CN107526810A - Establish method and device, methods of exhibiting and the device of clicking rate prediction model - Google Patents

Establish method and device, methods of exhibiting and the device of clicking rate prediction model Download PDF

Info

Publication number
CN107526810A
CN107526810A CN201710729982.1A CN201710729982A CN107526810A CN 107526810 A CN107526810 A CN 107526810A CN 201710729982 A CN201710729982 A CN 201710729982A CN 107526810 A CN107526810 A CN 107526810A
Authority
CN
China
Prior art keywords
mrow
event
clicking rate
msub
feature
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
Application number
CN201710729982.1A
Other languages
Chinese (zh)
Other versions
CN107526810B (en
Inventor
潘岸腾
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alibaba China Co Ltd
Original Assignee
Guangzhou Youshi Network Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangzhou Youshi Network Technology Co Ltd filed Critical Guangzhou Youshi Network Technology Co Ltd
Priority to CN201710729982.1A priority Critical patent/CN107526810B/en
Publication of CN107526810A publication Critical patent/CN107526810A/en
Application granted granted Critical
Publication of CN107526810B publication Critical patent/CN107526810B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • 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
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0263—Targeted advertisements based upon Internet or website rating
    • 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
    • G06Q30/0277—Online advertisement

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the present application discloses a kind of method and device for establishing clicking rate prediction model, methods of exhibiting and device, terminal and storage medium, wherein, methods described includes：The type of each displaying event and the clicking rate discreet value of the displaying event based on acquisition establish error function, wherein, the value of the type of the displaying event determines that clicking rate prediction model of the clicking rate discreet value based on structure of the displaying event is calculated according to the actual click situation of user；The error loss function of the clicking rate prediction model is established based on the error function；The value of type based on the error loss function and each displaying event, solve the value of the weight of the feature of each characteristic set in the clicking rate prediction model；The value of each weight obtained according to solving determines clicking rate prediction model.

Description

Establish method and device, methods of exhibiting and the device of clicking rate prediction model
Technical field
The application is related to network technique field, more particularly to a kind of method and device for establishing clicking rate prediction model, exhibition Show method and device.
Background technology
Network technical development needs to open up to today, such as increasing event, Domestic News, article, music, picture etc. Show to user, and the effect that this displaying will obtain, it is necessary to user is targetedly found for different displaying events, is entered The personalized displaying of row, the core technology difficult point of personalization displaying are how to accurately determine event to be presented.
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 establishing clicking rate prediction model Put, a kind of methods of exhibiting and device, terminal and storage medium, improve the accuracy that event is shown to user.
On the one hand, the application provides a kind of method for establishing clicking rate prediction model, including：
The type of each displaying event and the clicking rate discreet value of the displaying event based on acquisition establish error function, its In, the value of the type of the displaying event determines that the clicking rate of the displaying event is pre- according to the actual click situation of user Clicking rate prediction model of the valuation based on structure is calculated；
The error loss function of the clicking rate prediction model is established based on the error function；
The value of type based on the error loss function and each displaying event, solves the clicking rate and estimates mould The value of the weight of the feature of each characteristic set in type；
The value of each weight obtained according to solving determines clicking rate prediction model.
Alternatively, the clicking rate prediction model is built as follows：
Gather the feature of the user of displaying event；
The feature of the user is sorted out, multiple characteristic sets will be divided into per category feature, count each feature set Feature in conjunction sets weight to the standardized value of each displaying event for the feature of each characteristic set；
Feature based on each characteristic set is to the feature of the standardized value of each displaying event and each characteristic set The clicking rate prediction model of each displaying event of weight structure.
Alternatively, methods described also includes：
The clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model.
Alternatively, after the clicking rate discreet value that event to be presented is calculated based on the clicking rate prediction model, in addition to：
Based on the clicking rate discreet value for the event to be presented being calculated, calculate the expected of the event to be presented and receive Benefit；
The prospective earnings of multiple events to be presented are ranked up；
Prospective earnings are located at the event to be presented in predeterminable area and show user.
Alternatively, the clicking rate discreet value that event to be presented is calculated based on the clicking rate prediction model is included：
By the feature of characteristic set to the standardized value of the event to be presented and the weight of the feature of each characteristic set Substitute into the clicking rate discreet value that the clicking rate prediction model be calculated the event to be presented.
Alternatively, the value of the type based on the error loss function and each displaying event, solves each weight Value include：
For each weight, initial value is set；
Calculating is iterated to the error loss function using the loss reduction of the error loss function as target；
Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now each Value of the value of weight as the weight.
Alternatively, the clicking rate prediction model is：
The error loss function is：
Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents exhibition Show total number of events, flagjThe type of j-th of displaying event is represented, 0 is negative displaying event, and 1 is positive displaying event, if j-th of exhibition It is then positive displaying event to show that event is clicked on by user, is not clicked on then as negative displaying event, p by userI, jRepresent the i-th feature in jth The standardized value of the value of individual displaying event, calculation formula are as follows：
Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jPoint Hit number, show (fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
On the other hand, the application also provides a kind of methods of exhibiting, including：
Obtain the clicking rate discreet value of event to be presented；
Calculated based on the clicking rate discreet value and click on the prospective earnings that the event to be presented obtains；
The prospective earnings of multiple events to be presented are ranked up；
Prospective earnings are located at the event to be presented in predeterminable area and show user.
Alternatively, described calculated based on the clicking rate discreet value clicks on the prospective earnings bag that the event to be presented obtains Include：
The monovalent acquisition prospective earnings that are multiplied are clicked on default with the clicking rate discreet value.
On the other hand, the application also provides a kind of device for establishing clicking rate prediction model, including：
Acquisition module, built for the type of each displaying event based on acquisition and the clicking rate discreet value of the displaying event Vertical error function, wherein, clicking rate prediction model of the clicking rate discreet value based on structure of the displaying event is calculated, institute The value for stating the type of displaying event determines according to the actual click situation of user；
Module is established, for establishing the error loss function of the clicking rate prediction model based on the error function；
Module is solved, for the value of the type based on the error loss function and each displaying event, solves institute State the value of the weight of the feature of each characteristic set in clicking rate prediction model；
Determining module, for determining clicking rate prediction model according to the value for solving obtained each weight.
Alternatively, the clicking rate prediction model is built by following module：
Acquisition module, the feature of the user for gathering displaying event；
Statistical module, for the feature of the user to be sorted out, multiple characteristic sets will be divided into per category feature, united Standardized value of the feature in each characteristic set to each displaying event is counted, and power is set for the feature of each characteristic set Weight；
Module is built, for standardized value of the feature based on each characteristic set to each displaying event and each feature The clicking rate prediction model of each displaying event of weight structure of the feature of set.
Alternatively, described device also includes：
First computing module, for calculating the clicking rate discreet value of event to be presented based on the clicking rate prediction model.
Alternatively, first computing module also includes：
Second computing module, for the clicking rate discreet value based on the event to be presented being calculated, described in calculating The prospective earnings of event to be presented；
Order module, for being ranked up to the prospective earnings of multiple events to be presented；
Display module, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
Alternatively, first computing module is specifically used for the standard by the feature of characteristic set to the event to be presented The weight of change value and the feature of each characteristic set substitutes into the clicking rate prediction model and carries out that the thing to be presented is calculated The clicking rate discreet value of part.
Alternatively, the solution module includes：
Initialization module, for setting initial value for each weight；
Module is iterated to calculate, for being target to the error loss function using the loss reduction of the error loss function Calculating is iterated, stops the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now Value of the value of each weight as the weight.
Alternatively, the clicking rate prediction model is：
The error loss function is：
Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents exhibition Show total number of events, flagjThe type of j-th of displaying event is represented, 0 is negative displaying event, and 1 is positive displaying event, if j-th of exhibition It is then positive displaying event to show that event is clicked on by user, is not clicked on then as negative displaying event, p by userI, jRepresent the i-th feature in jth The standardized value of the value of individual displaying event, calculation formula are as follows：
Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jPoint Hit number, show (fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
On the other hand, the application also provides a kind of exhibiting device, including：
Discreet value acquisition module, for obtaining the clicking rate discreet value of event to be presented；
Income calculation module, for calculating the expection clicked on the event to be presented and obtained based on the clicking rate discreet value Income；
Income order module, for being ranked up to the prospective earnings of multiple events to be presented；
Event display module, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
Alternatively, the income calculation module is specifically used for being multiplied with default unit price of clicking on the clicking rate discreet value Obtain prospective earnings.
On the other hand, the application also provides a kind of terminal, including：Processor and the storage for being stored with computer instruction Device；
The processor reads the computer instruction, and performs a kind of foregoing clicking rate prediction model of establishing Method.
On the other hand, the application also provides a kind of terminal, including：Processor and the storage for being stored with computer instruction Device；
The processor reads the computer instruction, and performs a kind of foregoing methods of exhibiting.
On the other hand, the application also provides a kind of storage medium, is stored with computer instruction, and the computer instruction is held The foregoing method for establishing clicking rate prediction model is realized during row.
On the other hand, the application also provides a kind of storage medium, is stored with computer instruction, and the computer instruction is held Foregoing methods of exhibiting is realized during row.
A kind of method and device for establishing clicking rate prediction model, a kind of methods of exhibiting and the dress that the embodiment of the present application provides Put, terminal, storage medium, based on the click behavior of the user of multiple displaying events, feature is further divided into multiple Characteristic set, the standardized value of the feature of each characteristic set is counted, and be the feature configuration weight in each characteristic set, with The weight of each feature establishes clicking rate prediction model for parameter.After each weight determines, the clicking rate prediction model is just Determine.The clicking rate prediction model of the determination can be utilized to calculate the clicking rate discreet value of any displaying event, so as to User is showed will click on the high event of rate discreet value, improves the degree of accuracy that event is shown to user.
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：
Fig. 1 is the flow chart for the method for establishing clicking rate prediction model that the embodiment of the application one provides；
Fig. 2 is the flow chart for the method for establishing clicking rate prediction model that the embodiment of the application one provides；
Fig. 3 is the flow chart for the method for establishing clicking rate prediction model that the embodiment of the application one provides；
Fig. 4 is the flow chart for the method for establishing clicking rate prediction model that the embodiment of the application one provides；
Fig. 5 is the schematic diagram of a scenario for the displaying advertisement applications clicking rate prediction model that the embodiment of the application one provides；
Fig. 6 is the flow chart for the methods of exhibiting that the embodiment of the application one provides；
The structural representation for the device for establishing clicking rate prediction model that the embodiment of Fig. 7 the application one provides；
Fig. 8 is the structural representation for the exhibiting device that the embodiment of the application one provides.
Embodiment
The various aspects of the application are described below.Teaching herein can be embodied in the form of varied, and Any concrete structure disclosed herein, function or two kinds are only representational.Based on teaching herein, people in the art Member it is to be understood that one aspect disclosed herein can independently of any other aspect realize, and these aspect in Two or more aspects can combine in various manners.It is, for example, possible to use any number of the aspects set forth herein, Realization device puts into practice method.Further, it is possible to use other mechanisms, function or except one or more side 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.
In this application, there is provided a kind of method and device, terminal, storage medium for establishing clicking rate prediction model.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 establishing clicking rate prediction model, including step 101 to 104。
Step 101：The clicking rate discreet value of the type and the displaying event of each displaying event based on acquisition, which is established, to be missed Difference function, wherein, clicking rate prediction model of the clicking rate discreet value based on structure of the displaying event is calculated, the exhibition Show that the value of the type of event determines according to the actual click situation of user.
In the embodiment of the present application, the displaying event can various show user and it is desirable that user carries out click behaviour The event of work, such as can be displaying advertisement, displaying article, displaying application, displaying music or exhibiting pictures, displaying film.
The type of each displaying event is obtained, exemplified by showing advertisement, advertisement is shown to some user, if the user clicks on The displaying advertisement is checked, then the type of the displaying advertisement is exactly positive sample, can be represented with numeral 1；If the user does not have The displaying advertisement is checked in click, then this shows that the type of advertisement is exactly negative sample, can be represented with numeral 0.
Referring to Fig. 2, in the embodiment of the application one, the clicking rate prediction model passes through step 1011 to step 1013 structure Build.
Step 1011：Gather the feature of the user of displaying event.
In the embodiment of the present application, the feature of the user can include self attributes category feature, such as age, educational background, place Mobile phone application type that city, occupation, income or user like etc.；Mobile phone application category feature can also be included, such as mobile phone application Type, mobile phone application clicking rate etc..
Also exemplified by showing advertisement, certain advertising display has given multiple users, by gathering the feature of the multiple user i.e. The feature for the user that the advertising display is crossed can be obtained.
In the embodiment of the present application, each user can be regarded as a characteristic set, such as user A is with 95 Afterwards, the characteristic set A of the feature such as male, online game, user B be with after 95, women, the feature set of the feature such as online game Close B, user C is the characteristic set C with features such as after 90s, male, online games, acquisition characteristics set A, characteristic set B and Characteristic set C all features, then classify according to age characteristics, i.e., with after 95 feature point in a set, have Feature after 90s is divided in a set, has dividing in a set for online game feature, can also be according to sex character Classification, i.e., dividing in a set with masculinity, there is dividing in a set for femaleness.
Step 1012：The feature of the user is sorted out, multiple characteristic sets will be divided into per category feature, statistics is every Feature in individual characteristic set sets weight to the standardized value of each displaying event for the feature of each characteristic set.
In the embodiment of the present application, for every category feature of the user, multiple differences can be divided into some way Characteristic set.By taking the age as an example, " children ", " teenager ", " youth ", " middle age ", " old age " can be such as divided into, can also It is divided into " after 70 ", " after 80s ", " after 90s ", " after 95 ", " after 00 " etc..
Dividing mode per category feature can determine that the application is not construed as limiting according to being actually needed.
Step 1013：Standardized value and each characteristic set of the feature based on each characteristic set to each displaying event Feature each displaying event of weight structure clicking rate prediction model.
In the embodiment of the present application, the feature of each characteristic set is by each feature to the standardized value of each displaying event Actual click number of the feature of set to each displaying event and the feature to each characteristic set show the reality of the event Displaying number is calculated in advance.
In the embodiment of the present application, the weight of the feature of each characteristic set represents the feature of this feature set to displaying event Credibility.
In the embodiment of the application one, the clicking rate prediction model can be：
Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, pI, jRepresent Standardized value of i-th feature in j-th of value for showing event.
By the clicking rate prediction model, it is recognised that the clicking rate of some displaying event is by each characteristic set In standardized value and each characteristic set of the feature to showing event in the weight of feature determine jointly, each feature set Feature in conjunction is to showing that the standardized value of event can be according to the feature in each characteristic set to showing the actual point of event Hit number and the feature to each characteristic set shows that the actual displaying number of the event is calculated in advance.
Step 102：The error loss function of the clicking rate prediction model is established based on the error function.
Step 103：The value of type based on the error loss function and each displaying event, solves the click The value of the weight of the feature of each characteristic set in rate prediction model.
In the embodiment of the application one, the value of the type based on the error loss function and each displaying event, ask Solving the value of each weight can include：
For each weight, initial value is set；
Calculating is iterated to the error loss function using the loss reduction of the error loss function as target；
Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now each Value of the value of weight as the weight.
It is to be determined according to user to showing true click situation and the clicking rate discreet value of event in the embodiment of the present application Each feature is to the weight of the displaying event.
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, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents exhibition Show total number of events, pI, jRepresent standardized value of i-th feature in j-th of value for showing event, flagjRepresent j-th of displaying thing The type of part, 0 is negative displaying event, and 1 is positive displaying event, is positive displaying event if j-th of displaying event is clicked on by user, Do not clicked on then as negative displaying event by user.
If user clicks displaying event j, flagj=1, otherwise flagj=0.
Wherein, pI, jCalculation formula it is as follows：
Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jPoint Hit number, show (fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
Still exemplified by showing advertisement, the displaying advertisement gives the user with feature " after 95 " to show 100 times, wherein producing Raw click behavior number be 10 times, then " after 95 " feature to it is described displaying advertisement standardized value be then：What if the weight of the feature in the characteristic set of displaying advertisement was to determine, then So that the clicking rate of the displaying advertisement is estimated to the weight of the displaying advertisement according to each feature.
According to formula 3 it is recognised that the embodiment of the present application can be according to the p calculated in advanceI, jAnd user is to showing event Type actual value, solve the value of weight corresponding to each feature.
In the embodiment of the present application, gradient descent method can be used, the actual click value of the user based on displaying event, is solved { the θ of formula 1i| 0≤i≤n, i ∈ Z } optimum value, that is, solve each feature to show event weight.The method for solving can be with Comprise the following steps：
The first step：Number { the θ between one group of 0-1 is given at randomi| 0≤i≤n, i ∈ Z }, it is set to θ(0), initialize iteration step Number k=0；
Second step：Iterative calculation
Wherein α is the step-length of iteration, takes 0.001；
3rd step：Judge whether the error loss function restrains
Δg(θ(k+1))=| g (θ(k+1))-g(θ(k))|
If | Δ g (θ(k+1))-Δg(θ(k)| ＜ β, then be returned to θ(k+1), θ(k+1)The parameter for the model as estimated, Otherwise return to second step to continue to calculate, wherein β is the value of a very little, can take the α of β=0.01.
Step 104：The value of each weight obtained according to solving determines clicking rate prediction model.
In the embodiment of the present application, the value that weight of each feature to showing event is determined is calculated by above-mentioned steps, from And show the clicking rate prediction model of event and just establish.
In the embodiment of the present application, the clicking rate that event to be presented can be calculated based on the clicking rate prediction model is estimated Value.
A kind of method for establishing clicking rate prediction model that the embodiment of the present application provides, with the user's of multiple displaying events Based on click behavior, feature is further divided into multiple characteristic sets, counts the standardization of the feature of each characteristic set Value, and be the feature configuration weight in each characteristic set, establish clicking rate prediction model by parameter of the weight of each feature. After each weight determines, the clicking rate prediction model determines that.The clicking rate prediction model of the determination can be utilized Calculate the clicking rate discreet value of any displaying event, so as to select the high displaying event of clicking rate discreet value, improve to Family shows the degree of accuracy of event.
Referring to Fig. 3, in one embodiment of the application, reality of the event to be presented based on the clicking rate prediction model Using, including step 301 is to step 304.
Step 301：The clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model.
Step 302：Based on the clicking rate discreet value for the event to be presented being calculated, the event to be presented is calculated Prospective earnings.
In the embodiment of the present application, the clicking rate discreet value based on the event to be presented being calculated includes：By feature The feature of set substitutes into the clicking rate to the weight of the standardized value of the event to be presented and the feature of each characteristic set Prediction model be calculated the clicking rate discreet value of the event to be presented.
Step 303：The prospective earnings of multiple events to be presented are ranked up.
In the embodiment of the present application, it is anticipated that income is ranked up to multiple events to be presented, can be by prospective earnings height Event to be presented be placed on preferential position, the low event to be presented of prospective earnings is placed on rearward position.
Step 304：Prospective earnings are located at the event to be presented in predeterminable area and show user.
For example, the event to be presented that prospective earnings are preceding 90 can be placed in predeterminable area, then it is anticipated that the height of income It is low to be placed on prospective earnings ranking in predeterminable area positioned at preceding 90 event to be presented, then by the predeterminable area 90 events to be presented show user.
In the embodiment of the present application, the clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model, will Event to be presented is ranked up according to clicking rate discreet value, and clicking rate discreet value highest event is shown to user, can be very big Click of the raising user to showing event check interest, lift Consumer's Experience.
Referring to Fig. 4, in one embodiment of the application, exemplified by showing event A for displaying advertisement, illustrate the application institute The method for building up of the clicking rate prediction model of offer, including step 401 is to step 407.
Step 401：The feature of the user of collection displaying advertisement.
The clicking rate prediction model of the displaying advertisement is established by gathering the feature for the multiple users for showing advertisement.Institute It can be multiple to state the user of displaying advertisement, and the feature of each user can also be multiple.
Step 402：The feature of the user is sorted out, multiple characteristic sets will be divided into per category feature, statistics is every Feature in individual characteristic set sets weight to the standardized value for showing advertisement, and for the feature of each characteristic set.
In the embodiment of the present application, the characteristic set for the user for portraying the displaying advertisement can be gathered from multiple dimensions.
Dimension 1：Portrayed by the preference to the displaying advertisement of user, such as the user to enjoy shopping portrays " shopping Fan ".
Dimension 2：Portrayed by the Regional Property of user, such as Beijing, Tianjin, Shanghai.
Dimension 3：Portrayed by user's natural quality, such as age, sex etc..
Dimension 4：Portrayed by the social property of user, such as educational level, occupation, region etc..
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 403：Standardized value and each characteristic set of the feature based on each characteristic set to the displaying advertisement The weight of feature build the clicking rate prediction model.
In the embodiment of the present application, the feature of each characteristic set is by each feature to the standardized value of the displaying advertisement The feature of set shows the reality of the advertisement to the actual click number of the displaying advertisement and the feature to each characteristic set Displaying number is calculated.
In the embodiment of the present application, the weight of the feature of each characteristic set represents the feature of each characteristic set to the exhibition Show the Different Reliability of advertisement.
By taking the age as an example, if the feature of this type of age further comprises " after 90s ", " after 95 ", " after 00 " 3 features Set, then the feature in each characteristic set has identical weight to same this displaying event of displaying advertisement.For example, The user of 26 years old and 24 years old belongs to " after 90s " this characteristic set, then the age of the user of 26 years old and the user of 24 years old this Feature is identical to the value of the weight of same this displaying event of displaying advertisement.
The clicking rate prediction model for the displaying advertisement established in the embodiment of the present application is as shown in Equation 5：
Wherein, A represents this displaying event of displaying advertisement, and i represents feature, θiRepresent feature i weight, θ0Represent constant, N represents feature sum, pI, AStandardized value of i-th feature in the value of the displaying advertisement is represented, calculation formula is as follows：
Wherein, fI, ARepresent value of the ith feature in displaying advertisement, click (fI, A) represent feature fI, AHits, show(fI, A) represent feature fI, ADisplaying number, pctr (fI, A) represent feature fI, AClicking rate.
Such as：The displaying advertisement gives the user with feature " after 95 " to show 100 times, wherein caused click behavior time Number be 10 times, then " after 95 " feature to it is described displaying advertisement standardized value be then：
Step 404：Obtain the exhibition of the type of the displaying advertisement and the clicking rate prediction model calculating based on structure Show the clicking rate discreet value of advertisement, the value of the displaying advertisement determines according to the actual click rate of user.
In the embodiment of the present application, if user, which clicks on, has checked the displaying advertisement, then the type of the displaying advertisement is just It is positive sample, can be represented with numeral 1；If user does not have click to check the displaying advertisement, then the class of the displaying advertisement Type is exactly negative sample, can be represented with numeral 0.
Step 405：The type and the error function of the clicking rate discreet value of the displaying advertisement are established, based on the mistake Difference function establishes the error loss function of the clicking rate prediction model.
Error function in the embodiment of the present application is as shown in Equation 7.
The error loss function is as shown in Equation 8.
Wherein, A represents this displaying event of displaying advertisement, and i represents feature, θiFeature i weight is represented, n represents feature Sum, m represent displaying advertisement sum, pI, ARepresent standardized value of i-th feature in the value of the displaying advertisement, flagARepresent The type of the displaying advertisement, 0 is negative displaying event, and 1 is positive displaying event, is just if the displaying advertisement is clicked on by user Displaying event, do not clicked on then as negative displaying event by user.
If user clicks the displaying advertisement, flagA=1, otherwise flagA=0.
Step 406：The value of type based on the error loss function and the displaying advertisement, solves each weight Value.
According to formula 8 as can be seen that the embodiment of the present application is based on the error loss function and the displaying advertisement The value of type is actual click value, solves value of the feature to the weight for showing advertisement of each characteristic set.
In the embodiment of the present application, gradient descent method can be used, the actual click value of the user based on the displaying advertisement, Solve weight of each feature to displaying event.The method for solving may include steps of：
The first step：Number { the θ between one group of 0-1 is given at randomi| 0≤i≤n, i ∈ Z }, it is set to θ(0), initialize iteration step Number k=0；
Second step：Iterative calculation
Wherein α is the step-length of iteration, takes 0.001；
3rd step：Judge whether the error loss function restrains
Δg(θ(k+1))=| g (θ(k+1))-g(θ(k))|
If | Δ g (θ(k+1))-Δg(θ(k)) | ＜ β, then be returned to θ(k+1), θ(k+1)The parameter for the model as estimated, Otherwise return to second step to continue to calculate, wherein β is the value of a very little, can take the α of β=0.01.
Step 407：The value of each weight obtained according to solving determines clicking rate prediction model.
In the embodiment of the present application, calculated by above-mentioned steps and determine each feature to the weight for showing advertisement Value, so as to which the clicking rate prediction model of the displaying advertisement just establishes.
Fig. 5 is the application scenarios schematic diagram to the displaying ad click rate prediction model.The advertisement was not shown for one User I, user I includes 3 features, respectively i1、i2、i3；The feature of multiple users of the displaying advertisement is gathered first, then will The feature of the multiple user is sorted out, and will be divided into multiple characteristic sets per category feature, and count the feature pair in each characteristic set The standardized value of the displaying advertisement, i.e., Wherein, i1、i2、i3The value of corresponding weight is respectively θ1、θ2、θ3, the i1、i2、i3The value of corresponding weight is by above-mentioned Modeling process determines.Then user I is to the clicking rate discreet value for showing advertisement：ctrA=θ0+p1, Aθ1+p2, Aθ2+p3, Aθ3。
By the feature of each characteristic set based on the displaying advertisement to the standardized value for showing advertisement and often The weight of the feature of individual characteristic set establishes the clicking rate prediction model of the displaying advertisement, and can be estimated using the clicking rate Model calculates the clicking rate discreet value of other displaying advertisements, then therefrom selects the high advertising display of clicking rate discreet value to use Family, the dispensing degree of accuracy that advertisement is shown to user can be greatly enhanced, the advertising display of low clicking rate is avoided, has saved displaying Advertisement putting cost, improve the economic benefit of displaying advertisement.
Referring to Fig. 6, the methods of exhibiting of the embodiment of the application one offer, including step 601 to step 604.
Step 601：Obtain the clicking rate discreet value of event to be presented.
In the embodiment of the present application, the formula of the clicking rate prediction model in formula 1, user is calculated to different things to be presented The clicking rate discreet value of part.
Step 602：Calculated based on the clicking rate discreet value and click on the prospective earnings that the event to be presented obtains.
In the embodiment of the present application, by the feature of each characteristic set to the standardized value of the event to be presented and each spy The clicking rate that the weight substitution clicking rate prediction model for collecting the feature closed be calculated the event to be presented is pre- Valuation, the expected receipts of the event to be presented can be obtained by being multiplied with the clicking rate discreet value with default click unit price Benefit.
Step 603：The prospective earnings of multiple events to be presented are ranked up.
In the embodiment of the present application, after can the prospective earnings of multiple events to be presented all be calculated, it is anticipated that receiving Benefit is sorted successively from high to low, and multiple events to be presented can also be ranked up in particular order.
Step 604：Prospective earnings are located at the event to be presented in predeterminable area and show user.
For example, all event sets to be presented of default resources bank are A, can estimate out user u by above-mentioned steps clicks on The event b to be presented of any one in set A probability Pu(xb), if xbClick unit price be priceb, the value has event to be presented Supplier provide, then event b to be presented is earn to user u prospective earningsb=ctrj·priceb, according to the value pair Set A does descending, shows user to intercept top100 the event to be presented that prospective earnings are located in predeterminable area and waits to open up Show that event shows user u.
In the embodiment of the present application, the clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model, will Event to be presented is ranked up according to clicking rate discreet value, and it is preferential that the high event to be presented of prospective earnings is placed on into high-quality position Show user, it is possible to achieve profit maximization.
The clicking rate prediction model that the application provides is applied in displaying musically, including the first step is to the 4th step.
The first step：Obtain the clicking rate discreet value of music to be presented.
In the embodiment of the present application, the formula of the clicking rate prediction model in formula 1, user is calculated to different to be presented The clicking rate discreet value of music.
Second step：Calculated based on the clicking rate discreet value and click on the prospective earnings that the music to be presented obtains.
In the embodiment of the present application, by standardized value of the feature of each characteristic set to the displaying music and each feature The weight of the feature of set substitutes into the clicking rate that above-mentioned clicking rate prediction model be calculated the music to be presented and estimated Value, the music to be presented can be obtained by being multiplied with the clicking rate discreet value with the click unit price of default music to be presented Prospective earnings.
3rd step：The prospective earnings of multiple music to be presented are ranked up.
In the embodiment of the present application, after can the prospective earnings of multiple music to be presented all be calculated, it is anticipated that receiving Benefit is sorted successively from high to low.
4th step：Prospective earnings are located at the music to be presented in predeterminable area and show user.
Such as：The music to be presented that prospective earnings are top80 can be placed in predeterminable area, then it is anticipated that the height of income The low music to be presented that prospective earnings ranking is located to top80 is placed in predeterminable area, then will be located in the predeterminable area 80 first music to be presented show user.
In the embodiment of the present application, the clicking rate discreet value of music to be presented is calculated based on the clicking rate prediction model, will Music to be presented is ranked up according to clicking rate discreet value, and it is preferential that the high music to be presented of prospective earnings is placed on into high-quality position User is showed, during music is shown, realizes the maximization for accomplishing profit inside limited resource, it is also maximum Raising user's using effect of degree.
The method for establishing clicking rate prediction model of the application, stage one：Sample data is collected first, and extracts feature, Then the standardization based on ctr is carried out to feature, the ctr for finally establishing standardized feature estimates linear model, and passes through gradient Descent method tries to achieve model parameter.Stage two：The model established according to the stage one, it is general to calculate click of the user to different displaying events Rate, user is clicked on the high event of probability and shows user.
The clicking rate prediction model that the application establishes is applied and promoted in business application (equivalent to this displaying of displaying advertisement Event) in, it is the major source of revenues for applying shop that business application, which is promoted,.In business application extension process, main target is How (to include user resources, using displaying position resource) inside limited resource and generate profit maximization.How to generate profit Maximize, be to need to predict that user (shows event, such as show this displaying of advertisement to the business application of displaying the problem of core Event) click probability, as long as calculating click probability, with click probability be multiplied by business application unit price, you can obtain showing this The prospective earnings of business application, the high business application of income is placed on into high-quality position, and preferentially displaying can generate profit maximization Target.
Fig. 7 is a kind of structural representation for device for establishing clicking rate prediction model that the embodiment of the present application provides.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 it is only schematical.
The device for establishing clicking rate prediction model of the application offer includes：
Acquisition module 701, estimated for the type of each displaying event based on acquisition and the clicking rate of the displaying event Value establishes error function, wherein, the value of the type of the displaying event determines according to the actual click situation of user, the exhibition Show that clicking rate prediction model of the clicking rate discreet value of event based on structure is calculated；
Module 702 is established, for establishing the error loss function of the clicking rate prediction model based on the error function；
Module 703 is solved, for the value of the type based on the error loss function and each displaying event, is solved The value of the weight of the feature of each characteristic set in the clicking rate prediction model；
Determining module 704, for determining clicking rate prediction model according to the value for solving obtained each weight.
In the embodiment of the application one, the clicking rate prediction model is built by following module：
Acquisition module, the feature of the user for gathering displaying event；
Statistical module, for the feature of the user to be sorted out, multiple characteristic sets will be divided into per category feature, united Standardized value of the feature in each characteristic set to each displaying event is counted, and power is set for the feature of each characteristic set Weight；
Module is built, for standardized value of the feature based on each characteristic set to each displaying event and each feature The clicking rate prediction model of each displaying event of weight structure of the feature of set.
In the embodiment of the application one, described device also includes：
First computing module, for calculating the clicking rate discreet value of event to be presented based on the clicking rate prediction model.
In the embodiment of the application one, first computing module also includes：
Second computing module, for the clicking rate discreet value based on the event to be presented being calculated, described in calculating The prospective earnings of event to be presented；
Order module, for being ranked up to the prospective earnings of multiple events to be presented；
Display module, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
In the embodiment of the application one, first computing module is specifically used for the feature of characteristic set to described to be presented The weight of the feature of the standardized value of event and each characteristic set substitutes into the clicking rate prediction model and institute is calculated State the clicking rate discreet value of event to be presented.
In the embodiment of the application one, the solution module 703 includes：
Initialization module, for setting initial value for each weight；
Module is iterated to calculate, for being target to the error loss function using the loss reduction of the error loss function Calculating is iterated, stops the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now Value of the value of each weight as the weight.
In the embodiment of the application one, the clicking rate prediction model is：
The error loss function is：
Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents exhibition Show total number of events, flagjThe type of j-th of displaying event is represented, 0 is negative displaying event, and 1 is positive displaying event, if j-th of exhibition It is then positive displaying event to show that event is clicked on by user, is not clicked on then as negative displaying event, p by userI, jRepresent the i-th feature in jth The standardized value of the value of individual displaying event, calculation formula are as follows：
Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jPoint Hit number, show (fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
A kind of device for establishing clicking rate prediction model that the embodiment of the present application provides, with the user's of multiple displaying events Based on click behavior, feature is further divided into multiple characteristic sets, counts the standardization of the feature of each characteristic set Value, and be the feature configuration weight in each characteristic set, establish clicking rate prediction model by parameter of the weight of each feature. After each weight determines, the clicking rate prediction model determines that.The clicking rate prediction model of the determination can be utilized The clicking rate discreet value of any displaying event is calculated, user is showed so as to will click on the high event of rate discreet value, is improved The degree of accuracy of event is shown to user.
The application also provides a kind of terminal, including：Processor and the memory for being stored with computer instruction；
The processor reads the computer instruction, and performs a kind of clicking rate prediction model of establishing as described above Method.
The application also provides a kind of storage medium, is stored with computer instruction, and the computer instruction is realized when being performed The method as described above for establishing clicking rate prediction model.
Fig. 8 is a kind of structural representation for exhibiting device that the embodiment of the present application provides.Due to the basic phase of device embodiment Embodiment of the method is similar to, the relevent part can refer to the partial explaination of embodiments of method.Device embodiment described below is only It is schematical.
A kind of exhibiting device that the application provides includes：
Discreet value acquisition module 801, for obtaining the clicking rate discreet value of event to be presented；
Income calculation module 802, click on what the event to be presented obtained for being calculated based on the clicking rate discreet value Prospective earnings；
Income order module 803, for being ranked up to the prospective earnings of multiple events to be presented；
Event display module 804, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
In the embodiment of the application one, the income calculation module 802 is specifically used for the clicking rate discreet value with presetting Click unit price be multiplied obtain prospective earnings.
In the embodiment of the present application, the clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model, will Event to be presented is ranked up according to clicking rate discreet value, gives user to show clicking rate discreet value highest displaying event, can be with Greatly improve click of the user to showing event and check interest, lift Consumer's Experience.
The application also provides a kind of terminal, including：Processor and the memory for being stored with computer instruction；
The processor reads the computer instruction, and performs a kind of methods of exhibiting as described above.
The application also provides a kind of storage medium, is stored with computer instruction, and the computer instruction is realized when being performed Methods of exhibiting as described above.
If it should be noted that the clicking rate prediction model device of establishing is realized simultaneously in the form of SFU software functional unit As independent production marketing or in use, can be stored in a computer read/write memory medium.Based on such reason Solution, the application realize 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 Computer perform when, can be achieved above-mentioned each embodiment of the method the step of.Wherein, the computer program includes computer program Code, the computer program code can be source code form, displaying event code form, executable file or some centres Form etc..The computer-readable medium can include：Can carry the computer program code any entity or device, Recording medium, USB flash disk, mobile hard disk, magnetic disc, CD, computer storage, read-only storage (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software Distribution medium etc..It should be noted that the content that includes of the computer-readable medium can be according to making laws in jurisdiction Appropriate increase and decrease is carried out with the requirement of patent practice, such as in some jurisdictions, according to legislation and patent practice, computer Computer-readable recording medium does not include electric carrier signal and telecommunication signal.
It should be noted that for foregoing each method embodiment, in order to which simplicity describes, 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, and 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 are described, it is only described embodiment also not limit this application.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 those skilled in the art's possessed knowledge, can also not departed from The application makes a variety of changes on the premise of conceiving.

Claims (22)

  1. A kind of 1. method for establishing clicking rate prediction model, it is characterised in that including：
    The type of each displaying event and the clicking rate discreet value of the displaying event based on acquisition establish error function, wherein, The value of the type of the displaying event is according to the determination of the actual click situation of user, the clicking rate discreet value of the displaying event Clicking rate prediction model based on structure is calculated；
    The error loss function of the clicking rate prediction model is established based on the error function；
    The value of type based on the error loss function and each displaying event, is solved in the clicking rate prediction model The value of the weight of the feature of each characteristic set；
    The value of each weight obtained according to solving determines clicking rate prediction model.
  2. 2. according to the method for claim 1, it is characterised in that the clicking rate prediction model is built as follows：
    Gather the feature of the user of displaying event；
    The feature of the user is sorted out, multiple characteristic sets will be divided into per category feature, counted in each characteristic set Feature to the standardized value of each displaying event, and for the feature of each characteristic set, weight is set；
    Feature based on each characteristic set is to the standardized value of each displaying event and the weight of the feature of each characteristic set The clicking rate prediction model of each displaying event of structure.
  3. 3. according to the method for claim 1, it is characterised in that also include：
    The clicking rate discreet value of event to be presented is calculated based on the clicking rate prediction model.
  4. 4. according to the method for claim 3, it is characterised in that event to be presented is calculated based on the clicking rate prediction model Clicking rate discreet value after, in addition to：
    Based on the clicking rate discreet value for the event to be presented being calculated, the prospective earnings of the calculating event to be presented；
    The prospective earnings of multiple events to be presented are ranked up；
    Prospective earnings are located at the event to be presented in predeterminable area and show user.
  5. 5. according to the method for claim 3, it is characterised in that described to be presented based on clicking rate prediction model calculating The clicking rate discreet value of event includes：
    The feature of characteristic set is substituted into the weight of the standardized value of the event to be presented and the feature of each characteristic set The clicking rate prediction model be calculated the clicking rate discreet value of the event to be presented.
  6. 6. according to the method for claim 1, it is characterised in that show based on the error loss function and each event Type value, solving the value of each weight includes：
    For each weight, initial value is set；
    Calculating is iterated to the error loss function using the loss reduction of the error loss function as target；
    Stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now each weight Value of the value as the weight.
  7. 7. according to the method for claim 6, it is characterised in that the clicking rate prediction model is：
    <mrow> <msub> <mi>ctr</mi> <mi>j</mi> </msub> <mo>=</mo> <msub> <mi>&amp;theta;</mi> <mn>0</mn> </msub> <mo>+</mo> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <msub> <mi>&amp;theta;</mi> <mi>i</mi> </msub> </mrow>
    The error loss function is：
    <mrow> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>flag</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>ctr</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>flag</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>&amp;theta;</mi> <mn>0</mn> </msub> <mo>-</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <msub> <mi>&amp;theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow>
    Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents displaying event Sum, flagjThe type of j-th of displaying event is represented, 0 is negative displaying event, and 1 is positive displaying event, if j-th of displaying event It is then positive displaying event to be clicked on by user, is not clicked on then as negative displaying event, p by userI, jRepresent the i-th feature in j-th of displaying The standardized value of the value of event, calculation formula are as follows：
    <mrow> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>=</mo> <mi>p</mi> <mi>c</mi> <mi>t</mi> <mi>r</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>c</mi> <mi>l</mi> <mi>i</mi> <mi>c</mi> <mi>k</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>s</mi> <mi>h</mi> <mi>o</mi> <mi>w</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>
    Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jHits, show(fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
  8. A kind of 8. methods of exhibiting, it is characterised in that including：
    Obtain the clicking rate discreet value of event to be presented；
    Calculated based on the clicking rate discreet value and click on the prospective earnings that the event to be presented obtains；
    The prospective earnings of multiple events to be presented are ranked up；
    Prospective earnings are located at the event to be presented in predeterminable area and show user.
  9. 9. according to the method for claim 8, it is characterised in that described to be calculated based on the clicking rate discreet value described in click The prospective earnings that event to be presented obtains include：
    The monovalent acquisition prospective earnings that are multiplied are clicked on default with the clicking rate discreet value.
  10. A kind of 10. device for establishing clicking rate prediction model, it is characterised in that including：
    Acquisition module, the clicking rate discreet value for the type and the displaying event of each displaying event based on acquisition, which is established, to be missed Difference function, wherein, clicking rate prediction model of the clicking rate discreet value based on structure of the displaying event is calculated, the exhibition Show that the value of the type of event determines according to the actual click situation of user；
    Module is established, for establishing the error loss function of the clicking rate prediction model based on the error function；
    Module is solved, for the value of the type based on the error loss function and each displaying event, solves the point Hit the value of the weight of the feature of each characteristic set in rate prediction model；
    Determining module, for determining clicking rate prediction model according to the value for solving obtained each weight.
  11. 11. device according to claim 10, it is characterised in that the clicking rate prediction model passes through following module structure Build：
    Acquisition module, the feature of the user for gathering displaying event；
    Statistical module, for the feature of the user to be sorted out, multiple characteristic sets will be divided into per category feature, statistics is every Feature in individual characteristic set sets weight to the standardized value of each displaying event for the feature of each characteristic set；
    Module is built, for standardized value of the feature based on each characteristic set to each displaying event and each characteristic set Feature each displaying event of weight structure clicking rate prediction model.
  12. 12. device according to claim 10, it is characterised in that also include：
    First computing module, for calculating the clicking rate discreet value of event to be presented based on the clicking rate prediction model.
  13. 13. device according to claim 12, it is characterised in that first computing module also includes：
    Second computing module, for the clicking rate discreet value based on the event to be presented being calculated, wait to open up described in calculating Show the prospective earnings of event；
    Order module, for being ranked up to the prospective earnings of multiple events to be presented；
    Display module, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
  14. 14. device according to claim 12, it is characterised in that first computing module is specifically used for characteristic set Feature the clicking rate substituted into the weight of the standardized value of the event to be presented and the feature of each characteristic set estimated Model be calculated the clicking rate discreet value of the event to be presented.
  15. 15. device according to claim 10, it is characterised in that the solution module includes：
    Initialization module, for setting initial value for each weight；
    Module is iterated to calculate, for being carried out using the loss reduction of the error loss function as target to the error loss function Iterative calculation, stop the iterative calculation when the rate of change of the error loss function is less than predetermined threshold value and with now each Value of the value of weight as the weight.
  16. 16. device according to claim 15, it is characterised in that the clicking rate prediction model is：
    <mrow> <msub> <mi>ctr</mi> <mi>j</mi> </msub> <mo>=</mo> <msub> <mi>&amp;theta;</mi> <mn>0</mn> </msub> <mo>+</mo> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <msub> <mi>&amp;theta;</mi> <mi>i</mi> </msub> </mrow>
    The error loss function is：
    <mrow> <mi>g</mi> <mrow> <mo>(</mo> <mi>&amp;theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>flag</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>ctr</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>flag</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>&amp;theta;</mi> <mn>0</mn> </msub> <mo>-</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <msub> <mi>&amp;theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow>
    Wherein, j represents displaying event, and i represents feature, θiFeature i weight is represented, n represents feature sum, and m represents displaying event Sum, flagjThe type of j-th of displaying event is represented, 0 is negative displaying event, and 1 is positive displaying event, if j-th of displaying event It is then positive displaying event to be clicked on by user, is not clicked on then as negative displaying event, p by userI, jRepresent the i-th feature in j-th of displaying The standardized value of the value of event, calculation formula are as follows：
    <mrow> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>=</mo> <mi>p</mi> <mi>c</mi> <mi>t</mi> <mi>r</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>c</mi> <mi>l</mi> <mi>i</mi> <mi>c</mi> <mi>k</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>s</mi> <mi>h</mi> <mi>o</mi> <mi>w</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>
    Wherein, fI, jRepresent that ith feature shows the value of event, click (f at j-thI, j) represent feature fI, jHits, show(fI, j) represent feature fI, jDisplaying number, pctr (fI, j) represent feature fI, jClicking rate.
  17. A kind of 17. exhibiting device, it is characterised in that including：
    Discreet value acquisition module, for obtaining the clicking rate discreet value of event to be presented；
    Income calculation module, for calculating the expected receipts clicked on the event to be presented and obtained based on the clicking rate discreet value Benefit；
    Income order module, for being ranked up to the prospective earnings of multiple events to be presented；
    Event display module, user is showed for prospective earnings to be located at into the event to be presented in predeterminable area.
  18. 18. device according to claim 17, it is characterised in that the income calculation module is specifically used for the click Rate discreet value clicks on the monovalent acquisition prospective earnings that are multiplied with default.
  19. A kind of 19. terminal, it is characterised in that including：Processor and the memory for being stored with computer instruction；
    The processor reads the computer instruction, and performs a kind of establish as described in claim any one of 1-7 and click on The method of rate prediction model.
  20. A kind of 20. terminal, it is characterised in that including：Processor and the memory for being stored with computer instruction；
    The processor reads the computer instruction, and performs a kind of methods of exhibiting as claimed in claim 8 or 9.
  21. 21. a kind of storage medium, it is characterised in that be stored with computer instruction, realized such as when the computer instruction is performed The method for establishing clicking rate prediction model described in claim any one of 1-7.
  22. 22. a kind of storage medium, it is characterised in that be stored with computer instruction, realized such as when the computer instruction is performed A kind of methods of exhibiting described in claim 8 or 9.
CN201710729982.1A 2017-08-23 2017-08-23 Method and device for establishing click rate estimation model and display method and device Active CN107526810B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710729982.1A CN107526810B (en) 2017-08-23 2017-08-23 Method and device for establishing click rate estimation model and display method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710729982.1A CN107526810B (en) 2017-08-23 2017-08-23 Method and device for establishing click rate estimation model and display method and device

Publications (2)

Publication Number Publication Date
CN107526810A true CN107526810A (en) 2017-12-29
CN107526810B CN107526810B (en) 2021-01-26

Family

ID=60682193

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710729982.1A Active CN107526810B (en) 2017-08-23 2017-08-23 Method and device for establishing click rate estimation model and display method and device

Country Status (1)

Country Link
CN (1) CN107526810B (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109389424A (en) * 2018-09-20 2019-02-26 北京达佳互联信息技术有限公司 Flow allocation method, device, electronic equipment and storage medium
WO2019144892A1 (en) * 2018-01-25 2019-08-01 腾讯科技（深圳）有限公司 Data processing method, device, storage medium and electronic device
CN110263959A (en) * 2018-05-21 2019-09-20 腾讯科技（深圳）有限公司 Clicking rate predictor method, device, machinery equipment and computer readable storage medium
CN110401855A (en) * 2018-04-25 2019-11-01 腾讯科技（深圳）有限公司 Information displaying method, processing platform, calculates equipment and storage medium at device
CN111126649A (en) * 2018-10-31 2020-05-08 北京字节跳动网络技术有限公司 Method and apparatus for generating information
CN112150182A (en) * 2019-06-28 2020-12-29 腾讯科技（深圳）有限公司 Multimedia file pushing method and device, storage medium and electronic device
CN112214665A (en) * 2019-07-12 2021-01-12 北京字节跳动网络技术有限公司 Content display method and device, electronic equipment and computer readable storage medium
CN112612951A (en) * 2020-12-17 2021-04-06 上海交通大学 Unbiased learning sorting method for income improvement

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060242139A1 (en) * 2005-04-21 2006-10-26 Yahoo! Inc. Interestingness ranking of media objects
CN104951965A (en) * 2015-06-26 2015-09-30 深圳市腾讯计算机系统有限公司 Advertisement delivery method and device
CN105809487A (en) * 2016-03-17 2016-07-27 广州筷子信息科技有限公司 Advertisement element based advertisement release method and apparatus

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060242139A1 (en) * 2005-04-21 2006-10-26 Yahoo! Inc. Interestingness ranking of media objects
CN104951965A (en) * 2015-06-26 2015-09-30 深圳市腾讯计算机系统有限公司 Advertisement delivery method and device
CN105809487A (en) * 2016-03-17 2016-07-27 广州筷子信息科技有限公司 Advertisement element based advertisement release method and apparatus

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019144892A1 (en) * 2018-01-25 2019-08-01 腾讯科技（深圳）有限公司 Data processing method, device, storage medium and electronic device
CN110401855A (en) * 2018-04-25 2019-11-01 腾讯科技（深圳）有限公司 Information displaying method, processing platform, calculates equipment and storage medium at device
CN110263959A (en) * 2018-05-21 2019-09-20 腾讯科技（深圳）有限公司 Clicking rate predictor method, device, machinery equipment and computer readable storage medium
CN110263959B (en) * 2018-05-21 2024-01-09 腾讯科技（深圳）有限公司 Click rate estimation method, click rate estimation device, machine equipment and computer readable storage medium
CN109389424A (en) * 2018-09-20 2019-02-26 北京达佳互联信息技术有限公司 Flow allocation method, device, electronic equipment and storage medium
CN109389424B (en) * 2018-09-20 2021-08-06 北京达佳互联信息技术有限公司 Flow distribution method and device, electronic equipment and storage medium
CN111126649B (en) * 2018-10-31 2023-08-11 北京字节跳动网络技术有限公司 Method and device for generating information
CN111126649A (en) * 2018-10-31 2020-05-08 北京字节跳动网络技术有限公司 Method and apparatus for generating information
CN112150182A (en) * 2019-06-28 2020-12-29 腾讯科技（深圳）有限公司 Multimedia file pushing method and device, storage medium and electronic device
CN112150182B (en) * 2019-06-28 2023-08-29 腾讯科技（深圳）有限公司 Multimedia file pushing method and device, storage medium and electronic device
CN112214665A (en) * 2019-07-12 2021-01-12 北京字节跳动网络技术有限公司 Content display method and device, electronic equipment and computer readable storage medium
CN112612951B (en) * 2020-12-17 2022-07-01 上海交通大学 Unbiased learning sorting method for income improvement
CN112612951A (en) * 2020-12-17 2021-04-06 上海交通大学 Unbiased learning sorting method for income improvement

Also Published As

Publication number Publication date
CN107526810B (en) 2021-01-26

Similar Documents

Publication Publication Date Title
CN107526810A (en) Establish method and device, methods of exhibiting and the device of clicking rate prediction model
US11070643B2 (en) Discovering signature of electronic social networks
CN106372249B (en) A kind of clicking rate predictor method, device and electronic equipment
CN108197330B (en) Data digging method and device based on social platform
CN103353920B (en) Method and device for recommending games based on SNS
CN110097066A (en) A kind of user classification method, device and electronic equipment
CN107301247A (en) Set up the method and device, terminal, storage medium of clicking rate prediction model
CN108256568A (en) A kind of plant species identification method and device
CN108230010A (en) A kind of method and server for estimating ad conversion rates
CN109409928A (en) A kind of material recommended method, device, storage medium, terminal
CN103353880B (en) A kind of utilization distinctiveness ratio cluster and the data digging method for associating
CN108629608A (en) User data processing method and processing device
CN106570718A (en) Information releasing method and releasing system
CN104185041B (en) The automatic generation method and system of video interactive advertisement
CN106776859A (en) Mobile solution App commending systems based on user preference
CN106951471A (en) A kind of construction method of the label prediction of the development trend model based on SVM
CN107391582A (en) The information recommendation method of user preference similarity is calculated based on context ontology tree
CN108596276A (en) The naive Bayesian microblog users sorting technique of feature based weighting
CN110070134A (en) A kind of recommended method and device based on user interest perception
CN106327211A (en) SCRM system based on social media and development method thereof
US20110251889A1 (en) Inventory clustering
CN112132634A (en) Virtual gift resource distribution method and device, computer equipment and storage medium
Zhang et al. Mining target users for mobile advertising based on telecom big data
CN109785002A (en) A kind of interior prediction technique of paying of user&#39;s game
CN111626767A (en) Resource data distribution method, device and equipment

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
TA01 Transfer of patent application right

Effective date of registration: 20200526

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.

GR01 Patent grant
GR01 Patent grant