed that influences.These factors not only with the category of message itself
Property is related, also related with the attribute of purpose user and some other association attributes.Specifically, the push attributive character of message
It may include the one or more of following information: type of message, message pattern, regional information, temporal information, the point of same type message
Hit rate, user's Figure Characteristics and product version.Wherein, type of message and message pattern are the attributes of message itself.User draws
Attribute as being characterized in purpose user, may include the characteristic informations such as user's gender, age, hobby.Regional information, time letter
Breath, the clicking rate of same type message, product version these be to belong to other association attributes corresponding with message.It is with weather forecast
Example, the message and regional information and temporal information are closely bound up, also should be used as push attributive character.Push attribute listed above
Feature is merely illustrative, and in actual application, push attributive character can be according to the actual content to PUSH message and towards group
Body is flexibly set, the invention is not limited in this regard.
Sample extraction module 302 is suitable for being pushed to the data that the case where having pushed user feeds back according to message and obtains training sample
Notebook data, the training sample data include at least and have pushed user to the click data of message.
The present apparatus unlike the prior art, the present apparatus no longer with static nature match purpose user, but mainly according to
According to having pushed user to the click data of message as training sample data training pattern, with dynamic adjustment model parameter, according to
Dynamic, which adjusts updated model parameter and determines to which purpose user, continues PUSH message.
The initial stage of completion has just been customized in message, server end can preassign a collection of user user as a purpose,
To after these user's PUSH messages, sample extraction module 302 collects the data of feedback as initial training sample data, the instruction
Practice sample data and includes at least these users to the click data of message.Preassigned user, which can be, to be randomly selected,
It can be as prior art through static nature matching selection.
The data that sample extraction module 302 collects feedback mainly include the click data for having pushed user to message, if
Some has pushed user and has clicked the message, and the click data of feedback is to have clicked, if some, which has pushed user, does not click this
Message, the click data of feedback are not click on.In addition to this, the data of collection also may include the clicking rate of the message.
Model training module 303 is suitable for utilizing training sample data training pattern, generates model parameter.
Model used by the present apparatus is the assessment according to constructed by the push attributive character parameter and model coefficient of message
The model of probability is pushed, wherein model coefficient is the weight for influencing push attributive character parameter, by having pushed user to message
Click data carry out training pattern, generate model parameter.Training process in this way, it is desirable to which obtained model parameter is screening
The index of subsequent push purpose user.
Message pushing module 304 is suitable for determining message according to the association attributes of message and the association attributes of user to be pushed
Attributive character parameter is pushed, the push attributive character parameter and model parameter of message are substituted into model, whether obtained to wait push away
The probability for sending user's PUSH message, according to determine the probability whether to user's PUSH message to be pushed.
After training obtains model parameter, message pushing module 304 goes out to push purpose according to trained model discrimination
User.Since selected push attributive character is related with the association attributes of user, for particular message and specific use
Family, it is first determined push attributive character parameter.After the push attributive character parameter for determining message, by the push attributive character of message
Parameter and model parameter substitute into model, obtain whether to user's PUSH message to be pushed probability, according to determine the probability whether
To user's PUSH message to be pushed.
Further, the message driving means further include: characteristic parameter determining module 305, suitable for having pushed user
Before PUSH message, determine that the push attributive character of message is joined according to the association attributes of message and the association attributes for having pushed user
Number.After determining purpose user, characteristic parameter determining module 305 belongs to according to the association attributes of message are related to these users'
Property determines the push attributive character parameter of message.For playing message safely, if the push attributive character chosen includes: message
Type, message pattern, regional information, temporal information, the clicking rate of same type message, user's Figure Characteristics and product version,
Then determine that each push attributive character parameter can specifically: type of message is security classes, and message pattern is exaggeration pattern, region letter
Breath is Beijing, and temporal information is * days * month * of * * *, and the clicking rate of same type message (is saved especially by inquiry and clicked for 30%
The database of rate data obtains), user's Figure Characteristics are the Figure Characteristics of the specific user, and product version is mobile phone bodyguard.
By taking model is two sorted logic regression models as an example, push attributive character parameter is indicated with x, x is n-dimensional vector ginseng
Number, x=(x1,x2,…,xn), wherein x1,x2,…,xnThe parameter value of attributive character is pushed for n.By taking type of message as an example, point
For security classes, red packet class, commercial paper etc., the corresponding specific value of each classification, security classes 1, red packet class is 2, advertisement
Class is 3 etc., which is exactly the parameter value for pushing attributive character.And so on, other push attributive character are also distinguished
It is corresponding with parameter value, can be determined according to the actual situation, the parameter value of each push attributive character is finally obtained, then obtains n
Dimensional vector x.
Further, model training module 303 is further adapted for: according to the association attributes of message and will push user's
The push attributive character parameter and training sample data for the message that association attributes determine substitute into model, calculate according to the optimization of model
Method carries out optimization processing, generates model parameter.Model training module 303 is further adapted for: according to batch gradient descent algorithm
Optimization processing is carried out, the model parameter comprising model coefficient and the push probability threshold value obtained according to model coefficient is generated.
By taking model is two sorted logic regression models as an example, formula is as follows:
P (y=1 | x;θ)=hθ(x)
P (y=0 | x;θ)=1-hθ(x) (5)
Wherein, x is the push attributive character parameter of message, and θ is model coefficient, hθIt (x) is push probability, y=1 expression pushes away
It send, y=0 expression does not push.
Model training module 303 determines the value of P according to the range of purpose user and the click data of feedback of push,
X and P is substituted into above-mentioned formula, carries out optimization processing according to batch gradient descent algorithm, with specific reference to following formula:
Above-mentioned gradient descent algorithm is repeated, until convergence, wherein α is constant.It is final to determine the θ optimized, and according to θ
Obtained push probability threshold value.
Further, message pushing module 304 is further adapted for: probability is compared with push probability threshold value, according to
Comparison result determines whether that user to be pushed pushes the message.
The mould that message pushing module 304 generates the x that characteristic parameter determining module 305 determines and model training module 303
Shape parameter substitutes into θ and substitutes into model, i.e., in substitution above-mentioned formula (4), obtains hθ(x).By hθ(x) general with predetermined push
Rate threshold value is compared, if hθ(x) it is greater than or equal to push probability threshold value, it is determined that user's PUSH message to be pushed;Such as
Fruit hθ(x) it is less than push probability threshold value, it is determined that not to user's PUSH message to be pushed.
Further, device further include: database 306 is suitable for storage in after user's PUSH message to be pushed, institute
The data fed back the case where pushing messages to user to be pushed obtained, for use as subsequent training sample data.
The message driving means provided according to the present invention, message processing module choose message push attributive character it
Afterwards, these Feature Selections push purpose user is not matched not instead of simply, and user couple has been pushed by model training module foundation
The click data of message is as training sample data training pattern, and with dynamic adjustment model parameter, message pushing module is according to dynamic
State, which adjusts updated model parameter and determines to which purpose user, continues PUSH message.Present apparatus consideration has pushed user and has offseted
The click data of breath, which is reflection user to the whether interested key factor of message, in the way of this push
Message can be accurately pushed to interested user, greatly improve pushing efficiency.
Fig. 4 shows the functional block diagram of message push system according to an embodiment of the invention.As shown in figure 4, this disappears
Ceasing supplying system includes: message management platform 401, message task scheduler 402 and above-mentioned message driving means 403.
Message management platform 401 is suitable for: customizing message according to message template, sends a message to message task scheduler;
Message task scheduler 402 is suitable for: scheduling message forms push task, is sent to message pushing module 304 and is pushed.
Message management platform 401 is responsible for customizing message according to message template.By taking red packet message as an example, message management platform is fixed
The attribute related with message itself such as type of message, message pattern, the official documents and correspondence content of red packet message processed.After customizing message, disappear
402 scheduling message of task dispatcher is ceased as push task.Each push task has the status information of itself, records message
The state whether being pushed.When message pushing module 304 receives push task, according to above-mentioned push mode PUSH message.
The message push system provided according to the present invention, message processing module choose message push attributive character it
Afterwards, these Feature Selections push purpose user is not matched not instead of simply, and user couple has been pushed by model training module foundation
The click data of message is as training sample data training pattern, and with dynamic adjustment model parameter, message pushing module is according to dynamic
State, which adjusts updated model parameter and determines to which purpose user, continues PUSH message.This system consideration has pushed user and has offseted
The click data of breath, which is reflection user to the whether interested key factor of message, in the way of this push
Message can be accurately pushed to interested user, greatly improve pushing efficiency.
Algorithm and display are not inherently related to any particular computer, virtual system, or other device provided herein.
Various general-purpose systems can also be used together with teachings based herein.As described above, it constructs required by this kind of system
Structure be obvious.In addition, the present invention is also not directed to any particular programming language.It should be understood that can use various
Programming language realizes summary of the invention described herein, and the description done above to language-specific is to disclose this hair
Bright preferred forms.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that implementation of the invention
Example can be practiced without these specific details.In some instances, well known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the disclosure and help to understand one or more of the various inventive aspects,
Above in the description of exemplary embodiment of the present invention, each feature of the invention is grouped together into single implementation sometimes
In example, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention: i.e. required to protect
Shield the present invention claims features more more than feature expressly recited in each claim.More precisely, as following
Claims reflect as, inventive aspect is all features less than single embodiment disclosed above.Therefore,
Thus the claims for following specific embodiment are expressly incorporated in the specific embodiment, wherein each claim itself
All as a separate embodiment of the present invention.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodiment
Change and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodiment
Member or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use any
Combination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed
All process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint power
Benefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purpose
It replaces.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments mean it is of the invention
Within the scope of and form different embodiments.For example, in the following claims, embodiment claimed is appointed
Meaning one of can in any combination mode come using.
Various component embodiments of the invention can be implemented in hardware, or to run on one or more processors
Software module realize, or be implemented in a combination thereof.It will be understood by those of skill in the art that can be used in practice
Microprocessor or digital signal processor (DSP) are realized in information push-delivery apparatus according to an embodiment of the present invention and system
The some or all functions of some or all components.The present invention is also implemented as executing method as described herein
Some or all device or device programs (for example, computer program and computer program product).Such reality
Existing program of the invention can store on a computer-readable medium, or may be in the form of one or more signals.
Such signal can be downloaded from an internet website to obtain, and perhaps be provided on the carrier signal or in any other forms
It provides.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and ability
Field technique personnel can be designed alternative embodiment without departing from the scope of the appended claims.In the claims,
Any reference symbol between parentheses should not be configured to limitations on claims.Word "comprising" does not exclude the presence of not
Element or step listed in the claims.Word "a" or "an" located in front of the element does not exclude the presence of multiple such
Element.The present invention can be by means of including the hardware of several different elements and being come by means of properly programmed computer real
It is existing.In the unit claims listing several devices, several in these devices can be through the same hardware branch
To embody.The use of word first, second, and third does not indicate any sequence.These words can be explained and be run after fame
Claim.
The invention discloses:
A1, a kind of information push method, comprising:
Message to be pushed is obtained, the push attributive character of the message is chosen;
The data that the case where having pushed user feeds back, which are pushed to, according to message obtains training sample data, the training sample
Data have pushed user to the click data of the message described in including at least;
Using the training sample data training pattern, model parameter is generated;
Determine that the push attribute of the message is special according to the association attributes of the association attributes of the message and user to be pushed
Parameter is levied, the push attributive character parameter of the message and the model parameter are substituted into model, whether obtained to wait push
Whether the probability of user's PUSH message to user to be pushed pushes the message according to the determine the probability.
A2, information push method according to a1, it is described to be pushed to what the case where having pushed user fed back according to message
Data obtain training sample data:
To it is described pushed user's PUSH message before, according to the association attributes of the message and pushed the phase of user
Close the push attributive character parameter that attribute determines the message;
User's push message is pushed to described, the data fed back are as training sample data.
A3, the information push method according to A2, it is described to utilize training sample data training pattern, generate model parameter
Further comprise:
It will be according to the push category of the association attributes of the message and the message for the association attributes determination for having pushed user
Property characteristic parameter and training sample data substitute into model, carry out optimization processing according to the optimization algorithm of model, generate model
Parameter.
A4, information push method according to a3, the optimization algorithm according to model optimize and are handled into one
Step includes: to carry out optimization processing according to batch gradient descent algorithm.
A5, according to the described in any item information push methods of A1-A4, the model parameter include model coefficient and according to
The push probability threshold value that model coefficient obtains.
A6, information push method according to a5, it is described according to determine the probability whether to user to be pushed push described in
Message further comprises:
The probability is compared with the push probability threshold value, user to be pushed is determined whether according to comparison result
Push the message.
A7, according to the described in any item information push methods of A1-A6, the model is two sorted logic regression models.
A8, according to the described in any item information push methods of A1-A7, the push attributive character of the message includes following letter
What is ceased is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, Yong Huhua
As feature and product version.
A9, information push method according to a1, it is described pushed user at the beginning server end preassign;
It is described whether the message is pushed to user to be pushed according to determine the probability after, the method also includes:
After pushing the message to user to be pushed, the data that the case where pushing messages to the user to be pushed is fed back are obtained,
It is stored in database for use as subsequent training sample data.
B10, a kind of message driving means, comprising:
Message processing module chooses the push attributive character of the message suitable for obtaining message to be pushed;
Sample extraction module obtains training sample suitable for being pushed to the data that the case where having pushed user feeds back according to message
Data, the training sample data have pushed user to the click data of the message including at least described;
Model training module is suitable for utilizing the training sample data training pattern, generates model parameter;
Message pushing module, suitable for according to the association attributes of the message and the determination of the association attributes of user to be pushed
The push attributive character parameter of message substitutes into the push attributive character parameter of the message and the model parameter in model,
Obtain whether to user's PUSH message to be pushed probability, according to the determine the probability whether to user to be pushed push described in disappear
Breath.
B11, message driving means according to b10, further includes: characteristic parameter determining module, be suitable for it is described
Before pushing user's PUSH message, the message is determined according to the association attributes of the message and the association attributes for having pushed user
Push attributive character parameter.
B12, the message driving means according to B11, the model training module are further adapted for: will be disappeared according to described
The push attributive character parameter and training sample for the message that the association attributes of breath and the association attributes for having pushed user determine
Data substitute into model, carry out optimization processing according to the optimization algorithm of model, generate model parameter.
B13, message driving means according to b12, the model training module are further adapted for: according to batch gradient
Descent algorithm carries out optimization processing.
B14, according to the described in any item message driving means of B10-B13, the model training module is further adapted for: raw
At the model parameter comprising model coefficient and the push probability threshold value obtained according to model coefficient.
B15, message driving means according to b14, the message pushing module are further adapted for: by the probability with
The push probability threshold value is compared, and determines whether that user to be pushed pushes the message according to comparison result.
B16, according to the described in any item message driving means of B10-B15, the model is two sorted logic regression models.
B17, according to the described in any item message driving means of B10-B16, the push attributive character of the message include with
Lower information it is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, use
Family Figure Characteristics and product version.
B18, message driving means according to b10, it is described to have pushed user and refer in advance in server end at the beginning
It is fixed;
Described device further include: database is suitable for storage in after user the to be pushed push message, acquired
The data that the case where pushing messages to the user to be pushed is fed back, for use as subsequent training sample data.
C19, a kind of message push system, including the described in any item message driving means of B10-B18, further includes: message
Manage platform and message task scheduler;
The message management platform is suitable for: customizing message according to message template, sends a message to message task scheduler;
The message task scheduler is suitable for: scheduling message forms push task, is sent to message pushing module and is pushed away
It send.
Claims (17)
1. a kind of information push method, comprising:
The push task of message task scheduler is received, the message to be pushed as push task is obtained, disappears described in selection
The push attributive character of breath;
The data that the case where having pushed user feeds back, which are pushed to, according to message obtains training sample data, the training sample data
User has been pushed to the click data of the message including at least described;Wherein, described to have pushed user and servicing at the beginning
Device end preassigns;
Using the training sample data training pattern, model parameter is generated;
The push attributive character ginseng of the message is determined according to the association attributes of the association attributes of the message and user to be pushed
Whether number substitutes into the push attributive character parameter of the message and the model parameter in model, obtain to user to be pushed
Whether the probability of PUSH message to user to be pushed pushes the message according to the determine the probability;
After pushing the message to user to be pushed, obtain what the case where pushing messages to the user to be pushed was fed back
Data are stored in database for use as subsequent training sample data.
2. information push method according to claim 1, described to be pushed to the case where having pushed user feedback according to message
Data obtain training sample data and further comprise:
To it is described pushed user's PUSH message before, according to the association attributes of the message and pushed user it is related category
Property determines the push attributive character parameter of the message;
User's push message is pushed to described, the data fed back are as training sample data.
3. information push method according to claim 2, described to utilize training sample data training pattern, model ginseng is generated
Number further comprises:
It will be special according to the push attribute of the association attributes of the message and the message for the association attributes determination for having pushed user
It levies parameter and training sample data substitutes into model, carry out optimization processing according to the optimization algorithm of model, generate model parameter.
4. information push method according to claim 3, the optimization algorithm according to model optimize handle into
One step includes: to carry out optimization processing according to batch gradient descent algorithm.
5. information push method according to claim 1-4, the model parameter includes model coefficient and root
The push probability threshold value obtained according to model coefficient.
It is described that whether institute to user to be pushed is pushed according to determine the probability 6. information push method according to claim 5
Stating message further comprises:
The probability is compared with the push probability threshold value, determines whether that user to be pushed pushes according to comparison result
The message.
7. information push method according to claim 1-4, the model is two sorted logic regression models.
8. information push method according to claim 1-4, the push attributive character of the message includes following
Information it is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, user
Figure Characteristics and product version.
9. a kind of message driving means, comprising:
Message processing module is obtained as push task suitable for receiving the push task of message task scheduler wait push
Message, choose the push attributive character of the message;
Sample extraction module obtains number of training suitable for being pushed to the data that the case where having pushed user feeds back according to message
According to the training sample data have pushed user to the click data of the message including at least described;Wherein, described to have pushed
User preassigns in server end at the beginning;
Model training module is suitable for utilizing the training sample data training pattern, generates model parameter;
Message pushing module, suitable for determining the message according to the association attributes of the message and the association attributes of user to be pushed
Push attributive character parameter, the push attributive character parameter of the message and the model parameter are substituted into model, obtained
Whether to user's PUSH message to be pushed probability, whether the message to user to be pushed is pushed according to the determine the probability;
Database is suitable for storage in after pushing the message to user to be pushed, it is acquired push messages to it is described to
The data that the case where pushing user is fed back, for use as subsequent training sample data.
10. message driving means according to claim 9, further includes: characteristic parameter determining module, be suitable for it is described
Before pushing user's PUSH message, the message is determined according to the association attributes of the message and the association attributes for having pushed user
Push attributive character parameter.
11. message driving means according to claim 10, the model training module are further adapted for: will be according to described
The push attributive character parameter and training sample for the message that the association attributes of message and the association attributes for having pushed user determine
Notebook data substitutes into model, carries out optimization processing according to the optimization algorithm of model, generates model parameter.
12. message driving means according to claim 11, the model training module are further adapted for: according to batch ladder
Degree descent algorithm carries out optimization processing.
13. the model training module is further adapted for according to claim 9-12 described in any item message driving means: raw
At the model parameter comprising model coefficient and the push probability threshold value obtained according to model coefficient.
14. message driving means according to claim 13, the message pushing module is further adapted for: by the probability
It is compared with the push probability threshold value, determines whether that user to be pushed pushes the message according to comparison result.
15. the model is two sorted logic regression models according to claim 9-12 described in any item message driving means.
16. according to the described in any item message driving means of claim 9-12, the push attributive character of the message include with
Lower information it is one or more: type of message, message pattern, regional information, temporal information, the clicking rate of same type message, use
Family Figure Characteristics and product version.
17. a kind of message push system, including the described in any item message driving means of claim 9-16, further includes: message
Manage platform and message task scheduler;
The message management platform is suitable for: customizing message according to message template, sends a message to message task scheduler;
The message task scheduler is suitable for: scheduling message forms push task, is sent to message pushing module and is pushed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610235874.4A CN105915438B (en) | 2016-04-15 | 2016-04-15 | Information push method, apparatus and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610235874.4A CN105915438B (en) | 2016-04-15 | 2016-04-15 | Information push method, apparatus and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105915438A CN105915438A (en) | 2016-08-31 |
CN105915438B true CN105915438B (en) | 2019-02-19 |
Family
ID=56747141
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610235874.4A Active CN105915438B (en) | 2016-04-15 | 2016-04-15 | Information push method, apparatus and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105915438B (en) |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107800746B (en) * | 2016-09-07 | 2022-01-21 | 百度在线网络技术(北京)有限公司 | Message pushing method and device |
CN106446195A (en) * | 2016-09-29 | 2017-02-22 | 北京百度网讯科技有限公司 | News recommending method and device based on artificial intelligence |
CN108229994A (en) * | 2016-12-21 | 2018-06-29 | 北京金山云网络技术有限公司 | A kind of information-pushing method and device |
CN108241616B (en) * | 2016-12-23 | 2023-07-25 | 阿里巴巴集团控股有限公司 | Message pushing method and device |
CN108259546A (en) * | 2017-01-16 | 2018-07-06 | 广州市动景计算机科技有限公司 | Information push method, equipment and programmable device |
CN108429666B (en) * | 2017-02-13 | 2021-06-15 | 江苏侨拓信息科技有限公司 | Self-media management system |
CN106982250B (en) * | 2017-03-03 | 2020-06-02 | 北京小米移动软件有限公司 | Information pushing method and device |
CN106982256B (en) * | 2017-03-31 | 2020-07-28 | 百度在线网络技术(北京)有限公司 | Information pushing method, device, equipment and storage medium |
CN107659609B (en) * | 2017-07-26 | 2021-01-12 | 北京天云融创软件技术有限公司 | Deep learning support platform based on cloud computing and deep learning training method |
CN107766580A (en) * | 2017-11-20 | 2018-03-06 | 北京奇虎科技有限公司 | The method for pushing and device of message |
CN108470023A (en) * | 2018-01-18 | 2018-08-31 | 阿里巴巴集团控股有限公司 | The recommendation method and device of business function |
CN109167816B (en) * | 2018-08-03 | 2021-11-16 | 广州虎牙信息科技有限公司 | Information pushing method, device, equipment and storage medium |
CN110855731A (en) * | 2018-08-21 | 2020-02-28 | 阿里巴巴集团控股有限公司 | Message processing, pushing and model training method, device, equipment and storage medium |
CN109460158A (en) * | 2018-10-29 | 2019-03-12 | 维沃移动通信有限公司 | Characters input method, character correction model training method and mobile terminal |
CN109889430A (en) * | 2019-01-23 | 2019-06-14 | 平安科技(深圳)有限公司 | Information push method, device, computer equipment and storage medium |
CN110532467B (en) * | 2019-08-22 | 2022-06-07 | 上海易点时空网络有限公司 | Activity recommendation method and device based on push model, equipment and storage medium |
CN110598016B (en) * | 2019-09-11 | 2021-08-17 | 腾讯科技(深圳)有限公司 | Method, device, equipment and medium for recommending multimedia information |
CN111385190B (en) * | 2020-03-06 | 2023-02-17 | 杭州威佩网络科技有限公司 | Message sending method and device, electronic equipment and storage medium |
CN112182381B (en) * | 2020-09-28 | 2022-09-13 | 上海嗨普智能信息科技股份有限公司 | Data processing method, electronic device, and medium |
CN112507237A (en) * | 2020-12-28 | 2021-03-16 | 山东茶急送茶业有限公司 | Intelligent algorithm for performing accurate pushing according to user data in retail APP |
CN116501972B (en) * | 2023-05-06 | 2024-01-05 | 广州市巨应信息科技有限公司 | Content pushing method and AI intelligent pushing system based on big data online service |
CN117009672B (en) * | 2023-10-08 | 2024-01-09 | 江西科技学院 | Activity recommendation method and system based on big data |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
CN104281622A (en) * | 2013-07-11 | 2015-01-14 | 华为技术有限公司 | Information recommending method and information recommending device in social media |
CN104572734A (en) * | 2013-10-23 | 2015-04-29 | 腾讯科技(深圳)有限公司 | Question recommendation method, device and system |
-
2016
- 2016-04-15 CN CN201610235874.4A patent/CN105915438B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
CN104281622A (en) * | 2013-07-11 | 2015-01-14 | 华为技术有限公司 | Information recommending method and information recommending device in social media |
CN104572734A (en) * | 2013-10-23 | 2015-04-29 | 腾讯科技(深圳)有限公司 | Question recommendation method, device and system |
Also Published As
Publication number | Publication date |
---|---|
CN105915438A (en) | 2016-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105915438B (en) | Information push method, apparatus and system | |
CN104065565B (en) | The method of PUSH message, server, client terminal device and system | |
US20190138586A1 (en) | Adaptive Modification of Content Presented in Electronic Forms | |
CN103493089B (en) | Method and apparatus for tracking the feeding in social networks | |
US9064212B2 (en) | Automatic event categorization for event ticket network systems | |
US8301420B2 (en) | Method and apparatus for creating a representation of a product or process | |
US20060168250A1 (en) | System and method for facilitating personalization of applications based on anticipation of users' interests | |
CN110288049A (en) | Method and apparatus for generating image recognition model | |
CN106062806B (en) | Generating relationships and promoting content using interaction signals | |
CN108021708B (en) | Content recommendation method and device and computer readable storage medium | |
CN108038161A (en) | Information recommendation method, device and computing device based on photograph album | |
CN109872796A (en) | A kind of menu recommended method and device | |
CN103577531A (en) | Message subscription system and method based on message aggregate characteristics | |
CN108377273A (en) | A kind of information-pushing method, server and device | |
CN109241455B (en) | Recommended object display method and device | |
CN109040317A (en) | A kind of information push method, client, server and system | |
CN107633433A (en) | The checking method and device of advertisement | |
CN104615452A (en) | Information providing method and device | |
CN108365989A (en) | Event-handling method and device | |
CN108009147A (en) | Electronics book cover generation method, electronic equipment and computer-readable storage medium | |
CN112085562A (en) | Information display method and device, electronic equipment and computer readable medium | |
CN104239421A (en) | Method and system for pushing application to terminal | |
CN109409419B (en) | Method and apparatus for processing data | |
CN113723926A (en) | Bank pipelining processing method and device combining RPA and AI and electronic equipment | |
CN105787379A (en) | Information management method and system as well as electronic device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220725 Address after: Room 801, 8th floor, No. 104, floors 1-19, building 2, yard 6, Jiuxianqiao Road, Chaoyang District, Beijing 100015 Patentee after: BEIJING QIHOO TECHNOLOGY Co.,Ltd. Address before: 100088 room 112, block D, 28 new street, new street, Xicheng District, Beijing (Desheng Park) Patentee before: BEIJING QIHOO TECHNOLOGY Co.,Ltd. Patentee before: Qizhi software (Beijing) Co.,Ltd. |