CN107301247A - Set up the method and device, terminal, storage medium of clicking rate prediction model - Google Patents
Set up the method and device, terminal, storage medium of clicking rate prediction model Download PDFInfo
- Publication number
- CN107301247A CN107301247A CN201710578583.XA CN201710578583A CN107301247A CN 107301247 A CN107301247 A CN 107301247A CN 201710578583 A CN201710578583 A CN 201710578583A CN 107301247 A CN107301247 A CN 107301247A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- user
- feature
- value
- 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.)
- Granted
Links
Classifications
-
- 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
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the present application discloses a kind of method and device, terminal, storage medium for setting up clicking rate prediction model, and methods described includes：The feature of multiple first users is gathered, first user is the user that object was recommended, and the object treats click on content for what is recommended；For each user in the multiple first user each feature-set to the evaluation index of the object, the clicking rate prediction model is built based on the evaluation index；According to the clicking rate prediction model, error function of each first user to the actual click value of the object with estimating clicks value is set up, error loss function is set up based on the error function；Based on the error loss function and each first user counted in advance to the actual click value of the object, the numerical value of the evaluation index set to the object is solved；The numerical value of the evaluation index to the object obtained according to solving determines the clicking rate prediction model.
Description
Technical field
The application is related to network technique field, more particularly to a kind of method and device for setting up clicking rate prediction model, end
End, storage medium.
Background technology
Network technical development is to today, increasing object, such as Domestic News, article, music, and picture etc. needs to push away
Recommend to user, and the effect that this recommendation will have been obtained, it is necessary to targetedly find recommended use for different objects
Family, carries out personalized recommendation, and the core technology difficult point of personalized recommendation is how to accurately determine object to be recommended.
The content of the invention
In view of the above problems, the purpose of the application is the provision of a kind of method and dress for setting up clicking rate prediction model
Put, terminal, storage medium, improve the accuracy to user's recommended.
On the one hand, the embodiment of the present application provides a kind of method for setting up clicking rate prediction model, including：
The feature of multiple first users is gathered, first user is the user that object was recommended, and the object is to push away
That recommends treats click on content；
For each user in the multiple first user each feature-set to the evaluation index of the object, be based on
The evaluation index builds the clicking rate prediction model；
According to the clicking rate prediction model, each first user is set up to the actual click value of the object and is estimated a little
The error function of value is hit, error loss function is set up based on the error function；
Based on the actual click value of the error loss function and each first user counted in advance to the object,
Solve the numerical value of the evaluation index set to the object；
The numerical value of the evaluation index to the object obtained according to solving determines the clicking rate prediction model.
Alternatively, methods described also includes：
Feature based on second user, the second user is obtained to the object using the clicking rate prediction model
The discreet value of clicking rate, the second user is the user that the object was not recommended.
After the feature for gathering multiple first users, in addition to：
Feature is sorted out, multiple characteristic sets will be divided into per category feature；
For each user in the multiple first user each feature-set to the evaluation index of the object after
Also include：
Identical numerical value is assigned to the identical evaluation index of the object for the feature in each characteristic set.
The evaluation index includes：Each feature estimates clicks value r and each feature pair to the object
The degree of reliability a for estimating clicks value of the object, wherein, r ∈ [0,1], a ∈ [0,1].
It is described based on the actual click value of the error loss function and each user counted in advance to the object,
Solving the numerical value of the evaluation index set to the object includes：
The initial value of the evaluation index is set；
Calculating is iterated to the error loss function by target of the loss reduction of the error loss function；
Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now described
The value of evaluation index as the evaluation index numerical value.
The clicking rate prediction model is：
The error function is：
The error loss function is：
Wherein, i represents object, and u represents user, and U represents the set of first user, ctru,iRepresent user u to object
I clicking rate discreet value, f represents user u feature, FuRepresent user u characteristic set；yu,iRepresent user u to object i's
Actual click value；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent estimate clicks values of the feature f to object i
The degree of reliability.
On the other hand, the application also provides a kind of device for setting up clicking rate prediction model, including：
Acquisition module, the feature for gathering multiple first users, first user is the user that object was recommended,
The object treats click on content for what is recommended；
First modeling module, for each feature-set for each user in the multiple first user to described right
The evaluation index of elephant, the clicking rate prediction model is built based on the evaluation index；
Second modeling module, for according to the clicking rate prediction model, setting up each first user to the object
Actual click value and the error function for estimating clicks value, error loss function is set up based on the error function；
Solve module, for based on the error loss function and each first user counted in advance to the object
Actual click value, solve to the object set evaluation index numerical value；
3rd modeling module, for determining the click according to the numerical value for solving the obtained evaluation index to the object
Rate prediction model.
Alternatively, described device also includes：
Computing module, for the feature based on second user, obtains described second using the clicking rate prediction model and uses
Discreet value of the family to the clicking rate of the object, the second user is the user that the object was not recommended.
Alternatively, the acquisition module, is additionally operable to be sorted out feature, will be divided into multiple feature sets per category feature
Close；
First modeling module, is additionally operable to the identical evaluation index to the object for the feature in each characteristic set
Assign identical numerical value.
The evaluation index includes：Each feature includes to the evaluation index of the object：Each feature is to described
Object estimates the degree of reliability a for estimating clicks value of clicks value r and each feature to the object, wherein, r ∈ [0,
1], a ∈ [0,1].
The solution module includes：
First solves submodule, the initial value for setting the evaluation index；
Second solves submodule, for losing letter to the error by target of the loss reduction of the error loss function
Number is iterated calculating；
3rd solves submodule, for stopping described changing when the rate of change of the error loss function is less than predetermined threshold value
In generation, is calculated and the numerical value of the evaluation index is used as using the value of the now evaluation index.
The clicking rate prediction model is：
The error function is：
The error loss function is：
Wherein, i represents object, and u represents user, and U represents the set of first user, ctru,iRepresent user u to object
I clicking rate discreet value, f represents user u feature, FuRepresent user u characteristic set；yu,iRepresent user u to object i's
Actual click value；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent estimate clicks values of the feature f to object i
The degree of reliability.
Another aspect, the application also provides a kind of terminal, including：The storage of processor and the computer instruction that is stored with
Device；
The processor reads the computer instruction, and performs a kind of foregoing clicking rate prediction model of setting up
Method.
On the other hand, the application also provides a kind of storage medium, and be stored with computer instruction, and the computer instruction is held
A kind of foregoing method for setting up clicking rate prediction model is realized during row.
A kind of method and device, terminal, storage medium for setting up clicking rate prediction model that the embodiment of the present application is provided, with
Based on the click behavior for the user that object i was recommended, the feature of each user is gathered, object i is commented with each feature
Valency index is that parameter sets up clicking rate prediction model.After each feature is determined to object i evaluation index, for one not by
Recommended object i user, as long as the characteristic set of the user is determined, you can this is estimated according to the clicking rate prediction model
Clicking rate of the user to object i.With reference to this method, object i can be recommended and estimate the higher user of clicking rate, this method is carried
The high degree of accuracy that object is pushed to user, in actual applications, reduces invalid recommendation, improves network utilization, can
It is embodied as different user and personalized object recommendation is provided.
Brief description of the drawings
According to following detailed descriptions carried out referring to the drawings, the above and other objects, features and advantages of the application will become
Obtain obviously.In the accompanying drawings：
The flow chart for the method for setting up clicking rate prediction model that Fig. 1 provides for the embodiment of the application one；
The flow chart for the method for hitting rate prediction model for setting up a web advertisement that Fig. 2 provides for the embodiment of the application one；
The schematic diagram of a scenario for the web advertisement clicking rate prediction model application that Fig. 3 provides for the embodiment of the application one；
The method flow diagram for setting up article clicking rate prediction model that Fig. 4 provides for the embodiment of the application one；
The schematic diagram of a scenario of the clicking rate prediction model application for the article that Fig. 5 provides for the embodiment of the application one；
The structural representation for the device for setting up clicking rate prediction model that Fig. 6 provides for the embodiment of the application one.
Embodiment
The various aspects of the application are described below.Teaching herein can be embodied in varied form, and
Any concrete structure disclosed herein, function or two kinds are only representational.Based on teaching herein, people in the art
Member is it is to be understood that one aspect disclosed herein can be independently of in any other aspect realization, and these aspects
Two or more aspects can combine in various manners.It is, for example, possible to use any number of the aspects set forth herein,
Realize device or put into practice method.Further, it is possible to use other mechanisms, function or except one or more sides described in this paper
It is not outside face or the 26S Proteasome Structure and Function of one or more aspects described herein, realizes this device or put into practice this side
Method.In addition, any aspect described herein can include at least one element of claim.
There is provided a kind of method and device, terminal, storage medium for setting up clicking rate prediction model in this application.Under
The embodiment of the application is described with reference to accompanying drawing for face.
Referring to Fig. 1, the embodiment of the application one provide the method for setting up clicking rate prediction model, including step 101 to
106。
Step 101：The feature of multiple first users is gathered, first user is the user that object was recommended.
As it was previously stated, in embodiments herein, the object treats click on content for what is recommended, such as can be that network is wide
Announcement, article, using, music or picture, film etc..
So that music is recommended as an example, some music has been recommended to numerous users, then these were recommended the use of the music
Family is the first user of the music.The spy for the user that the music was recommended can be obtained by the feature for gathering the first user
Levy.
In the present embodiment, the feature of first user can include age, educational background, place city, occupation or income
Deng polytype.
In embodiments herein, after the feature for gathering multiple first users, it can also include：
Feature is sorted out, multiple characteristic sets will be divided into per category feature.
For every category feature, multiple different characteristic sets can be divided into some way.By taking the age as an example, such as
" children ", " teenager ", " youth ", " middle age ", " old age " etc. can be divided into, " after 60 ", " after 70 ", " 80 can also be divided into
Afterwards ", " after 90s ", " after 00 " etc..
Dividing mode per category feature can determine that the application is not construed as limiting according to actual needs.
Step 102：For evaluation of each feature-set to the object of each user in the multiple first user
Index, the clicking rate prediction model is built based on the evaluation index.
In the embodiment of the application one, the evaluation index to the object can include：Each feature is to the pre- of object
Estimate the degree of reliability a for estimating clicks value of clicks value r and each feature to object；Wherein, r ∈ [0,1], a ∈ [0,1].
I.e. in the embodiment of the present application, by estimating the clicks value r and degree of reliability a for estimating clicks value, the two refer to each feature
Mark to evaluate an object.
When feature is sorted out, and when being divided into multiple characteristic sets per category feature, the embodiment of the present application is provided
Method also include：
Identical numerical value is assigned to the identical evaluation index of the object for the feature in each characteristic set.
By taking the feature of this type of age as an example, this feature be further divided into " children ", " teenager ", " youth ", " in
Year ", " old age " multiple characteristic sets, then value phase of the feature to the evaluation index r and a of same object in each set
Together.
Still so that music is recommended as an example, a recommended is music i, if the feature pair in " teenager " this characteristic set
The music i evaluation index r and a value are respectively r1And a1, then any one user, if its age this feature category
In " teenager " this characteristic set, then this feature of the age of user is r to music i evaluation index1And a1。
In the embodiment of the application one, the clicking rate prediction model can be：
Wherein, ctr represents clicking rate discreet value；U represents user；I represents object；F represents the feature of user；FuRepresent to use
The characteristic set at family；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent that feature f estimates clicks value to object i can
By degree.
By the clicking rate prediction model, it is recognised that a user u is by the use to some object i clicking rate
Each feature in the characteristic set F at family estimates clicks value r and the reliable journey for estimating clicks value to object i to object i
Degree a is determined jointly.
If having " youth ", " university student ", " permanent Guangzhou ", " scientific and technological fan " this 4 spies in user A characteristic set F
Levy, this 4 features are respectively a to certain object i two evaluation indexes1And r1、a2And r2、a3And r3And a4And r4, then user A
To object i clicking rate discreet value ctrA,i=(a1*r1+a2*r2+a3*r3+a4*r4)/(a1+a2+a3+a4).I.e. as the spy of user
Collection is closed after F determinations, if what each feature in the characteristic set F of user was to determine to object i evaluation index, then
To estimate clicking rate of the user to the object to object i evaluation index according to each feature.
Step 103：According to the clicking rate prediction model, actual click value of each first user to the object is set up
Error function with estimating clicks value, error loss function is set up based on the error function.
Step 104：Based on the reality of the error loss function and each first user counted in advance to the object
Border clicks value, solves the numerical value of the evaluation index set to the object.
In the embodiment of the application one, based on the error loss function and each user counted in advance to the object
Actual click value and estimate the difference of clicks value, solving the numerical value of the evaluation index set to the object can include：
The initial value of the evaluation index is set；
Calculating is iterated to the error loss function by target of the loss reduction of the error loss function；
Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now described
The value of evaluation index as the evaluation index numerical value.
In the embodiment of the present application, it is according to the object i multiple user u being recommended true clicks value and estimates clicks value
To determine each feature to the evaluation index of the object.
In the embodiment of the application one, if clicking rate prediction model is as shown in Equation 1, then corresponding error function is：
Error loss function is：
Wherein, u represents user；I represents object；F represents the feature of user；ctru,iRepresent clicks of the user u to object i
Rate discreet value, FuRepresent user u characteristic set；yu,iRepresent actual click values of the user u to object i；rf,iRepresent f pairs of feature
Object i estimates clicks value；af,iRepresent that feature f estimates the degree of reliability of clicks value to object i, U represents what object i was recommended
The set of user.
If user's point u has hit object i, yu,i=1, otherwise yu,i=0.
According to formula 3 it is recognised that the embodiment of the present application be feature based on multiple users being recommended to object i and
The analysis of actual click value determines each feature f evaluation index a to object if,iAnd rf,i。
In the embodiment of the present application, gradient descent method, the actual click based on the object i users being recommended can be used
Value, with error L, (r, a) minimum target solve evaluation indexes of each feature f to object i.The method for solving can be included such as
Lower step：
1st step：Vectorial r, a of the random given one group decimal composition between 0-1, are set to r(0),a(0), initialize iteration step
Number k=0；
2nd step：Iterative calculation
Wherein θ is the step-length of iteration, takes 0.01
3rd step：Judge whether the error loss function restrains
ΔL(r(k+1), a(k+1))=| L (r(k+1), a(k+1))-L(r(k), a(k))|
If | Δ L (r(k+1), a(k+1))-ΔL(r(k), a(k)) | ＜ α, then be returned to r(k+1), a(k+1)The as ginseng of model
Number, otherwise returns to step 2 and continues to calculate, wherein α is the value of a very little, can take the θ of α=0.01
Step 105：The numerical value of the evaluation index to the object obtained according to solving determines that the clicking rate estimates mould
Type.
Calculated by above-mentioned steps and evaluation index as of each feature f to object i is determinedF, iAnd rF, i, so that object i
Clicking rate prediction model is determined that.
Step 106：Feature based on second user, the second user is obtained to institute using the clicking rate prediction model
The discreet value of the clicking rate of object is stated, the second user is the user that the object was not recommended.
It is true according to the feature of collection by gathering the feature of the user for a user B for not being recommended object i
Make corresponding evaluation index, you can clicking rate discreet values of the user B to object i is calculated according to the clicking rate prediction model.
The method for setting up clicking rate prediction model that the embodiment of the present application is provided, with the point of the object i users being recommended
Hit based on behavior, gather the feature of each user, clicking rate is set up as parameter to object i evaluation index using each feature pre-
Estimate model.After each feature is determined to object i evaluation index, for a user for not being recommended object i, as long as
The characteristic set of the user is determined, you can clicking rate of the user to object i is estimated according to the clicking rate prediction model.With reference to
This method, object i can be recommended estimate the higher user of clicking rate, be the method increase and pushed the accurate of object to user
Degree, in actual applications, reduces invalid recommendation, improves network utilization, can be implemented as different user and provides personalized
Object recommendation.
Referring to Fig. 2, in one embodiment of the application, the object is web advertisement c, foundation provided herein
The method of clicking rate prediction model includes step 201 to 206.
Step 201：The web advertisement quilt is portrayed in the feature for the user that the multiple web advertisements of collection were recommended, foundation
The characteristic set of the user recommended.
In the embodiment of the present application, the net is set up by gathering the feature for the user that the web advertisement c was recommended
Network advertisement c clicking rate prediction model.
In the embodiment of the present application, the spy for portraying the user that the web advertisement was recommended can be gathered from multiple dimensions
Collection is closed.
Dimension 1：Portrayed by the preference to the web advertisement of user, the user for example enjoyed shopping portrays " shopping fan ".
Dimension 2：Portrayed by the Regional Property of user, for example Beijing, Tianjin, Shanghai.
Dimension 3：Portrayed by user's natural quality, such as age, sex.
Dimension 4：Portrayed by the social property of user, such as educational level, occupation, region.
In practical application, according to the difference of object, selected dimension can also be different.The application is not construed as limiting to this.
To every category feature, different characteristic sets can be further divided into.
Step 202：For each characteristic set feature-set to the evaluation index of the web advertisement, evaluated based on described
Index builds the clicking rate prediction model.
In the embodiment of the present application, the evaluation index to the web advertisement can include：The feature of each feature set
The feature for estimating clicks value r and each characteristic set to the web advertisement estimates clicks value to the web advertisement
Degree of reliability a；Wherein, r ∈ [0,1], a ∈ [0,1].
It should be noted that the feature of each characteristic set is identical to the value of the evaluation index of the web advertisement.
By taking the age as an example, if the feature of this type of age further comprise " children ", " teenager ", " youth ", " middle age ",
" old age " 5 characteristic sets, then the feature in each characteristic set has identical evaluation index to the same web advertisement.
For example, the user of 30 years old and 32 years old belongs to " youth " this characteristic set, then the year of the user of 30 years old and the user of 32 years old
This feature of age is identical to the evaluation index value of the same web advertisement.
The clicking rate prediction model for the web advertisement set up in the embodiment of the present application is as shown in Equation 4.
Wherein, c represents the web advertisement, and u represents user, and f represents the feature of user, ctrU, c,Represent user u to net
Network advertisement c clicking rate discreet values, FuRepresent user u characteristic set, rF, cRepresent that feature f estimates clicks value to web advertisement c；
aF, cRepresent that feature f estimates the degree of reliability of clicks value to web advertisement c.
Step 203：According to the web advertisement clicking rate prediction model, the use that each web advertisement was recommended is set up
Error function of the family to the actual click value of the web advertisement with estimating clicks value, sets up error based on the error function and damages
Lose function.
Error function in the embodiment of the present application is as shown in Equation 5：
Wherein, yU, cRepresent actual click values of the user u to web advertisement c.If user u clicks web advertisement c,
yU, c=1, otherwise yU, c=0.
Step 204：Based on the reality of the error loss function and each user counted in advance to the web advertisement
Border clicks value and the difference for estimating clicks value, solve the numerical value of the evaluation index set to the object.
The error loss function is as shown in Equation 6：
Wherein, (r, a) represents error loss function to L, and U represents the set for the user u that the web advertisement c was recommended.
From formula 6 as can be seen that the embodiment of the present application is the user being recommended with reference to multiple web advertisement c to described
Web advertisement c actual click value and the feature f of each user determine that evaluation of each feature to the web advertisement c refers to
Target aF, cAnd rF, c's.
Gradient descent method, the actual click value y based on the web advertisement c user u being recommended can be usedU, c, with
(r, a) minimum target solve evaluations of each feature f of each user u in user's set U to the web advertisement c to error L
Index.The method for solving may include steps of：
1st step：Vectorial r, a of the random given one group decimal composition between 0-1, are set to r(0), a(0), initialize iteration step
Number k=0；
2nd step：Iterative calculation
Wherein θ is the step-length of iteration, takes 0.01
3rd step：Judge whether the error loss function restrains
ΔL(r(k+1), a(k+1))=| L (r(k+1), a(k+1))-L(r(k), a(k))|
If | Δ L (r(k+1), a(k+1))-ΔL(r(k), a(k)) | ＜ α, then be returned to r(k+1), a(k+1)The as ginseng of model
Number, otherwise returns to step 2 and continues to calculate, wherein α is the value of a very little, can take the θ of α=0.01.
Step 205：The numerical value of the evaluation index to the web advertisement obtained according to solving determines the web advertisement
Clicking rate prediction model.
Calculated by above-mentioned steps and evaluation index as of each feature f to the web advertisement c is determinedF, cAnd rF, c, so that
The clicking rate prediction model of the web advertisement is determined that.
Step 206：The corresponding evaluation index of feature for the user that the web advertisement was not recommended inputs the point
The rate prediction model of hitting obtains discreet value of the user to the clicking rate of the web advertisement.
Fig. 3 is the application scenarios schematic diagram to the web advertisement clicking rate prediction model.Institute was not recommended for one
Web advertisement c user B is stated, by gathering the n feature of the user, the corresponding evaluation index of each feature is respectively a1And r1、
a2And r2、a3And r3……anAnd rn, the corresponding evaluation index of each feature determined by above-mentioned modeling process.Then use
Clicking rate discreet value ctrs of the family B to the web advertisementB,c=(a1*r1+a2*r2+a3*r3+……+an*rn)/(a1+a2+a3
+……+an)。
The point of the web advertisement is set up by the feature and actual click value for the user being recommended based on the web advertisement
Rate prediction model is hit, and the clicking rate for calculating the user that the web advertisement was not recommended using the clicking rate prediction model is pre-
Valuation, then therefrom selects the high user of clicking rate discreet value and sends the web advertisement to these users, then can greatly carry
The degree of accuracy that the high web advertisement is delivered, it is to avoid the dispensing of the advertisement of low clicking rate, has saved advertisement putting cost, has improved
The economic benefit of the web advertisement.
Fig. 4 is the application that the clicking rate prediction model that the application is provided estimates aspect in article clicking rate, and this method includes
Step 401 is to 406.
Step 401：The feature for multiple users that the article was recommended is gathered, the feature set for portraying multiple users is set up
Close.
In the embodiment of the present application, the point of the article is set up by gathering the feature for the user that the article was recommended
Hit rate prediction model.
The determination mode of the characteristic set of user is similar to embodiment before, and here is omitted.
Step 402：For each characteristic set feature-set to the evaluation index of the article, based on the evaluation index
Build the clicking rate prediction model of the article.
In the embodiment of the present application, the evaluation index to the article can include：The feature of each feature set is to institute
State the degree of reliability a that estimates clicks value of the feature for estimating clicks value r and each characteristic set to the article of article；
Wherein, r ∈ [0,1], a ∈ [0,1].
It should be noted that the feature of each characteristic set is identical to the value of the evaluation index of the article.
The clicking rate prediction model of this article set up in the embodiment of the present application is as shown in Equation 7.
Wherein, d represents the article, and u represents user, and f represents feature；ctrU, dRepresent clicking rates of the user u to article d
Discreet value, FuRepresent user u characteristic set, rF, dRepresent that feature f estimates clicks value to article d；aF, dRepresent feature f to text
Chapter d estimates the degree of reliability of clicks value.
Step 403：According to the clicking rate prediction model of the article, actual click of each user to the article is set up
It is worth and estimates the error function of clicks value, error loss function is set up based on the error function.
Error function in the embodiment of the present application is as shown in Equation 8：
Wherein, yU, dActual click values of the user u to article d is represented, if user's point u clicks this article d, yU, d=1,
Otherwise yU, d=0.
Step 404：Based on the actual point of the error loss function and each user counted in advance to the article
Hit value and estimate the difference of clicks value, solve the numerical value of the evaluation index set to the object.
The error loss function is as shown in Equation 9.
Wherein, (r, a) represents error loss to L, and U represents the set for the user u that the article d was recommended.
From formula 9 as can be seen that the embodiment of the present application is the user being recommended with reference to multiple article d to the article
D actual click value and the feature of each user determine a of each feature to the evaluation index of the article dF, dAnd rF, d
's.
Gradient descent method, the actual click value y based on the article d user u being recommended can be usedU, d, with error
(r, a) minimum target solve evaluation indexes of each feature f in characteristic set to the article d to L.The method for solving can
To comprise the following steps：
1st step：Vectorial r, a of the random given one group decimal composition between 0-1, are set to r(0), a(0), initialize iteration step
Number k=0；
2nd step：Iterative calculation
Wherein θ is the step-length of iteration, takes 0.01
3rd step：Judge whether the error loss function restrains
ΔL(r(k+1), a(k+1))=| L (r(k+1), a(k+1))-L(r(k), a(k))|
If | Δ L (r(k+1), a(k+1))-ΔL(r(k), a(k)) | ＜ α, then be returned to r(k+1), a(k+1)The as ginseng of model
Number, otherwise returns to step 2 and continues to calculate, wherein α is the value of a very little, can take the θ of α=0.01.
Step 405：The numerical value of the evaluation index to the article obtained according to solving determines the clicking rate of the article
Prediction model.
Calculated by above-mentioned steps and evaluation index as of each feature f to the article d is determinedF, dAnd rF, d, so that described
The clicking rate prediction model of article is determined that.
Step 406：The corresponding evaluation index of feature for the user that the article was not recommended inputs the clicking rate
Prediction model obtains discreet value of the user to the clicking rate of the article.
Fig. 5 is the application scenarios schematic diagram to this article clicking rate prediction model.The text was not recommended for one
Chapter d user M, by gathering 4 features (by taking 4 features as an example) of the user, the corresponding evaluation index difference of each feature
For a1And r1、a2And r2、a3And r3And a4And r4, the corresponding evaluation index of each feature passes through above-mentioned modeling process
It is determined that.Then clicking rate discreet value ctrs of the user M to the article dM,d=(a1*r1+a2*r2+a3*r3+a4*r4)/(a1+a2+a3+
a4)。
The clicking rate for setting up this article d by feature and actual click value based on this article d users being recommended is pre-
Estimate model, and the clicking rate discreet value for the user that this article d was not recommended is calculated using the clicking rate prediction model, then
Therefrom select the high user of clicking rate discreet value and send this article d to these users, then can be greatly enhanced the article d and push away
The degree of accuracy recommended, is greatly enhanced the amount of reading of article.
A kind of structural representation for device for setting up clicking rate prediction model that Fig. 6 provides for the embodiment of the present application.Due to
Device embodiment is substantially similar to embodiment of the method, and the relevent part can refer to the partial explaination of embodiments of method.It is described below
Device embodiment be only schematical.
The device for setting up clicking rate prediction model of the application offer includes：
Acquisition module 601, the feature for gathering multiple first users, first user is the use that object was recommended
Family；
First modeling module 602, for each feature-set for each user in the multiple first user to institute
The evaluation index of object is stated, the clicking rate prediction model is built based on the evaluation index；
Second modeling module 603, for according to the clicking rate prediction model, setting up each first user to the object
Actual click value and estimate the error function of clicks value, error loss function is set up based on the error function；
Solve module 604, for based on the error loss function and each first user counted in advance to described
The actual click value of object, solves the numerical value of the evaluation index set to the object；
3rd modeling module 605, described in being determined according to the numerical value for solving the obtained evaluation index to the object
Clicking rate prediction model.
In the embodiment of the application one, described device also includes：
Computing module 606, for the feature based on second user, described second is obtained using the clicking rate prediction model
Discreet value of the user to the clicking rate of the object, the second user is the user that the object was not recommended.
In the embodiment of the application one, the acquisition module 601 is additionally operable to be sorted out feature, will be divided per category feature
For multiple characteristic sets；
First modeling module 602, is additionally operable to the identical evaluation to the object for the feature in each characteristic set
Index assigns identical numerical value.
In the embodiment of the present application, the evaluation index includes：Each feature includes to the evaluation index of the object：It is described
Each feature estimates the degree of reliability for estimating clicks value of clicks value r and each feature to the object to the object
A, wherein, r ∈ [0,1], a ∈ [0,1].
In the embodiment of the application one, the solution module 604 includes：
First solves submodule, the initial value for setting the evaluation index；
Second solves submodule, for being entered using the loss reduction of the error loss function as target to the loss function
Row iteration is calculated；
3rd solves submodule, for stopping described changing when the rate of change of the error loss function is less than predetermined threshold value
In generation, is calculated and the numerical value of the evaluation index is used as using the value of the now evaluation index.
In the embodiment of the application one, the clicking rate prediction model is：
The error function is：
The error loss function is：
Wherein, i represents object, and u represents user, and U represents the set of first user, ctru,iRepresent user u to object
I clicking rate discreet value, f represents user u feature, FuRepresent user u characteristic set；yu,iRepresent user u to object i's
Actual click value；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent estimate clicks values of the feature f to object i
The degree of reliability.
A kind of device for setting up clicking rate prediction model that the embodiment of the present application is provided, the user being recommended with object i
Click behavior based on, gather the feature of each user, click set up for parameter to object i evaluation index using each feature
Rate prediction model.After each feature is determined to object i evaluation index, for a user for not being recommended object i,
As long as the characteristic set of the user is determined, you can estimate clicking rate of the user to object i according to the clicking rate prediction model.
By the device, object i can be recommended and estimate the higher user of clicking rate, the method increase and object is pushed to user
The degree of accuracy, in actual applications, reduces invalid recommendation, improves network utilization, can be implemented as different user and provides individual
The object recommendation of property.
Present invention also provides a kind of terminal, including：The memory of processor and the computer instruction that is stored with；It is described
Processor reads the computer instruction, and performs one kind as described above and set up clicking rate prediction model method.
Present invention also provides a kind of storage medium, be stored with computer instruction, real when the computer instruction is performed
Now one kind as described above sets up clicking rate prediction model method.
If it should be noted that the clicking rate prediction model device of setting up is realized simultaneously in the form of SFU software functional unit
As independent production marketing or in use, it can be stored in a computer read/write memory medium.Based on such reason
Solution, the application realizes all or part of flow in above-described embodiment method, can also instruct correlation by computer program
Hardware complete, described computer program can be stored in a computer-readable recording medium, the computer program is in quilt
When computer is performed, the step of each above-mentioned embodiment of the method can be achieved.Wherein, the computer program includes computer program
Code, the computer program code can be source code form, object identification code form, executable file or some intermediate forms
Deng.The computer-readable medium can include：Any entity or device, record of the computer program code can be carried
Medium, USB flash disk, mobile hard disk, magnetic disc, CD, computer storage, read-only storage (ROM, Read-Only Memory), with
Machine access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..
It should be noted that the content that the computer-readable medium is included can be according to legislation in jurisdiction and patent practice
It is required that carrying out appropriate increase and decrease, such as, in some jurisdictions, according to legislation and patent practice, computer-readable medium is not wrapped
Include electric carrier signal and telecommunication signal.
It should be noted that for foregoing each method embodiment, for simplicity description, therefore it is all expressed as a series of
Combination of actions, but those skilled in the art should know, the application is not limited by described sequence of movement because
According to the application, some steps can use other orders or carry out simultaneously.Secondly, those skilled in the art should also know
Know, embodiment described in this description belongs to preferred embodiment, involved action and module might not all be this Shens
Please be necessary.
In the above-described embodiments, the description to each embodiment all emphasizes particularly on different fields, and does not have the portion being described in detail in some embodiment
Point, it may refer to the associated description of other embodiments.
The application preferred embodiment disclosed above is only intended to help and illustrates the application.Alternative embodiment is not detailed
All details of narration, it is only described embodiment that this application is not limited yet.Obviously, according to the content of this specification,
It can make many modifications and variations.This specification is chosen and specifically describes these embodiments, is to preferably explain the application
Principle and practical application so that skilled artisan can be best understood by and utilize the application.The application is only
Limited by claims and its four corner and equivalent.
The application preferred embodiment and embodiment are explained in detail above in conjunction with accompanying drawing, but applied simultaneously
The above-described embodiment and examples are not limited to, in the knowledge that those skilled in the art possess, can also not departed from
The application makes a variety of changes on the premise of conceiving.
Claims (14)
1. a kind of method for setting up clicking rate prediction model, it is characterised in that including：
The feature of multiple first users is gathered, first user is the user that object was recommended, and the object is recommendation
Treat click on content；
For each user in the multiple first user each feature-set to the evaluation index of the object, based on described
Evaluation index builds the clicking rate prediction model；
According to the clicking rate prediction model, each first user is set up to the actual click value of the object and clicks value is estimated
Error function, error loss function is set up based on the error function；
Based on the error loss function and each first user counted in advance to the actual click value of the object, solve
The numerical value of the evaluation index set to the object；
The numerical value of the evaluation index to the object obtained according to solving determines the clicking rate prediction model.
2. according to the method described in claim 1, it is characterised in that also include：
Feature based on second user, click of the second user to the object is obtained using the clicking rate prediction model
The discreet value of rate, the second user is the user that the object was not recommended.
3. according to the method described in claim 1, it is characterised in that after the feature of multiple first users of collection, in addition to：
Feature is sorted out, multiple characteristic sets will be divided into per category feature；
For each user in the multiple first user each feature-set to also being wrapped after the evaluation index of the object
Include：
Identical numerical value is assigned to the identical evaluation index of the object for the feature in each characteristic set.
4. according to the method described in claim 1, it is characterised in that the evaluation index includes：Each feature is to described
Object estimates the degree of reliability a for estimating clicks value of clicks value r and each feature to the object, wherein, r ∈ [0,
1], a ∈ [0,1].
5. according to the method described in claim 1, it is characterised in that described to be based on the error loss function and in advance statistics
Each user to the actual click value of the object, solving the numerical value of the evaluation index set to the object includes：
The initial value of the evaluation index is set；
Calculating is iterated to the error loss function by target of the loss reduction of the error loss function；
Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and evaluated with now described
The value of index as the evaluation index numerical value.
6. method according to claim 5, it is characterised in that the clicking rate prediction model is：
<mrow>
<msub>
<mi>ctr</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
The error function is：
<mrow>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>ctr</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
The error loss function is：
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>,</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>u</mi>
<mo>&Element;</mo>
<mi>U</mi>
</mrow>
</munder>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
Wherein, i represents object, and u represents user, and U represents the set of first user, ctru,iRepresent user u to object i's
Clicking rate discreet value, f represents user u feature, FuRepresent user u characteristic set；yu,iRepresent reality of the user u to object i
Clicks value；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent that feature f estimates the reliable of clicks value to object i
Degree.
7. a kind of device for setting up clicking rate prediction model, it is characterised in that including：
Acquisition module, the feature for gathering multiple first users, first user is the user that object was recommended, described
Object treats click on content for what is recommended；
First modeling module, for each feature-set for each user in the multiple first user to the object
Evaluation index, the clicking rate prediction model is built based on the evaluation index；
Second modeling module, for according to the clicking rate prediction model, setting up reality of each first user to the object
Clicks value and the error function for estimating clicks value, error loss function is set up based on the error function；
Module is solved, for based on the reality of the error loss function and each first user counted in advance to the object
Border clicks value, solves the numerical value of the evaluation index set to the object；
3rd modeling module, for determining that the clicking rate is pre- according to the numerical value for solving the obtained evaluation index to the object
Estimate model.
8. device according to claim 7, it is characterised in that also include：
Computing module, for the feature based on second user, the second user pair is obtained using the clicking rate prediction model
The discreet value of the clicking rate of the object, the second user is the user that the object was not recommended.
9. device according to claim 7, it is characterised in that
The acquisition module, is additionally operable to be sorted out feature, will be divided into multiple characteristic sets per category feature；
First modeling module, is additionally operable to assign phase to the identical evaluation index of the object for the feature in each characteristic set
Same numerical value.
10. device according to claim 7, it is characterised in that the evaluation index includes：Each feature is to the object
Evaluation index include：Each feature estimates clicks value r and each feature to the object to the object
The degree of reliability a of clicks value is estimated, wherein, r ∈ [0,1], a ∈ [0,1].
11. device according to claim 7, it is characterised in that the solution module includes：
First solves submodule, the initial value for setting the evaluation index；
Second solves submodule, for being changed using the loss reduction of the error loss function as target to the loss function
In generation, calculates；
3rd solves submodule, based on stopping the iteration when the rate of change of the error loss function is less than predetermined threshold value
Calculate and the numerical value of the evaluation index is used as using the value of the now evaluation index.
12. device according to claim 11, it is characterised in that the clicking rate prediction model is：
<mrow>
<msub>
<mi>ctr</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
The error function is：
<mrow>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>ctr</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
The error loss function is：
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>,</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>u</mi>
<mo>&Element;</mo>
<mi>U</mi>
</mrow>
</munder>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>f</mi>
<mo>&Element;</mo>
<msub>
<mi>F</mi>
<mi>u</mi>
</msub>
</mrow>
</msub>
<msub>
<mi>a</mi>
<mrow>
<mi>f</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
Wherein, i represents object, and u represents user, and U represents the set of first user, ctru,iRepresent user u to object i's
Clicking rate discreet value, f represents user u feature, FuRepresent user u characteristic set；yu,iRepresent reality of the user u to object i
Clicks value；rf,iRepresent that feature f estimates clicks value to object i；af,iRepresent that feature f estimates the reliable of clicks value to object i
Degree.
13. a kind of terminal, it is characterised in that including：The memory of processor and the computer instruction that is stored with；
The processor reads the computer instruction, and performs a kind of foundation click as described in claim any one of 1-6
The method of rate prediction model.
14. a kind of storage medium, it is characterised in that be stored with computer instruction, is realized such as when the computer instruction is performed
A kind of method for setting up clicking rate prediction model described in claim any one of 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710578583.XA CN107301247B (en) | 2017-07-14 | 2017-07-14 | Method and device for establishing click rate estimation model, terminal and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710578583.XA CN107301247B (en) | 2017-07-14 | 2017-07-14 | Method and device for establishing click rate estimation model, terminal and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107301247A true CN107301247A (en) | 2017-10-27 |
| CN107301247B CN107301247B (en) | 2020-12-18 |
Family
ID=60132942
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710578583.XA Active CN107301247B (en) | 2017-07-14 | 2017-07-14 | Method and device for establishing click rate estimation model, terminal and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107301247B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886361A (en) * | 2017-11-14 | 2018-04-06 | 深圳市金立通信设备有限公司 | A kind of method and server for assessing ad conversion rates prediction model |
| CN107977859A (en) * | 2017-11-14 | 2018-05-01 | 广州优视网络科技有限公司 | Advertisement placement method, device, computing device and storage medium |
| CN108053050A (en) * | 2017-11-14 | 2018-05-18 | 广州优视网络科技有限公司 | Clicking rate predictor method, device, computing device and storage medium |
| CN109829116A (en) * | 2019-02-14 | 2019-05-31 | 北京达佳互联信息技术有限公司 | A kind of content recommendation method, device, server and computer readable storage medium |
| CN111160638A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Conversion estimation method and device |
| CN111159241A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Click conversion estimation method and device |
| CN111598638A (en) * | 2019-02-21 | 2020-08-28 | 北京沃东天骏信息技术有限公司 | Click rate determination method, device and equipment |
| CN111861542A (en) * | 2020-06-15 | 2020-10-30 | 北京雷石天地电子技术有限公司 | Method, device, terminal and non-transitory computer readable storage medium for delivering advertisement |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080275775A1 (en) * | 2007-05-04 | 2008-11-06 | Yahoo! Inc. | System and method for using sampling for scheduling advertisements in an online auction |
| CN104536983A (en) * | 2014-12-08 | 2015-04-22 | 北京掌阔技术有限公司 | Method and device for predicting advertisement click rate |
| CN104834641A (en) * | 2014-02-11 | 2015-08-12 | 腾讯科技（北京）有限公司 | Processing method of network medium information and relevant system |
| CN105260471A (en) * | 2015-10-19 | 2016-01-20 | 广州唯品会信息科技有限公司 | Training method and system of commodity personalized ranking model |
| CN105589971A (en) * | 2016-01-08 | 2016-05-18 | 车智互联（北京）科技有限公司 | Method and device for training recommendation model, and recommendation system |
| CN105678335A (en) * | 2016-01-08 | 2016-06-15 | 车智互联（北京）科技有限公司 | Click rate pre-estimation method, device and calculating equipment |
| CN105701191A (en) * | 2016-01-08 | 2016-06-22 | 腾讯科技（深圳）有限公司 | Push information click rate estimation method and device |
| CN105808762A (en) * | 2016-03-18 | 2016-07-27 | 北京百度网讯科技有限公司 | Resource sequencing method and device |
-
2017
- 2017-07-14 CN CN201710578583.XA patent/CN107301247B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080275775A1 (en) * | 2007-05-04 | 2008-11-06 | Yahoo! Inc. | System and method for using sampling for scheduling advertisements in an online auction |
| CN104834641A (en) * | 2014-02-11 | 2015-08-12 | 腾讯科技（北京）有限公司 | Processing method of network medium information and relevant system |
| CN104536983A (en) * | 2014-12-08 | 2015-04-22 | 北京掌阔技术有限公司 | Method and device for predicting advertisement click rate |
| CN105260471A (en) * | 2015-10-19 | 2016-01-20 | 广州唯品会信息科技有限公司 | Training method and system of commodity personalized ranking model |
| CN105589971A (en) * | 2016-01-08 | 2016-05-18 | 车智互联（北京）科技有限公司 | Method and device for training recommendation model, and recommendation system |
| CN105678335A (en) * | 2016-01-08 | 2016-06-15 | 车智互联（北京）科技有限公司 | Click rate pre-estimation method, device and calculating equipment |
| CN105701191A (en) * | 2016-01-08 | 2016-06-22 | 腾讯科技（深圳）有限公司 | Push information click rate estimation method and device |
| CN105808762A (en) * | 2016-03-18 | 2016-07-27 | 北京百度网讯科技有限公司 | Resource sequencing method and device |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886361A (en) * | 2017-11-14 | 2018-04-06 | 深圳市金立通信设备有限公司 | A kind of method and server for assessing ad conversion rates prediction model |
| CN107977859A (en) * | 2017-11-14 | 2018-05-01 | 广州优视网络科技有限公司 | Advertisement placement method, device, computing device and storage medium |
| CN108053050A (en) * | 2017-11-14 | 2018-05-18 | 广州优视网络科技有限公司 | Clicking rate predictor method, device, computing device and storage medium |
| CN109829116A (en) * | 2019-02-14 | 2019-05-31 | 北京达佳互联信息技术有限公司 | A kind of content recommendation method, device, server and computer readable storage medium |
| CN109829116B (en) * | 2019-02-14 | 2021-07-30 | 北京达佳互联信息技术有限公司 | Content recommendation method and device, server and computer readable storage medium |
| CN111598638A (en) * | 2019-02-21 | 2020-08-28 | 北京沃东天骏信息技术有限公司 | Click rate determination method, device and equipment |
| CN111598638B (en) * | 2019-02-21 | 2023-11-07 | 北京沃东天骏信息技术有限公司 | Click rate determination method, device and equipment |
| CN111159241A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Click conversion estimation method and device |
| CN111160638A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Conversion estimation method and device |
| CN111160638B (en) * | 2019-12-20 | 2022-09-02 | 深圳前海微众银行股份有限公司 | Conversion estimation method and device |
| CN111159241B (en) * | 2019-12-20 | 2023-04-07 | 深圳前海微众银行股份有限公司 | Click conversion estimation method and device |
| CN111861542A (en) * | 2020-06-15 | 2020-10-30 | 北京雷石天地电子技术有限公司 | Method, device, terminal and non-transitory computer readable storage medium for delivering advertisement |
| CN111861542B (en) * | 2020-06-15 | 2024-02-02 | 北京雷石天地电子技术有限公司 | Method, device, terminal and non-transitory computer readable storage medium for advertising |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107301247B (en) | 2020-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107301247A (en) | Set up the method and device, terminal, storage medium of clicking rate prediction model | |
| CN103353920B (en) | Method and device for recommending games based on SNS | |
| CN105373597B (en) | The user collaborative filtered recommendation method merging based on k medoids item cluster and partial interest | |
| CN107491813A (en) | A kind of long-tail group recommending method based on multiple-objection optimization | |
| CN107526810A (en) | Establish method and device, methods of exhibiting and the device of clicking rate prediction model | |
| CN104317959B (en) | Data digging method based on social platform and device | |
| CN106372249B (en) | A kind of clicking rate predictor method, device and electronic equipment | |
| CN103823908B (en) | Content recommendation method and server based on user preference | |
| CN103309967B (en) | Collaborative filtering method based on similarity transmission and system | |
| Wu et al. | Deep censored learning of the winning price in the real time bidding | |
| KR20200003109A (en) | Method and apparatus for setting sample weight, electronic device | |
| CN106484777A (en) | A kind of multimedia data processing method and device | |
| CN107194430A (en) | A kind of screening sample method and device, electronic equipment | |
| CN107016569A (en) | The targeted customer's account acquisition methods and device of a kind of networking products | |
| CN108665120A (en) | The foundation of scoring model, the appraisal procedure of user credit and device | |
| CN106815347A (en) | Improvement slope one Collaborative Filtering Recommendation Algorithms based on commodity similarity | |
| CN108198041A (en) | A kind of Method of Commodity Recommendation based on time attenuator, terminal and medium | |
| CN106846094A (en) | A kind of method and apparatus for recommending application message based on application has been installed | |
| CN102339448B (en) | Group purchase platform information processing method and device | |
| CN109711883A (en) | Internet advertising clicking rate predictor method based on U-Net network | |
| CN105469279A (en) | Commodity quality evaluation method and apparatus thereof | |
| CN107918825A (en) | A kind of method and apparatus that age of user section is judged based on application installation preference | |
| CN103366009B (en) | A kind of book recommendation method based on self-adaption cluster | |
| CN106919611A (en) | Product information method for pushing and device | |
| CN109460519A (en) | Browse object recommendation method and device, storage medium, server |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200527 Address after: 310051 room 508, floor 5, building 4, No. 699, Wangshang Road, Changhe street, Binjiang District, Hangzhou City, Zhejiang Province Applicant after: Alibaba (China) Co.,Ltd. Address before: 510627 unit 02, floor 15, Tower B, Pingyun Plaza, radio and television, No. 163, xipingyun Road, Huangpu Avenue, Tianhe District, Guangzhou City, Guangdong Province (for office use only) Applicant before: GUANGZHOU UC NETWORK TECHNOLOGY Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |