CN106709755A - Method of predicting user frequency and apparatus thereof - Google Patents
Method of predicting user frequency and apparatus thereof Download PDFInfo
- Publication number
- CN106709755A CN106709755A CN201611070600.0A CN201611070600A CN106709755A CN 106709755 A CN106709755 A CN 106709755A CN 201611070600 A CN201611070600 A CN 201611070600A CN 106709755 A CN106709755 A CN 106709755A
- Authority
- CN
- China
- Prior art keywords
- users
- sample
- user
- class
- subscriber
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/955—Retrieval from the web using information identifiers, e.g. uniform resource locators [URL]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Finance (AREA)
- Databases & Information Systems (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method of predicting a user frequency and an apparatus thereof. The method comprises the following steps of setting at least one user type and setting a corresponding relation of the user type and the user frequency; setting a target prediction model corresponding to each user type; acquiring characteristic data of a target user; according to the characteristic data of the target user and each target prediction model, predicting a target user type where a target user belongs; according to the corresponding relation, determining the user frequency corresponding to the target user type; and taking the user frequency corresponding to the target user type as the user frequency of the target user. The invention provides the method of predicting the user frequency and the apparatus thereof and the user frequency can be predicted.
Description
Technical field
The present invention relates to network technique field, more particularly to a kind of method and device for predicting user's frequency.
Background technology
As internet occupies the increasing proportion of people's daily life, Internet advertising has also obtained hair at full speed
Exhibition.If used as the standard of differentiation, Internet advertising can be divided into the purpose from advertisement:Brand advertising and effect advertisement.Effect is wide
It is to promote consumer behavior to accuse main purpose, and evaluation index is usually CPA (Cost Per Action, each cost of activities) index:
Click, download, registration, phone, on-line consulting, or purchase etc.;Brand advertising main purpose is to set up brand recognition, lifting
Brand influence, evaluation index is usually:TA (Target Audiences, target audience) and N+UV (user view, user
The frequency), user's frequency is primarily referred to as user's access times within a certain period of time.For example:User was broadcast in one day using youku.com
Put the number of times that device plays video, user's number of times of browsing objective website etc. in a week.
If the user's frequency in a period of time can be predicted, user can be flowed according to the user's frequency for predicting
Amount carries out optimal allocation, carries out advertisement putting, can greatly improve the income of customer flow.The unpredictable user of prior art is frequently
It is secondary.
The content of the invention
A kind of method and device for predicting user's frequency is the embodiment of the invention provides, user's frequency can be predicted.
On the one hand, a kind of method for predicting user's frequency is the embodiment of the invention provides, including:
At least one class of subscriber is set, the corresponding relation of the class of subscriber and user's frequency is set;
The corresponding target prediction model of each described class of subscriber is set;
Including:
Obtain the characteristic of targeted customer;
Characteristic and each described target prediction model according to the targeted customer, predict belonging to the targeted customer
Targeted customer's classification;
According to the corresponding relation, the corresponding user's frequency of targeted customer's classification is determined;
Using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency.
Further, the corresponding target prediction model of described each class of subscriber of setting, including:
The corresponding initial predicted model of each described class of subscriber is set;
Obtain the characteristic and user's frequency of sample of users;
User's frequency and the corresponding relation according to each sample of users, determine belonging to each described sample of users
Class of subscriber;
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, to each
The initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber.
Further, it is described that the corresponding initial predicted model of each described class of subscriber is set, including:
The corresponding initial predicted function of each described class of subscriber is set, wherein, active user's classification it is corresponding it is described just
Beginning anticipation function is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is used for current sample
The characteristic vector at family, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h
X () is the probability that the current sample of users belongs to active user's classification, belonging to described in the current sample of users works as
During preceding class of subscriber, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
Class of subscriber belonging to the characteristic according to each sample of users and each described sample of users, it is right
Each described initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that often
Each feature weight in the individual initial predicted function;
According to each feature weight in the corresponding initial predicted function of each described class of subscriber, each institute is determined
The corresponding target prediction function of class of subscriber is stated, wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiFor active user's classification is corresponding initial pre-
The ith feature weight surveyed in function, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,
yn), yjIt is the ith feature data of the targeted customer, H (y) is that the targeted customer belongs to the general of active user's classification
Rate;
The characteristic according to the targeted customer and each described target prediction model, predict the targeted customer
Affiliated targeted customer's classification, including:
Characteristic and each described target prediction function according to the targeted customer, determine each described target prediction
The corresponding prediction probability of function;
Using the corresponding class of subscriber of target prediction probability maximum in each described prediction probability as the targeted customer
Classification.
Further, the characteristic according to each sample of users and the use belonging to each described sample of users
Family classification, determines each feature weight in each described initial predicted function, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that
Each feature weight of described current initial predicted function when the weight of current initial predicted function determines that parameter takes maximum,
Wherein,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)It is k-th characteristic vector of sample of users, when k-th sample of users belongs to the current initial predicted function pair
During the class of subscriber answered, z(k)=1, when k-th sample of users is not belonging to the corresponding user of the current initial predicted function
During classification, z(k)=0.
Further, the characteristic, including:User cookie, access time, access times, attribute tags, login
Annual distribution, access are spaced apart, the average of access time, the variance of access time, the average of access times, access times
One or more in variance.
Further, it is described using the corresponding user's frequency of targeted customer's classification as the targeted customer user
After the frequency, further include:
The control frequency of user's frequency, at least one advertiser according to the targeted customer is required and default principle, for institute
The advertisement that targeted customer delivers at least one advertiser is stated, wherein, the default principle includes:Meet most advertisers
Control frequency require that the control of advertiser frequency is required to include:The advertisement of the advertiser at least browses pre- by the targeted customer
If value time.
On the other hand, a kind of device for predicting user's frequency is the embodiment of the invention provides, including:
First setting unit, for setting at least one class of subscriber, sets the class of subscriber right with user's frequency
Should be related to
Second setting unit, for setting the corresponding target prediction model of each described class of subscriber;
Target Acquisition unit, the characteristic for obtaining targeted customer;
Predicting unit, for the characteristic according to the targeted customer and each described target prediction model, predicts institute
State the targeted customer's classification belonging to targeted customer;
Frequency determining unit, for according to the corresponding relation, determining the corresponding user's frequency of targeted customer's classification,
Using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency.
Further, second setting unit, including:
Subelement is set, for setting the corresponding initial predicted model of each described class of subscriber;
Sample acquisition subelement, characteristic and user's frequency for obtaining sample of users;
Sample class determination subelement, for user's frequency and the corresponding relation according to each sample of users,
Determine the class of subscriber belonging to each described sample of users;
Training subelement, for belonging to the characteristic according to each sample of users and each described sample of users
Class of subscriber, is trained to initial predicted model each described, obtains the corresponding target prediction mould of each described class of subscriber
Type.
Further, the setting subelement, for setting the corresponding initial predicted function of each described class of subscriber, its
In, the corresponding initial predicted function of active user's classification is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is used for current sample
The characteristic vector at family, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h
X () is the probability that the current sample of users belongs to active user's classification, belonging to described in the current sample of users works as
During preceding class of subscriber, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
The training subelement, for the characteristic according to each sample of users and each described sample of users institute
The class of subscriber of category, determines each feature weight in each described initial predicted function;According to each class of subscriber pair
Each feature weight in the initial predicted function answered, determines the corresponding target prediction function of each described class of subscriber,
Wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiFor active user's classification is corresponding initial pre-
The ith feature weight surveyed in function, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,
yn), yjIt is the ith feature data of the targeted customer, H (y) is that the targeted customer belongs to the general of active user's classification
Rate;
The predicting unit, for the characteristic according to the targeted customer and each described target prediction function, really
The corresponding prediction probability of fixed each described target prediction function, by target prediction probability pair maximum in each described prediction probability
The class of subscriber answered is used as targeted customer's classification.
Further, the training subelement, in the execution characteristic according to each sample of users and respectively
Class of subscriber belonging to the individual sample of users, when determining each feature weight in each described initial predicted function, is used for
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that current initial
Each feature weight of described current initial predicted function when the weight of anticipation function determines that parameter takes maximum, wherein,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)It is k-th characteristic vector of sample of users, when k-th sample of users belongs to the current initial predicted function pair
During the class of subscriber answered, z(k)=1, when k-th sample of users is not belonging to the corresponding user of the current initial predicted function
During classification, z(k)=0.
Further, the characteristic, including:User cookie, access time, access times, attribute tags, login
Annual distribution, access are spaced apart, the average of access time, the variance of access time, the average of access times, access times
One or more in variance.
Further, the device is further included:
Advertisement putting unit, the control frequency for the user's frequency according to the targeted customer, at least one advertiser is required
With default principle, the advertisement of at least one advertiser is delivered for the targeted customer, wherein, the default principle bag
Include:The control frequency for meeting most advertisers requires that the control frequency of the advertiser is required to include:The minimum quilt of the advertisement of the advertiser
The targeted customer browses preset value.
In embodiments of the present invention, the corresponding target prediction model of each class of subscriber is set, by the feature of targeted customer
Data input determines targeted customer's classification of targeted customer in each target prediction model, by targeted customer's classification correspondence
User's frequency as user's frequency of targeted customer, realize the prediction to user's frequency of targeted customer.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are the present invention
Some embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis
These accompanying drawings obtain other accompanying drawings.
Fig. 1 is a kind of flow chart of the method for prediction user frequency that one embodiment of the invention is provided;
Fig. 2 is the flow chart of the method for another prediction user's frequency that one embodiment of the invention is provided;
Fig. 3 is the flow chart of the method for another prediction user's frequency that one embodiment of the invention is provided;
Fig. 4 is a kind of schematic diagram of the device of prediction user frequency that one embodiment of the invention is provided;
Fig. 5 is the schematic diagram of the device of another prediction user's frequency that one embodiment of the invention is provided.
Specific embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention, rather than whole embodiments, based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained on the premise of creative work is not made, belongs to the scope of protection of the invention.
As shown in figure 1, the embodiment of the invention provides a kind of method for predicting user's frequency, the method can include following
Step:
Step 101:At least one class of subscriber is set, the corresponding relation of the class of subscriber and user's frequency is set;
Step 102:The corresponding target prediction model of each described class of subscriber is set;
Step 103:Obtain the characteristic of targeted customer;
Step 104:Characteristic and each described target prediction model according to the targeted customer, predict the target
Targeted customer's classification belonging to user;
Step 105:According to the corresponding relation, the corresponding user's frequency of targeted customer's classification is determined;
Step 106:Using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency.
In embodiments of the present invention, the corresponding target prediction model of each class of subscriber is set, by the feature of targeted customer
Data input determines targeted customer's classification of targeted customer in each target prediction model, by targeted customer's classification correspondence
User's frequency as user's frequency of targeted customer, realize the prediction to user's frequency of targeted customer.
In order to user's frequency is more accurately predicted, in an embodiment of the present invention, described each class of subscriber pair of setting
The target prediction model answered, including:
The corresponding initial predicted model of each described class of subscriber is set;
Obtain the characteristic and user's frequency of sample of users;
User's frequency and the corresponding relation according to each sample of users, determine belonging to each described sample of users
Class of subscriber;
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, to each
The initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber.
In embodiments of the present invention, according to the characteristic and class of subscriber of sample of users, each is initial pre- to default
Survey model to be trained, obtain the corresponding target prediction model of each initial predicted model, goal forecast model is base
Obtained in the characteristic training of sample of users, when the prediction of user's frequency of targeted customer is carried out, can obtained more accurate
True predicts the outcome.
In order to further predict user's frequency exactly, in an embodiment of the present invention, the spy for obtaining sample of users
Data and user's frequency are levied, including:
Obtain characteristic of the sample of users within the cycle very first time, and the user within the second time cycle
The frequency, wherein, the cycle very first time is before second time cycle, and the cycle very first time and described second
It is separated by the preset value time cycle between time cycle;
The characteristic for obtaining targeted customer, including:
Characteristic of the targeted customer in targeted time period is obtained, wherein, the targeted time period is being treated
Before the predicted time cycle, and when being separated by the preset value between the targeted time period and the time cycle to be predicted
Between the cycle;
It is described using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency, including:
Using the corresponding user's frequency of targeted customer's classification as the targeted customer in the time cycle to be predicted
User's frequency.
In embodiments of the present invention, multiple time cycles are marked off in advance, when initial predicted model is trained, using first
The user's frequency in characteristic and the second time cycle in time cycle trains initial predicted model, for the second time
For cycle, the characteristic in the cycle very first time is historical data, for the time cycle to be predicted, week object time
The characteristic of phase is historical data, and the time interval in the second time cycle and the cycle very first time is equal to the time to be predicted
The time interval of cycle and targeted time period, for train the data of initial predicted model with for predicting the spy of targeted customer
Levy data has similitude in time, therefore, it is possible to improve the accuracy of prediction.For example, in training initial predicted mould
During type, using sample of users in the characteristic in January and user's frequency in March, in order to improve the accuracy of prediction, pre-
When surveying user's frequency of targeted customer, because the time cycle to be predicted is August part, data during in order to training have in time
There is similitude, treatment is predicted in the characteristic in June using targeted customer, the two is identical in time interval.
In an embodiment of the present invention, the characteristic and user's frequency for obtaining sample of users, including:
User access logses are obtained from destination server, the sample of users is extracted from the user access logses
Characteristic and user's frequency;
The characteristic for obtaining targeted customer, including:
The user access logses of the targeted customer are obtained from the destination server, from the user of the targeted customer
The characteristic of the targeted customer is extracted in access log.
In embodiments of the present invention, destination server can be the server of network player, for example:The service of youku.com
The server of device, server of Tengxun's video etc., or software, for example:The server of wechat, the server of Taobao's software
Deng, or website server, for example:Sina website, Tengxun website etc..From the user access logses in destination server
In extract the characteristic and user's frequency of sample of users, train the target prediction model for obtaining can to use by these data
Targeted customer in destination server carries out the prediction of user's frequency.
For example, in order to deliver advertisement in Tengxun's video, user access logses are obtained from the server of Tengxun's video,
Sample of users is extracted from user access logses in the characteristic in January and user's frequency in March, is instructed by these data
Get each target prediction model;The user access logses of targeted customer are obtained from the server of Tengxun's video, is used from target
Characteristic of the targeted customer in June is extracted in the user access logses at family, it is pre- using these characteristics and each target
Model is surveyed, prediction targeted customer can deliver in August part according to user's frequency in user's frequency of August part for targeted customer
Advertisement.For example:It is 4 times that targeted customer is predicted in user's frequency of August part, that is, predicts targeted customer in August part meeting
Using 4 Tengxun's videos, there are 2 advertisers to need to deliver advertisement in Tengxun's video, each advertiser requires advertisement by extremely
See 2 times less, at this moment, the advertisement 2 times of each advertiser in August part, can be respectively delivered for the targeted customer.
In an embodiment of the present invention, the characteristic, including:User cookie, access time, access times, category
Property label, login time distribution, access be spaced apart, the average of access time, the variance of access time, access times it is equal
Value, one or more in the variance of access times.
In embodiments of the present invention, each user can be identified by user cookie, attribute tags can include:With
The information such as sex, age, the location at family.These characteristics can be obtained by cleaning daily record, and for cleaning daily record
The preliminary characteristic for obtaining afterwards, determines the statistical indicator of each preliminary characteristic, and by the system of each preliminary characteristic
Meter index and preliminary characteristic are used as characteristic.Here statistical indicator includes:Average, variance etc..
In an embodiment of the present invention, described using the corresponding user's frequency of targeted customer's classification as the target
After user's frequency of user, further include:
The control frequency of user's frequency, at least one advertiser according to the targeted customer is required and default principle, for institute
The advertisement that targeted customer delivers at least one advertiser is stated, wherein, the default principle includes:Meet most advertisers
Control frequency require that the control of advertiser frequency is required to include:The advertisement of the advertiser at least browses pre- by the targeted customer
If value time.
In embodiments of the present invention, after the user's frequency for predicting targeted customer, the user's frequency that will be predicted is used for
Advertisement is delivered for the targeted customer, when advertisement is delivered, is delivered according to default principle, can realized to user's frequency
Optimal allocation, can maximize the income of customer flow.
For example, there are three advertisers, be respectively advertiser A, advertiser B and advertiser C, their control frequency requires to divide
It is not:Control frequency is for 3, control frequency is 1, control frequency is 2, wherein, the control frequency of advertiser refers to that at least see n times this is wide for requirement user for a
Accuse main advertisement.It is 4 times in user's frequency of Tengxun's video to predict targeted customer within next week by the embodiment of the present invention,
That is, user can use Tengxun's video 4 times within next week.According to default principle, carry out delivering extensively for targeted customer
The mode of announcement is:Deliver the advertisement of advertiser A and the advertisement of advertiser B.So disclosure satisfy that the control frequency of 2 advertisers is required.
If do not delivered according to default principle, dispensing advertisement can be in the following manner carried out:Deliver the advertisement of advertiser A and wide
Accuse the advertisement of main C, in this case, because the two is needed altogether by targeted customer's flow 5 times, therefore, can only meet one it is wide
Master is accused, the income of the customer flow than being delivered according to default principle is low.
As shown in Fig. 2 the embodiment of the invention provides a kind of method for predicting user's frequency, the method can include following
Step:
Step 201:At least one class of subscriber is set, the corresponding relation of class of subscriber and user's frequency is set.
For example, user's frequency is access times of the user in one month, such as use of the user in one month
The access times of Tengxun's video.4 class of subscribers are set, and the corresponding relation is:The corresponding user's frequency of class of subscriber 1 is less than or equal to
2 times;The corresponding user's frequency of class of subscriber 2 is equal to 3 times;The corresponding user's frequency of class of subscriber 3 is equal to 4 times;4 pairs of applications of class of subscriber
The family frequency is more than or equal to 5 times.
Step 202:The corresponding initial predicted model of each class of subscriber is set.
Step 203:Obtain the characteristic and user's frequency of sample of users.
Specifically, the characteristic and user's frequency of multiple sample of users are obtained.
Characteristic, including:User cookie, access time, access times, attribute tags, login time distribution, access
Be spaced apart, one in the average of access time, the variance of access time, the average of access times, the variance of access times
Or it is multiple.
For example, user access logses are obtained from the server of Tengxun's video, sample is extracted from user access logses
User is in the characteristic in January and user's frequency in March.Here sample of users is the user of Tengxun's video.Here
User access logses can from the user of Tengxun's video randomly selected multiple users user access logses.
Step 204:User's frequency and corresponding relation according to each sample of users, determine the use belonging to each sample of users
Family classification.
For example, user's frequency of sample of users A is 3, then sample of users belongs to class of subscriber 2;Sample of users B's
User's frequency is 7, then sample of users belongs to class of subscriber 4.
Step 205:The class of subscriber belonging to characteristic and each sample of users according to each sample of users, to each
Initial predicted model is trained, and obtains the corresponding target prediction model of each class of subscriber.
For example, by after training, the corresponding target prediction model 1 of class of subscriber 1, the corresponding target of class of subscriber 2
Forecast model 2, the corresponding target prediction model 3 of class of subscriber 3, the corresponding target prediction model 4 of class of subscriber 4.
Step 206:Obtain the characteristic of targeted customer.
For example, the user access logses of targeted customer are obtained from the server of Tengxun's video, from targeted customer's
User access logses extract characteristic of the sample of users in June.Here sample of users is the user of Tengxun's video.
Step 207:Characteristic and each target prediction model according to targeted customer, the mesh belonging to prediction targeted customer
Mark class of subscriber.
For example, the characteristic in June of targeted customer is input in each target prediction model, obtains every
Individual target prediction mode input predicts the outcome, and it can be targeted customer's input current goal forecast model correspondence that this predicts the outcome
Active user's classification probability, this is, the corresponding class of subscriber of target prediction model of the maximum probability that will be input into is used as mesh
Mark class of subscriber.For example:What target prediction model 1, target prediction model 2, target prediction model 3, target prediction model 4 were exported
Probability is respectively:0.5、0.4、0.7、0.2.So, the corresponding class of subscriber 3 of target prediction model 3 is targeted customer's classification,
That is, targeted customer belongs to class of subscriber 3.
Step 208:According to corresponding relation, the corresponding user's frequency of targeted customer's classification is determined.
For example, when targeted customer's classification is class of subscriber 3, according to corresponding relation, the corresponding user of class of subscriber 3
The frequency is equal to 4 times.
Step 209:Using the corresponding user's frequency of targeted customer's classification as targeted customer user's frequency.
For example, the corresponding user's frequency of class of subscriber 3 is equal to 4 user's frequencys as targeted customer, also
It is to say, predicts targeted customer and the access times of Tengxun's video are used within the time cycle to be predicted for 4 times, here to be predicted
Time cycle can be August part.
Furthermore it is also possible to including:The control frequency of user's frequency, at least one advertiser according to targeted customer is required and default
Principle, the advertisement of at least one advertiser is delivered for targeted customer, wherein, default principle includes:Meet most advertisers
Control frequency require that the control of advertiser frequency is required to include:The advertisement of advertiser at least browses preset value by the targeted customer.
For example, for user's frequency of targeted customer, when targeted customer uses Tengxun's video, for targeted customer
Advertisement is delivered, specifically, 4 advertisements is delivered to targeted customer, specifically, can be delivered according to default principle.
In order to further provide for the accuracy of prediction, in an embodiment of the present invention, described each described user class of setting
Not corresponding initial predicted model, including:
The corresponding initial predicted function of each described class of subscriber is set, wherein, active user's classification it is corresponding it is described just
Beginning anticipation function is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is used for current sample
The characteristic vector at family, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h
X () is the probability that the current sample of users belongs to active user's classification, belonging to described in the current sample of users works as
During preceding class of subscriber, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
Class of subscriber belonging to the characteristic according to each sample of users and each described sample of users, it is right
Each described initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that often
Each feature weight in the individual initial predicted function;
According to each feature weight in the corresponding initial predicted function of each described class of subscriber, each institute is determined
The corresponding target prediction function of class of subscriber is stated, wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiFor active user's classification is corresponding initial pre-
The ith feature weight surveyed in function, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,
yn), yjIt is the ith feature data of the targeted customer, H (y) is that the targeted customer belongs to the general of active user's classification
Rate;
The characteristic according to the targeted customer and each described target prediction model, predict the targeted customer
Affiliated targeted customer's classification, including:
Characteristic and each described target prediction function according to the targeted customer, determine each described target prediction
The corresponding prediction probability of function;
Using the corresponding class of subscriber of target prediction probability maximum in each described prediction probability as the targeted customer
Classification.
In embodiments of the present invention, the initial predicted function of each class of subscriber can with identical, but, obtain after training
The corresponding target prediction function of each initial predicted function be probably different.When corresponding relation is set, each user class
The scope of user's frequency or user's frequency can not corresponded to.For example, the corresponding user's frequency of class of subscriber 1 is small
In equal to 2 times;The corresponding user's frequency of class of subscriber 2 is equal to 3 times;The corresponding user's frequency of class of subscriber 3 is equal to 4 times;Class of subscriber 4
The correspondence user frequency is more than or equal to 5 times.When being trained, user's frequency of sample of users A is 3 times, the feature of sample of users A
Data vector is XA.It is initial pre- because sample of users A is not belonging to class of subscriber 1 for the initial predicted function 1 of class of subscriber 1
The h (x) surveyed in function 1 is X for 0, xA;Similarly, for class of subscriber 3 initial predicted function 3 and class of subscriber 4 it is initial pre-
Function 4 is surveyed, h (x) is X for 0, xA.For the initial predicted function 2 of class of subscriber 1, because sample of users A belongs to class of subscriber
2, h (x) in initial predicted function 2 are X for 1, xA.It is also using same assignment for other sample of users.Namely
Say, for same x, the h (x) in the initial predicted function of different class of subscribers is probably different, so, in training
Afterwards, the target prediction function of different class of subscribers is probably different.In training, each feature of each initial predicted function
Weight is unknown, it is necessary to pass through training acquisition.After each feature weight for determining each initial predicted function, will determine
Go out feature weight to be updated in corresponding initial predicted function, you can obtain the corresponding target prediction function of each class of subscriber.
For target prediction function, feature weight is known, and H (y) is the result obtained according to characteristic.For example,
By after training, the corresponding target prediction function 1 of class of subscriber 1, the corresponding target prediction function 2 of class of subscriber 2, class of subscriber
3 corresponding target prediction functions 3, the corresponding target prediction function 4 of class of subscriber 4;For the characteristic of targeted customer, target
Anticipation function 1, target prediction function 2, target prediction function 3, the H (y) of the output of target prediction function 4 are respectively:0.5、0.4、
0.7th, 0.2, then, the H (y) of the output of target prediction function 3 is maximum, and the corresponding class of subscriber 3 of target prediction function 3 is used for target
Family classification, that is to say, that targeted customer belongs to class of subscriber 3.
In embodiments of the present invention, initial predicted model includes:Initial predicted function, target prediction model includes:Target
Anticipation function,
It is in an embodiment of the present invention, described to be used according to each described sample in order to further provide for the accuracy of prediction
The characteristic at family and the class of subscriber belonging to each described sample of users, determine each in each described initial predicted function
Feature weight, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that
Each feature weight of described current initial predicted function when the weight of current initial predicted function determines that parameter takes maximum,
Wherein,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)It is k-th characteristic vector of sample of users, when k-th sample of users belongs to the current initial predicted function pair
During the class of subscriber answered, z(k)=1, when k-th sample of users is not belonging to the corresponding user of the current initial predicted function
During classification, z(k)=0.
In embodiments of the present invention, parameter is determined by the weight of each initial predicted function, it may be determined that go out more preferably
Feature weight, more accurate target prediction function can be determined by more preferably feature weight, and then causes to predict the outcome
It is more accurate.
As shown in figure 3, the embodiment of the invention provides a kind of method for predicting user's frequency, the method can include following
Step:
Step 301:At least one class of subscriber is set, the corresponding relation of class of subscriber and user's frequency is set.
For example, user's frequency is access times of the user in one month, such as use of the user in one month
The access times of Tengxun's video.4 class of subscribers are set, and the corresponding relation is:The corresponding user's frequency of class of subscriber 1 is less than or equal to
2 times;The corresponding user's frequency of class of subscriber 2 is equal to 3 times;The corresponding user's frequency of class of subscriber 3 is equal to 4 times;4 pairs of applications of class of subscriber
The family frequency is more than or equal to 5 times.
Step 302:The corresponding initial predicted function of each class of subscriber is set, wherein, active user's classification is corresponding just
Beginning anticipation function is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is used for current sample
The characteristic vector at family, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h
X () is the probability that the current sample of users belongs to active user's classification, belonging to described in the current sample of users works as
During preceding class of subscriber, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0.
For example, the corresponding initial predicted function 1 of class of subscriber 1;The corresponding initial predicted function 2 of class of subscriber 2;User class
Other 3 corresponding initial predicted function 3;The corresponding initial predicted function 4 of class of subscriber 4.
Step 303:Obtain the characteristic and user's frequency of sample of users.
Specifically, the categorized data set of standard can according to the characteristic of sample of users and user's frequency, be generated.Citing
For, to be concentrated in the grouped data of standard, each sample of users one element of correspondence, the element is x, and x is the spy of sample of users
Data vector is levied, wherein, the characteristic vector of each sample of users is set according to default form, for example, x1It is spy
Levy data 1, x2It is characterized data 2, xjIt is characterized data j, xnData n is characterized, the like.For example:x1It is sample of users
Access time, x2Access times, x for sample of users3Average, x for the access time of sample of users4It is the visit of sample of users
Ask the average of number of times.
Step 304:User's frequency and corresponding relation according to each sample of users, determine the use belonging to each sample of users
Family classification.
Step 305:The class of subscriber belonging to characteristic and each sample of users according to each sample of users, it is determined that
Each feature weight of each initial predicted function when the weight of each initial predicted function determines that parameter takes maximum, its
In,
Wherein, l (θ) is that the weight of current initial predicted function determines parameter, and m is the quantity of sample of users, x(k)It is kth
The characteristic vector of individual sample of users, when k-th sample of users belongs to the corresponding class of subscriber of current initial predicted function,
z(k)=1, when k-th sample of users is not belonging to the corresponding class of subscriber of current initial predicted function, z(k)=0.
Step 306:According to each feature weight in the corresponding initial predicted function of each class of subscriber, determine that each is used
The corresponding target prediction function of family classification, wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiIt is the corresponding initial predicted letter of active user's classification
Ith feature weight in number, y is the characteristic vector of targeted customer, y=(y1,y2,...,yj,...,yn), yjIt is target
The ith feature data of user, H (y) belongs to the probability of active user's classification for targeted customer.
For example, the corresponding target prediction function 1 of class of subscriber 1, the corresponding target prediction function 2 of class of subscriber 2 is used
The corresponding target prediction function 3 of family classification 3, the corresponding target prediction function 4 of class of subscriber 4.
Step 307:Obtain the characteristic of targeted customer.
Specifically, the characteristic vector of targeted customer can be generated according to the characteristic of targeted customer, wherein, target
The characteristic vector of user is set according to default form, and the default form is identical with the characteristic vector of sample of users, lifts
For example, y1It is characterized data 1, y2It is characterized data 2, yjIt is characterized data j, ynData n is characterized, the like.For example:y1
Access time, y for targeted customer2Access times, y for targeted customer3Average, y for the access time of targeted customer4For
The average of the access times of targeted customer.
Step 308:Characteristic and each target prediction function according to targeted customer, determine each target prediction function
Corresponding prediction probability.
For example, for the characteristic of targeted customer, target prediction function 1, target prediction function 2, target prediction
Function 3, the H (y) of the output of target prediction function 4 are respectively:0.5th, 0.4,0.7,0.2,
Step 309:Using the corresponding class of subscriber of target prediction probability maximum in each prediction probability as targeted customer
Classification.
For example, the H (y) of the output of target prediction function 3 is maximum, and the corresponding class of subscriber 3 of target prediction function 3 is mesh
Mark class of subscriber, that is to say, that targeted customer belongs to class of subscriber 3.
Step 310:According to corresponding relation, the corresponding user's frequency of targeted customer's classification is determined.
For example, according to corresponding relation, determine that the corresponding user's frequency of class of subscriber 3 is equal to 4 times.
Step 311:Using the corresponding user's frequency of targeted customer's classification as targeted customer user's frequency.
For example, because the corresponding user's frequency of targeted customer's classification is 4 times, so user's frequency of targeted customer is
4 times.
As shown in figure 4, a kind of device for predicting user's frequency is the embodiment of the invention provides, including:
First setting unit 401, for setting at least one class of subscriber, sets the class of subscriber with user's frequency
Corresponding relation
Second setting unit 402, for setting the corresponding target prediction model of each described class of subscriber;
Target Acquisition unit 403, the characteristic for obtaining targeted customer;
Predicting unit 404, for the characteristic according to the targeted customer and each described target prediction model, prediction
Targeted customer's classification belonging to the targeted customer;
Frequency determining unit 405, for according to the corresponding relation, determining the corresponding user of targeted customer's classification frequently
It is secondary, using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency.
As shown in figure 5, the accuracy in order to improve prediction, in an embodiment of the present invention, second setting unit, bag
Include:
Subelement 501 is set, for setting the corresponding initial predicted model of each described class of subscriber;
Sample acquisition subelement 502, characteristic and user's frequency for obtaining sample of users;
Sample class determination subelement 503, closes for the user's frequency according to each sample of users and the correspondence
System, determines the class of subscriber belonging to each described sample of users;
Training subelement 504, for the characteristic according to each sample of users and each described sample of users institute
The class of subscriber of category, is trained to initial predicted model each described, obtains the corresponding target of each described class of subscriber pre-
Survey model.
Specifically, sample class determination subelement is connected with the first setting unit;Training subelement is connected with predicting unit.
In order to further improve the accuracy of prediction, in an embodiment of the present invention, the setting subelement, for setting
The corresponding initial predicted function of each described class of subscriber, wherein, the corresponding initial predicted function of active user's classification is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is used for current sample
The characteristic vector at family, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h
X () is the probability that the current sample of users belongs to active user's classification, belonging to described in the current sample of users works as
During preceding class of subscriber, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
The training subelement, for the characteristic according to each sample of users and each described sample of users institute
The class of subscriber of category, determines each feature weight in each described initial predicted function;According to each class of subscriber pair
Each feature weight in the initial predicted function answered, determines the corresponding target prediction function of each described class of subscriber,
Wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiFor active user's classification is corresponding initial pre-
The ith feature weight surveyed in function, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,
yn), yjIt is the ith feature data of the targeted customer, H (y) is that the targeted customer belongs to the general of active user's classification
Rate;
The predicting unit, for the characteristic according to the targeted customer and each described target prediction function, really
The corresponding prediction probability of fixed each described target prediction function, by target prediction probability pair maximum in each described prediction probability
The class of subscriber answered is used as targeted customer's classification.
In an embodiment of the present invention, the training subelement, is performing the spy according to each sample of users
Data and the class of subscriber belonging to each described sample of users are levied, each feature power in each described initial predicted function is determined
During weight, for the class of subscriber belonging to the characteristic according to each sample of users and each described sample of users, it is determined that
Each feature power of the described current initial predicted function when the weight of current initial predicted function determines that parameter takes maximum
Weight, wherein,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)It is k-th characteristic vector of sample of users, when k-th sample of users belongs to the current initial predicted function pair
During the class of subscriber answered, z(k)=1, when k-th sample of users is not belonging to the corresponding user of the current initial predicted function
During classification, z(k)=0.
In an embodiment of the present invention, the characteristic, including:User cookie, access time, access times, category
Property label, login time distribution, access be spaced apart, the average of access time, the variance of access time, access times it is equal
Value, one or more in the variance of access times;
In an embodiment of the present invention, the device is further included:
Advertisement putting unit, the control frequency for the user's frequency according to the targeted customer, at least one advertiser is required
With default principle, the advertisement of at least one advertiser is delivered for the targeted customer, wherein, the default principle bag
Include:The control frequency for meeting most advertisers requires that the control frequency of the advertiser is required to include:The minimum quilt of the advertisement of the advertiser
The targeted customer browses preset value.
In an embodiment of the present invention, the sample acquisition subelement, accesses for obtaining user from destination server
Daily record, extracts the characteristic and user's frequency of the sample of users from the user access logses;
The Target Acquisition unit, the user for obtaining the targeted customer from the destination server accesses day
Will, extracts the characteristic of the targeted customer from the user access logses of the targeted customer.
In order to further predict user's frequency exactly, in an embodiment of the present invention, the sample acquisition subelement,
For obtaining characteristic of the sample of users within the cycle very first time, and user within the second time cycle is frequently
It is secondary, wherein, the cycle very first time before second time cycle, and the cycle very first time with described second when
Between be separated by the preset value time cycle between the cycle;
The Target Acquisition unit, for obtaining characteristic of the targeted customer in targeted time period, wherein,
The targeted time period before the time cycle to be predicted, and the targeted time period and the time cycle to be predicted it
Between be separated by the preset value time cycle;
The frequency determining unit, it is described using the corresponding user's frequency of targeted customer's classification as the mesh performing
Mark user user's frequency when, for using the corresponding user's frequency of targeted customer's classification as the targeted customer described
User's frequency of time cycle to be predicted.
The contents such as the information exchange between each unit, implementation procedure in said apparatus, due to implementing with the inventive method
Example is based on same design, and particular content can be found in the narration in the inventive method embodiment, and here is omitted.
Each embodiment of the invention at least has the advantages that:
1st, in embodiments of the present invention, the corresponding target prediction model of each class of subscriber is set, by the spy of targeted customer
Data input is levied in each target prediction model, targeted customer's classification of targeted customer is determined, by targeted customer's classification pair
The user's frequency answered realizes the prediction to user's frequency of targeted customer as user's frequency of targeted customer.
2nd, in embodiments of the present invention, according to the characteristic and class of subscriber of sample of users, each is initial to default
Forecast model is trained, and obtains the corresponding target prediction model of each initial predicted model, and goal forecast model is
Characteristic training based on sample of users is obtained, and when the prediction of user's frequency of targeted customer is carried out, can be obtained more
Accurately predict the outcome.
3rd, in embodiments of the present invention, multiple time cycles are marked off in advance, when initial predicted model is trained, using the
The user's frequency in characteristic and the second time cycle in a period of time trains initial predicted model, during for second
Between for the cycle, the characteristic in the cycle very first time is historical data, for the time cycle to be predicted, the object time
The characteristic in cycle is historical data, and the second time cycle with the time interval in the cycle very first time when being equal to be predicted
Between cycle and targeted time period time interval, for train the data of initial predicted model with for predicting targeted customer's
Characteristic has similitude in time, therefore, it is possible to improve the accuracy of prediction.
4th, in embodiments of the present invention, after the user's frequency for predicting targeted customer, the user's frequency that will be predicted is used
In advertisement is delivered for the targeted customer, when advertisement is delivered, delivered according to default principle, can be realized to user's frequency
Optimal allocation, the income of customer flow can be maximized.
It should be noted that herein, such as first and second etc relational terms are used merely to an entity
Or operation makes a distinction with another entity or operation, and not necessarily require or imply these entities or exist between operating
Any this actual relation or order.And, term " including ", "comprising" or its any other variant be intended to it is non-
It is exclusive to include, so that process, method, article or equipment including a series of key elements not only include those key elements,
But also other key elements including being not expressly set out, or also include by this process, method, article or equipment are solid
Some key elements.In the absence of more restrictions, the key element limited by sentence " including ", does not arrange
Except also there is other identical factor in the process including the key element, method, article or equipment.
One of ordinary skill in the art will appreciate that:Realizing all or part of step of above method embodiment can pass through
Programmed instruction related hardware is completed, and foregoing program can be stored in the storage medium of embodied on computer readable, the program
Upon execution, the step of including above method embodiment is performed;And foregoing storage medium includes:ROM, RAM, magnetic disc or light
Disk etc. is various can be with the medium of store program codes.
It is last it should be noted that:Presently preferred embodiments of the present invention is the foregoing is only, skill of the invention is merely to illustrate
Art scheme, is not intended to limit the scope of the present invention.All any modifications made within the spirit and principles in the present invention,
Equivalent, improvement etc., are all contained in protection scope of the present invention.
Claims (10)
1. it is a kind of predict user's frequency method, it is characterised in that
At least one class of subscriber is set, the corresponding relation of the class of subscriber and user's frequency is set;
The corresponding target prediction model of each described class of subscriber is set;
Including:
Obtain the characteristic of targeted customer;
Characteristic and each described target prediction model according to the targeted customer, predict the mesh belonging to the targeted customer
Mark class of subscriber;
According to the corresponding relation, the corresponding user's frequency of targeted customer's classification is determined;
Using the corresponding user's frequency of targeted customer's classification as the targeted customer user's frequency.
2. method according to claim 1, it is characterised in that
The setting corresponding target prediction model of each class of subscriber, including:
The corresponding initial predicted model of each described class of subscriber is set;
Obtain the characteristic and user's frequency of sample of users;
User's frequency and the corresponding relation according to each sample of users, determine the use belonging to each described sample of users
Family classification;
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, to described in each
Initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber.
3. method according to claim 2, it is characterised in that
It is described that the corresponding initial predicted model of each described class of subscriber is set, including:
The corresponding initial predicted function of each described class of subscriber is set, wherein, active user's classification is corresponding described initial pre-
Surveying function is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is current sample of users
Characteristic vector, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h(x)
Belong to the probability of active user's classification for the current sample of users, the current use is belonged in the current sample of users
During the classification of family, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
Class of subscriber belonging to the characteristic according to each sample of users and each described sample of users, to each
The initial predicted model is trained, and obtains the corresponding target prediction model of each described class of subscriber, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, determines each institute
State each feature weight in initial predicted function;
According to each feature weight in the corresponding initial predicted function of each described class of subscriber, each described use is determined
The corresponding target prediction function of family classification, wherein, the corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiIt is the corresponding initial predicted letter of active user's classification
Ith feature weight in number, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,yn), yjFor
The ith feature data of the targeted customer, H (y) is the probability that the targeted customer belongs to active user's classification;
The characteristic according to the targeted customer and each described target prediction model, predict belonging to the targeted customer
Targeted customer's classification, including:
Characteristic and each described target prediction function according to the targeted customer, determine each described target prediction function
Corresponding prediction probability;
Using the corresponding class of subscriber of target prediction probability maximum in each described prediction probability as targeted customer's classification.
4. method according to claim 3, it is characterised in that
Class of subscriber belonging to the characteristic according to each sample of users and each described sample of users, it is determined that often
Each feature weight in the individual initial predicted function, including:
The class of subscriber belonging to characteristic and each described sample of users according to each sample of users, it is determined that current
Each feature weight of described current initial predicted function when the weight of initial predicted function determines that parameter takes maximum, its
In,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)For
The characteristic vector of k-th sample of users, when k-th sample of users, to belong to the current initial predicted function corresponding
During class of subscriber, z(k)=1, when k-th sample of users is not belonging to the corresponding class of subscriber of the current initial predicted function
When, z(k)=0.
5. according to any described method in claim 1-4, it is characterised in that
The characteristic, including:User cookie, access time, access times, attribute tags, login time distribution, access
Be spaced apart, one in the average of access time, the variance of access time, the average of access times, the variance of access times
Or it is multiple;
And/or,
It is described using the corresponding user's frequency of targeted customer's classification as after user's frequency of the targeted customer, enter one
Step includes:
The control frequency of user's frequency, at least one advertiser according to the targeted customer is required and default principle, for the mesh
Mark user delivers the advertisement of at least one advertiser, wherein, the default principle includes:Meet the control of most advertisers
Frequency requires that the control frequency of the advertiser is required to include:The advertisement of the advertiser at least browses preset value by the targeted customer
It is secondary.
6. it is a kind of predict user's frequency device, it is characterised in that including:
First setting unit, for setting at least one class of subscriber, sets class of subscriber pass corresponding with user's frequency
System
Second setting unit, for setting the corresponding target prediction model of each described class of subscriber;
Target Acquisition unit, the characteristic for obtaining targeted customer;
Predicting unit, for the characteristic according to the targeted customer and each described target prediction model, predicts the mesh
Targeted customer's classification belonging to mark user;
Frequency determining unit, for according to the corresponding relation, determining the corresponding user's frequency of targeted customer's classification, by institute
State the user frequency of the corresponding user's frequency of targeted customer's classification as the targeted customer.
7. device according to claim 6, it is characterised in that
Second setting unit, including:
Subelement is set, for setting the corresponding initial predicted model of each described class of subscriber;
Sample acquisition subelement, characteristic and user's frequency for obtaining sample of users;
Sample class determination subelement, for user's frequency and the corresponding relation according to each sample of users, it is determined that
Class of subscriber belonging to each described sample of users;
Training subelement, for the user belonging to the characteristic according to each sample of users and each described sample of users
Classification, is trained to initial predicted model each described, obtains the corresponding target prediction model of each described class of subscriber.
8. device according to claim 7, it is characterised in that
The setting subelement, for setting the corresponding initial predicted function of each described class of subscriber, wherein, active user's class
The not corresponding initial predicted function is:
Wherein, θTX=θ0+θ1x1+,...,+θixi+,...,+θnxn, θiIt is ith feature weight, x is current sample of users
Characteristic vector, x=(x1,x2,...,xj,...,xn), xjIt is j-th characteristic of the current sample of users;h(x)
Belong to the probability of active user's classification for the current sample of users, the current use is belonged in the current sample of users
During the classification of family, h (x) is 1;When the current sample of users is not belonging to active user's classification, h (x) is 0;
The training subelement, for belonging to the characteristic according to each sample of users and each described sample of users
Class of subscriber, determines each feature weight in each described initial predicted function;It is corresponding according to each described class of subscriber
Each feature weight in the initial predicted function, determines the corresponding target prediction function of each described class of subscriber, wherein,
The corresponding target prediction function of active user's classification is:
Wherein, θTY=θ0+θ1y1+,...,+θiyi+,...,+θnyn, θiIt is the corresponding initial predicted letter of active user's classification
Ith feature weight in number, y is the characteristic vector of the targeted customer, y=(y1,y2,...,yj,...,yn), yjFor
The ith feature data of the targeted customer, H (y) is the probability that the targeted customer belongs to active user's classification;
The predicting unit, for the characteristic according to the targeted customer and each described target prediction function, it is determined that often
The corresponding prediction probability of the individual target prediction function, target prediction probability maximum in each described prediction probability is corresponding
Class of subscriber is used as targeted customer's classification.
9. device according to claim 8, it is characterised in that
The training subelement, is performing the characteristic according to each sample of users and each described sample of users
Affiliated class of subscriber, when determining each feature weight in each described initial predicted function, for according to each sample
Class of subscriber belonging to the characteristic of this user and each described sample of users, it is determined that in the weight of current initial predicted function
Each feature weight of described current initial predicted function when determining that parameter takes maximum, wherein,
Wherein, l (θ) is that the weight of the current initial predicted function determines parameter, and m is the quantity of the sample of users, x(k)For
The characteristic vector of k-th sample of users, when k-th sample of users, to belong to the current initial predicted function corresponding
During class of subscriber, z(k)=1, when k-th sample of users is not belonging to the corresponding class of subscriber of the current initial predicted function
When, z(k)=0.
10. according to any described device in claim 6-9, it is characterised in that
The characteristic, including:User cookie, access time, access times, attribute tags, login time distribution, access
Be spaced apart, one in the average of access time, the variance of access time, the average of access times, the variance of access times
Or it is multiple;
And/or,
Further include:
Advertisement putting unit, the control frequency for the user's frequency according to the targeted customer, at least one advertiser is required and pre-
If principle, the advertisement of at least one advertiser is delivered for the targeted customer, wherein, the default principle includes:It is full
The control frequency of the most advertiser of foot requires that the control frequency of the advertiser is required to include:The advertisement of the advertiser is at least described
Targeted customer browses preset value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611070600.0A CN106709755A (en) | 2016-11-28 | 2016-11-28 | Method of predicting user frequency and apparatus thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611070600.0A CN106709755A (en) | 2016-11-28 | 2016-11-28 | Method of predicting user frequency and apparatus thereof |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106709755A true CN106709755A (en) | 2017-05-24 |
Family
ID=58935016
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611070600.0A Pending CN106709755A (en) | 2016-11-28 | 2016-11-28 | Method of predicting user frequency and apparatus thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106709755A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107608792A (en) * | 2017-09-12 | 2018-01-19 | 中国联合网络通信集团有限公司 | Resource regulating method and device |
CN113987482A (en) * | 2021-12-28 | 2022-01-28 | 中孚信息股份有限公司 | IP first access detection method, system and equipment based on FM |
WO2023103527A1 (en) * | 2021-12-10 | 2023-06-15 | 深圳前海微众银行股份有限公司 | Access frequency prediction method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104216884A (en) * | 2013-05-29 | 2014-12-17 | 酷盛(天津)科技有限公司 | Collaborative filtering system and method on basis of time decay |
CN104602042A (en) * | 2014-12-31 | 2015-05-06 | 合一网络技术(北京)有限公司 | User behavior based label setting method |
CN105631538A (en) * | 2015-12-23 | 2016-06-01 | 北京奇虎科技有限公司 | User activity prediction method and device, and application method and system thereof |
CN105701498A (en) * | 2015-12-31 | 2016-06-22 | 腾讯科技(深圳)有限公司 | User classification method and server |
-
2016
- 2016-11-28 CN CN201611070600.0A patent/CN106709755A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104216884A (en) * | 2013-05-29 | 2014-12-17 | 酷盛(天津)科技有限公司 | Collaborative filtering system and method on basis of time decay |
CN104602042A (en) * | 2014-12-31 | 2015-05-06 | 合一网络技术(北京)有限公司 | User behavior based label setting method |
CN105631538A (en) * | 2015-12-23 | 2016-06-01 | 北京奇虎科技有限公司 | User activity prediction method and device, and application method and system thereof |
CN105701498A (en) * | 2015-12-31 | 2016-06-22 | 腾讯科技(深圳)有限公司 | User classification method and server |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107608792A (en) * | 2017-09-12 | 2018-01-19 | 中国联合网络通信集团有限公司 | Resource regulating method and device |
WO2023103527A1 (en) * | 2021-12-10 | 2023-06-15 | 深圳前海微众银行股份有限公司 | Access frequency prediction method and device |
CN113987482A (en) * | 2021-12-28 | 2022-01-28 | 中孚信息股份有限公司 | IP first access detection method, system and equipment based on FM |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109657138A (en) | A kind of video recommendation method, device, electronic equipment and storage medium | |
US8732015B1 (en) | Social media pricing engine | |
Tamaddoni Jahromi et al. | Modeling customer churn in a non-contractual setting: the case of telecommunications service providers | |
CN106803190A (en) | A kind of ad personalization supplying system and method | |
CN106997549A (en) | The method for pushing and system of a kind of advertising message | |
WO2014124310A1 (en) | A system for improving shape-based targeting by using interest level data | |
CN107146089A (en) | The single recognition methods of one kind brush and device, electronic equipment | |
CN106503006A (en) | The sort method and device of application App neutron applications | |
CN108062678A (en) | A kind of advertisement bit allocation method, device and advertisement delivery system | |
CN104035926B (en) | A kind of dispensing of internet information and system | |
US20140310098A1 (en) | System for Improving Shape-Based Targeting By Using Interest Level Data | |
CN106294830A (en) | The recommendation method and device of multimedia resource | |
CN108509583A (en) | A kind of information-pushing method, server and computer readable storage medium | |
Miklošík | Changes in purchasing decision-making process of consumers in the digital era | |
JP2002092305A (en) | Score calculating method, and score providing method | |
CN101464984A (en) | Advertisement publishing control system and method for publishing advertisement by using advertisement space tag | |
CN110097394A (en) | The latent objective recommended method of product and device | |
CN107895213A (en) | Forecasting Methodology, device and the electronic equipment of spending limit | |
CN108777701A (en) | A kind of method and device of determining receiver | |
CN106709755A (en) | Method of predicting user frequency and apparatus thereof | |
CN108446944A (en) | A kind of the determination method, apparatus and electronic equipment in resident city | |
CN107992500A (en) | A kind of information processing method and server | |
US8296180B1 (en) | System for improving shape-based targeting by using interest level data | |
CN110162609A (en) | For recommending the method and device asked questions to user | |
CN112132634A (en) | Virtual gift resource distribution method and device, computer equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |
|
RJ01 | Rejection of invention patent application after publication |