CN108520034A - Using recommendation method, apparatus and computer equipment - Google Patents
Using recommendation method, apparatus and computer equipment Download PDFInfo
- Publication number
- CN108520034A CN108520034A CN201810272484.3A CN201810272484A CN108520034A CN 108520034 A CN108520034 A CN 108520034A CN 201810272484 A CN201810272484 A CN 201810272484A CN 108520034 A CN108520034 A CN 108520034A
- Authority
- CN
- China
- Prior art keywords
- application
- pond
- installation
- recommendation
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/60—Software deployment
- G06F8/61—Installation
Abstract
The present invention provides a kind of application recommendation method, including：According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；Pond is applied according to the set of applications of target user's installation and recommendation, is generated comprising the second set for being combined into element using an application in pond by an application of set of applications and recommendation successively；Wherein, the element representation recommends another application after having installed an application；Obtain at least two basic forecast models；According to first set, second set and at least two basic forecast models, feature vector is obtained；Feature vector is inputted Fusion Model, obtains recommending the installation predicted value using applying in pond；Wherein, Fusion Model applies the incidence relation for the installation predicted value applied in pond for characteristic feature vector with recommendation；According to installation predicted value, recommend to apply to target user.Compared with traditional recommended models, the installation probability for the recommendation application that the present invention obtains is more acurrate, it can be achieved that exposure installation conversion ratio is obviously improved.
Description
Technical field
The present invention relates to field of computer technology, specifically, the present invention relates to a kind of applications to recommend method, apparatus and meter
Calculate machine equipment.
Background technology
It is various to disclosure satisfy that user in different necks as the fast development of Internet technology and the quick of intelligent terminal are popularized
Domain, application demand in different problem application software emerge one after another.In order to concentrate displaying application software to user and meet use
Installation requirements are downloaded in the application at family, such as application shop or application market can recommend application software to user and provide application software
The application platform for downloading channel is come into being.However, due to application software substantial amounts, application platform needs to recommend using application
Algorithm can be recommended its possible interested, liking and installation application software to user.
In the prior art, the application proposed algorithm that above application platform uses has very much, such as Ye Beisi, decision tree, god
Through network algorithm etc..These algorithms are each has something to recommend him, but also respectively have drawback.It has been seen for example, bayesian algorithm can stress recommended user
The application of classification, and neural network algorithm can then stress the application that recommended user has not seen classification.Application in the prior art
Proposed algorithm is excessively single, it may be exposed to user and loses interest in, install the low application software of wish degree, using recommendation
The exposure installation conversion ratio of algorithm is not high.
Invention content
It is that can at least solve above-mentioned one of technological deficiency, recommendation side is applied the present invention provides following technical scheme
Method, device and computer equipment.
The embodiment of the present invention provides a kind of application recommendation method according to one side, including：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；It is used according to target
The set of applications and recommend that pond, generation is applied to include the application by set of applications successively and recommend to answer using the one of pond that family is installed
With the second set for being combined into element；Wherein, the element representation recommends another application after having installed an application；
Obtain at least two basic forecast models；
According to the first set, the second set and at least two basic forecasts model, feature vector is obtained；
Described eigenvector is inputted Fusion Model, obtains recommending the installation predicted value using applying in pond；Wherein, described
Fusion Model is used to characterize described eigenvector and the incidence relation recommended using the installation predicted value applied in pond；
According to the installation predicted value, recommend to apply to the target user.
Preferably, the Fusion Model is generated by following steps：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；
Obtain the set of applications of historical user's installation；The corresponding recommendation of historical user is obtained using the peace applied in pond
Dress value；
It generates the application comprising the set of applications installed successively by the historical user and recommends the application using pond
It is combined into the third set of element；
Obtain at least two basic forecast models；
According to the first set, the third set and at least two basic forecasts model, feature vector is obtained；
According to described eigenvector and the corresponding installation value, training sample is generated；
According to the training sample, the Fusion Model is generated by regression algorithm training.
Further, the regression algorithm is logistic regression algorithm.
Preferably, at least two basic forecasts model is generated by following steps：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；
Obtain the set of applications of historical user's installation；The corresponding recommendation of historical user is obtained using the peace applied in pond
Dress value；
It generates the application comprising the set of applications installed successively by the historical user and recommends the application using pond
It is combined into the third set of element；
According to the first set, the third set, foundation characteristic vector is obtained；
According to described eigenvector and the corresponding installation value, basic model training sample is generated；
According to the multiple basic model training sample, training generates at least two basic forecast models.
Preferably, described to obtain the corresponding recommendation of historical user using the installation value applied in pond, including：
Historical user is obtained to recommending using the installation behavior applied in pond；
Corresponding installation value is obtained according to the installation behavior；Wherein, recommend using in pond if historical user's installation is described
Application then installation value be 1, if historical user do not install it is described recommend using pond application if installation value be 0.
Preferably, described according to the first set, the second set and at least two basic forecasts model, it obtains
To feature vector, including：
According to the first set and the second set, foundation characteristic vector is obtained；
The foundation characteristic vector is inputted at least two basic forecasts model respectively, is obtained and described at least two
Basic forecast model is corresponding to be recommended using at least two installation predicted values applied in pond；
Predicted value is installed according at least two basic models described in the foundation characteristic vector sum, obtains feature vector.
Preferably, at least two basic forecasts model, including：Bayes predictive model, decision tree prediction model, god
Through Network Prediction Model and SVM prediction models.
Preferably, described according to the installation predicted value, recommend to apply to the target user, including：
It determines that the installation predicted value is more than preset threshold value, recommends described recommend using Chi Zhongying to the target user
With.
Preferably, described according to the installation predicted value, recommend to apply to the target user, including：
The correspondence recommendation is ranked up using the multiple installation predicted values applied in pond by sequence from big to small, to
The target user recommends to come the corresponding application of installation predicted value of the preset quantity of foremost.
The embodiment of the present invention is a kind of using recommendation apparatus according on the other hand, additionally providing, including：
Gather generation module, for according to the application for always applying pond, generating comprising being combined into the of element by application two-by-two
One set；According to the set of applications of target user's installation and recommend to apply pond, generate comprising successively by an application of set of applications
With the second set for recommending to be combined into element using an application in pond；Wherein, the element representation is recommended after having installed an application
Another application；
Prediction model acquisition module, for obtaining at least two basic forecast models；
Feature vector acquisition module, for according to the first set, the second set and at least two basis
Prediction model obtains feature vector；
Predicted value acquisition module is installed, for described eigenvector to be inputted Fusion Model, obtains recommending to apply Chi Zhongying
Installation predicted value；Wherein, the Fusion Model is used to characterizing described eigenvector and recommends using applying in pond with described
The incidence relation of predicted value is installed；
Recommending module, for according to the installation predicted value, recommending to apply to the target user.
The embodiment of the present invention additionally provides a kind of computer equipment according to another aspect, and the terminal includes one
Or multiple processors；Memory；One or more application program, wherein one or more of application programs be stored in it is described
It in memory and is configured as being executed by one or more of processors, one or more of programs are configured to：It executes
Above-mentioned applies recommendation method.
Compared with prior art, the present invention having the advantages that：
Method, apparatus and computer equipment are recommended in application provided by the invention, by using by least two basic forecasts
The Fusion Model that Model Fusion obtains recommends to apply to realize to target user, the recommended models with traditional only single algorithm
It compares, the installation probability for the recommendation application that Fusion Model obtains through the invention is more acurrate, and then can realize application platform application
The exposure installation conversion ratio of proposed algorithm is obviously improved.
In addition, application provided by the invention recommends method, apparatus and computer equipment to be calculated also according to by Fusion Model
The installation predicted value obtained, to recommending using using screening, installing that probability is big to answer to expose it to target user in pond
With software, it can be achieved that further increasing the exposure installation conversion ratio of application platform application proposed algorithm.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Description of the drawings
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, wherein：
Fig. 1 is the method flow diagram that method is recommended in application provided in an embodiment of the present invention；
Fig. 2 is the method flow diagram provided in an embodiment of the present invention for generating Fusion Model；
Fig. 3 is the method flow diagram provided in an embodiment of the present invention for generating basic forecast model；
Fig. 4 is the structural schematic diagram provided in an embodiment of the present invention using recommendation apparatus；
Fig. 5 is the structural schematic diagram of computer equipment provided in an embodiment of the present invention.
Specific implementation mode
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, and is only used for explaining the present invention, and is not construed as limiting the claims.
Embodiment one
An embodiment of the present invention provides a kind of applications to recommend method, as shown in Figure 1, this method includes：
Step S101：According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；Root
According to the set of applications and recommendation of target user's installation using pond, generate comprising successively by an application of set of applications and recommendation application
One application in pond is combined into the second set of element；Wherein, the element representation recommends another application after having installed an application.
For the present embodiment, it can recommend application software to user in such as application shop or application market etc. and provide using soft
Part is downloaded in the application platform of channel, includes various application software, the class software of such as doing shopping, audio broadcast message class software,
Game class software etc..
For the present embodiment, the basis always applies the application in pond, generates comprising being combined into the of element by application two-by-two
One set, specifically, acquisition are described always using the application in pond, and described is always in presently described application platform using the application in pond
All application software, by described always using the application combination of two in pond at element, according to multiple Element generation first sets.Its
In, the application that the first application in two applications in each element is installed as platform user, the second application is used as to platform
The application that family is recommended.
For example, it includes being taken using a, using b and using c according to total application for applying pond always to apply pond (being denoted as set A)
Any two of which application is combined into the element, including application a＆ applications b, using a＆ applications c, using b＆ applications a etc., then give birth to
At comprising by using the first set F for being combined into element, the expression formula of the first set F is two-by-two：F=i＆ | i ∈ A, j ∈
A }, wherein element i＆j is for indicating：A certain user has installed applies j using after i to its recommendation.
It is described that pond is applied according to the set of applications and recommendation of target user's installation for the present embodiment, it generates comprising successively
The second set for being combined into element using an application in pond by an application and recommendation for set of applications specifically obtains target and uses
It includes the mounted one or more of user that pond, the set of applications are applied in the set of applications of family installation and pre-set recommendation
Application software, the recommendation include that pre-set current application platform is intended to one or more application recommended to the user using pond
One application of the set of applications and the application for recommending to apply pond are combined into element, according to multiple members by software successively
Element generates second set.Wherein, the application of first application expression target user's installation in two applications in each element, second
The application recommended to target user using expression.
Need it is clear that, the recommendation of different target user using pond can it is identical can be different, the recommendation can using pond
Pond is applied to be directed to the pre-set recommendation of single target user, can also be to be set in advance for all users of the application platform
Pond is applied in the recommendation set, and can also be that current goal user is pre-set and rejected for all users of the application platform have been pacified
The recommendation of dress application applies pond, the present embodiment not to limit this.
Step S102：Obtain at least two basic forecast models.
For the present embodiment, the basic forecast model may include bayes predictive model, can also include decision tree
Prediction model, can also be pre- including neural network prediction model, SVM (Support Vector Machine, support vector machines)
It surveys model etc. and applies proposed algorithm, the concrete model of at least two basic forecasts model is not limited in the present embodiment.
Step S103：According to the first set, the second set and at least two basic forecasts model, obtain
Feature vector.
Foundation characteristic vector can be generated according to the first set, the second set for the present embodiment, it is described
Foundation characteristic is used for the change for being included as the input feature vector variable of basic prediction model and the input feature vector variable of Fusion Model
Amount；By at least two basic forecasts model and the foundation characteristic Vector Fusion, described eigenvector can be obtained, wherein
Described eigenvector is for the input feature vector variable as Fusion Model.
Step S104：Described eigenvector is inputted Fusion Model, obtains recommending the installation predicted value using applying in pond；
Wherein, the Fusion Model is used to characterize described eigenvector and recommends to apply being associated with for the installation predicted value applied in pond with described
Relationship.
For the present embodiment, the Fusion Model is used to characterizing described eigenvector and recommends using applying in pond with described
The incidence relation of predicted value is installed, i.e., inputs trained Fusion Model using as the described eigenvector of input feature vector variable,
It can obtain according to the calculated recommendation of described eigenvector using the installation predicted value applied in pond.The installation prediction
Value is used to indicate on the basis of the target user has installed the application software that described eigenvector characterization has, then to the mesh
After the application in pond is applied in the recommendation that mark user recommends described eigenvector characterization to have, the target user can install institute
Recommend the probability of application.The installation predicted value is bigger, then it represents that target user installs the institute and recommends the probability of application bigger.
Step S105：According to the installation predicted value, recommend to apply to the target user.
For the present embodiment, according to the installation predicted value, recommend one using in pond to the target user
Or multiple applications.
The present invention using recommendation method by using the Fusion Model for obtaining at least two basic forecast Model Fusions
Recommend to apply to realize to target user, compared with traditional only recommended models of single algorithm, merges mould through the invention
The installation probability for the recommendation application that type obtains is more acurrate, and then can realize the exposure installation conversion of application platform application proposed algorithm
Rate is obviously improved.
Embodiment two
The alternatively possible realization method of the embodiment of the present invention further includes implementing on the basis of shown in embodiment one
Step shown in example two, wherein
As shown in Fig. 2, the Fusion Model in the step S104 is generated by following steps：
Step S201：According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two.
For the present embodiment, acquisition is described always using the application in pond, and described is always presently described application using the application in pond
All application software on platform, by described always using the application combination of two in pond at element, according to multiple Element generations first
Set.Wherein, the first application for install as platform user of application in two applications in each element, second apply conduct to
The application that platform user is recommended.
For example, it includes being taken using a, using b and using c according to total application for applying pond always to apply pond (being denoted as set A)
Any two of which application is combined into the element, including application a＆ applications b, using a＆ applications c, using b＆ applications a etc., then give birth to
At comprising by using the first set F for being combined into element, the expression formula of the first set F is two-by-two：F=i＆j | i ∈ A, j
∈ A }, wherein element i＆j is for indicating：A certain user has installed applies j using after i to its recommendation.
Step S202：Obtain the set of applications of historical user's installation；It obtains the corresponding recommendation of historical user and applies pond
The installation value of middle application.
For the present embodiment, the historical user indicates have execution to apply installation behavior in preset time in application platform
Validated user.Wherein, the preset time can be one day, one week, the times such as 30 days, and the present embodiment does not limit this.
The set of applications of historical user's installation includes the mounted one or more application software of the user.
Wherein, described to obtain the corresponding recommendation of historical user using the installation value applied in pond, including：Obtain history
User is to recommending using the installation behavior applied in pond；Corresponding installation value is obtained according to the installation behavior；Wherein, if history
Using the application in pond, then installation value is 1 for recommendation described in user installation, if historical user does not install recommendation the answering using pond
It is 0 with then installation value.
For the present embodiment, the installation value is for characterizing whether historical user in preset time installs recommendation application, tool
Body, if historical user installs the recommendation using the application in pond, installation value is 1, if historical user does not install described push away
The application using pond is recommended, then installation value is 0.
Step S203：It generates the application comprising the set of applications installed successively by the historical user and recommends to apply pond
An application be combined into the third set of element.
It is described that pond is applied according to the set of applications and recommendation of historical user's installation for the present embodiment, it generates comprising successively
The third set for being combined into element using an application in pond by an application and recommendation for set of applications specifically obtains history and uses
It includes the mounted one or more of user that pond, the set of applications are applied in the set of applications of family installation and pre-set recommendation
Application software, the recommendation include that pre-set current application platform is intended to one or more application recommended to the user using pond
One application of the set of applications and the application for recommending to apply pond are combined into element, according to multiple members by software successively
Element generates third set.Wherein, the application of first application expression historical user's installation in two applications in each element, second
The application recommended to historical user using expression.
Need it is clear that, the recommendation of different historical users using pond can it is identical can be different, the recommendation can using pond
Pond is applied to be directed to the pre-set recommendation of single historical user, can also be to be set in advance for all users of the application platform
Pond is applied in the recommendation set, and can also be that current historical user is pre-set and rejected for all users of the application platform have been pacified
The recommendation of dress application applies pond, the present embodiment not to limit this.
Step S204：Obtain at least two basic forecast models.
For the present embodiment, the basic forecast model may include bayes predictive model, can also include decision tree
Prediction model, can also be pre- including neural network prediction model, SVM (Support Vector Machine, support vector machines)
It surveys model etc. and applies proposed algorithm, the concrete model of at least two basic forecasts model is not limited in the present embodiment.
Step S205：According to the first set, the third set and at least two basic forecasts model, obtain
Feature vector.
Foundation characteristic vector can be generated according to the first set, the third set for the present embodiment, it is described
Foundation characteristic is used for the change for being included as the input feature vector variable of basic prediction model and the input feature vector variable of Fusion Model
Amount；By at least two basic forecasts model and the foundation characteristic Vector Fusion, feature vector can be obtained, wherein described
Feature vector is for the input feature vector variable as Fusion Model.
Step S206：According to described eigenvector and the corresponding installation value, training sample is generated.
For the present embodiment, according to the feature vector for the input feature vector variable as Fusion Model, and with it is described
Installation value described in the corresponding step S202 of the affiliated validated user of feature vector, produces for training Fusion Model
Training sample.
For the present embodiment, the historical user includes having execution to apply installation behavior in preset time in application platform
Multiple validated users.Therefore multiple training samples of corresponding multiple validated users are also produced according to the present embodiment above-mentioned steps.
Step S207：According to the training sample, the Fusion Model is generated by regression algorithm training.
For the present embodiment, it is preferable that the regression algorithm is logistic regression algorithm.
For the present embodiment, the normalized form of the logistic regression algorithm is：Wherein, X
Indicate that feature vector, b indicate that the biasing coefficient of Fusion Model, w indicate the feature weight coefficient of Fusion Model.
For the present embodiment, by the way that multiple training samples comprising described eigenvector and the corresponding installation value are defeated
Enter into the normalized form of above-mentioned logistic regression algorithm, and Fusion Model loss function los is solved by gradient descent method
Homographic solution w, b when (w, b) is minimum obtain parameter w, the b of Fusion Model, and then training generates the Fusion Model.
Need it is clear that, it is above-mentioned that generate the Fusion Model using the training of logistic regression algorithm be only one kind
The method for obtaining Fusion Model can also use the training of other algorithms to generate in addition to using above-mentioned logistic regression algorithm
Fusion Model.
In the present embodiment, by the way that at least two basic forecast Model Fusions are obtained Fusion Model, recommend to answer to improve
With the accuracy of installation probability, the exposure installation conversion ratio for then improving application platform application proposed algorithm provides strong support.
Embodiment three
The alternatively possible realization method of the embodiment of the present invention further includes implementing on the basis of shown in embodiment one
Step shown in example three, wherein
As shown in figure 3, at least two basic forecast models described in the step S102 are generated by following steps：
Step S301：According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two.
For the present embodiment, acquisition is described always using the application in pond, and described is always presently described application using the application in pond
All application software on platform, by described always using the application combination of two in pond at element, according to multiple Element generations first
Set.Wherein, the first application for install as platform user of application in two applications in each element, second apply conduct to
The application that platform user is recommended.
For example, it includes being taken using a, using b and using c according to total application for applying pond always to apply pond (being denoted as set A)
Any two of which application is combined into the element, including application a＆ applications b, using a＆ applications c, using b＆ applications a etc., then give birth to
At comprising by using the first set F for being combined into element, the expression formula of the first set F is two-by-two：F=i＆j | i ∈ A, j
∈ A }, wherein element i＆j is for indicating：A certain user has installed applies j using after i to its recommendation.
Step S302：Obtain the set of applications of historical user's installation；It obtains the corresponding recommendation of historical user and applies pond
The installation value of middle application.
For the present embodiment, the historical user indicates have execution to apply installation behavior in preset time in application platform
Validated user.Wherein, the preset time can be one day, one week, the times such as 30 days, and the present embodiment does not limit this.
The set of applications of historical user's installation includes the mounted one or more application software of the user.
Wherein, described to obtain the corresponding recommendation of historical user using the installation value applied in pond, including：Obtain history
User is to recommending using the installation behavior applied in pond；Corresponding installation value is obtained according to the installation behavior；Wherein, if history
Using the application in pond, then installation value is 1 for recommendation described in user installation, if historical user does not install recommendation the answering using pond
It is 0 with then installation value.
For the present embodiment, the installation value is for characterizing whether historical user in preset time installs recommendation application, tool
Body, if historical user installs the recommendation using the application in pond, installation value is 1, if historical user does not install described push away
The application using pond is recommended, then installation value is 0.
Step S303：It generates the application comprising the set of applications installed successively by the historical user and recommends to apply pond
An application be combined into the third set of element.
It is described that pond is applied according to the set of applications and recommendation of historical user's installation for the present embodiment, it generates comprising successively
The third set for being combined into element using an application in pond by an application and recommendation for set of applications specifically obtains history and uses
It includes the mounted one or more of user that pond, the set of applications are applied in the set of applications of family installation and pre-set recommendation
Application software, the recommendation include that pre-set current application platform is intended to one or more application recommended to the user using pond
One application of the set of applications and the application for recommending to apply pond are combined into element, according to multiple members by software successively
Element generates third set.Wherein, the application of first application expression historical user's installation in two applications in each element, second
The application recommended to historical user using expression.
Need it is clear that, the recommendation of different historical users using pond can it is identical can be different, the recommendation can using pond
Pond is applied to be directed to the pre-set recommendation of single historical user, can also be to be set in advance for all users of the application platform
Pond is applied in the recommendation set, and can also be that current historical user is pre-set and rejected for all users of the application platform have been pacified
The recommendation of dress application applies pond, the present embodiment not to limit this.
Step S304：According to the first set, the third set, foundation characteristic vector is obtained.
Foundation characteristic vector can be generated according to the first set, the third set for the present embodiment, it is described
Foundation characteristic is used for the input feature vector variable as basic prediction model.
Step S305：According to described eigenvector and the corresponding installation value, basic model training sample is generated.
It is vectorial according to the foundation characteristic for the input feature vector variable as basic prediction model for the present embodiment, with
And the installation value described in the step 302 corresponding with validated user belonging to the foundation characteristic vector, it produces for instructing
Practice the basic model training sample of basic forecast model.
For the present embodiment, the historical user includes having execution to apply installation behavior in preset time in application platform
Multiple validated users.Therefore multiple basic models that corresponding multiple validated users are also produced according to the present embodiment above-mentioned steps are instructed
Practice sample.
Step S306：According to the basic model training sample, training generates at least two basic forecast models.
For the present embodiment, by the way that multiple basic model training samples are input to pre-set basic forecast mould
In the corresponding normalized form of type, then it can train and generate the basic forecast model.Wherein, the basic forecast model may include
Bayes predictive model can also include decision tree prediction model, can also include neural network prediction model, SVM
(Support Vector Machine, support vector machines) prediction model etc. applies proposed algorithm, in the present embodiment to it is described extremely
The concrete model of few two basic forecast models does not limit.
Example IV
The alternatively possible realization method of the embodiment of the present invention further includes implementing on the basis of shown in embodiment one
Step shown in example four, wherein
The step S103 includes：
According to the first set and the second set, foundation characteristic vector is obtained；The foundation characteristic vector point
At least two basic forecasts model is not inputted, obtains recommend using pond corresponding at least two basic forecasts model
At least two installation predicted values of middle application；According at least two installation predicted value described in the foundation characteristic vector sum, obtain
Feature vector.
Wherein, as a preferred embodiment, at least two basic forecasts model, including：Bayes predictive model,
Decision tree prediction model, neural network prediction model and SVM prediction models.
Hereinafter, a practical application example is provided, to for the feature vector as Fusion Model input feature vector variable
Generating process is described in detail.
First, the first set and the second set, the first set and the second set are obtained, base is obtained
Plinth feature vector.
Specifically, it according to total application for applying pond A, generates comprising the first set F for being combined into element by application two-by-two, institute
The expression formula for stating first set F is：F=i＆j | i ∈ A, j ∈ A }, wherein element i＆j is for indicating：A certain user has installed
J is applied to its recommendation using after i, the element number of the first set F is n.
According to the set of applications I of target user's u installationsuPond J is applied with recommendation, is generated comprising successively by the one of set of applications
Using and recommend to be combined into the second set IJ of element using an application in pondu, such as { i1＆j, i2＆j ... }；Wherein, the member
Element indicates that target user u has been installed and applies j to its recommendation using after i.
According to the first set F and the second set IJu, obtain foundation characteristic vector xu,j, wherein the xu,jFor
To the mode input data that target user u recommends to generate using j, the foundation characteristic vector is n-dimensional vector, the foundation characteristic
The value of element is as follows in vector：
Wherein, the aindex (l) indicates the corresponding elements of serial number l in the first set F.
Then, the foundation characteristic vector is inputted at least two basic forecasts model respectively, obtain with it is described extremely
Few two basic forecast models are corresponding to be recommended using at least two installation predicted values applied in pond.
Specifically, at least two basic forecasts model includes：Bayes predictive model bayes, decision tree predict mould
Type dt, neural network prediction model bp and SVM prediction model svm.The foundation characteristic vector x that above-mentioned steps are obtainedu,jIt is input to
Aforementioned four basic forecast model obtains corresponding basic model installation predicted value bayes (xu,j)、dt(xu,j)、bp(xu,j)、
svm(xu,j)。
Finally, according at least two installation predicted value described in the foundation characteristic vector sum, feature vector is obtained.
Specifically, melt at least two basic forecast Model Fusions to the foundation characteristic vector, obtaining being used for conduct
The feature vector of molding type input feature vector variable, expression formula are：
uXu,j=(xu,j,bayes(xu,j),dt(xu,j),bp(xu,j),svm(xu,j))。
Embodiment five
The alternatively possible realization method of the embodiment of the present invention further includes implementing on the basis of shown in embodiment one
Step shown in example five, wherein
The step S105, including：It determines that the installation predicted value is more than preset threshold value, recommends institute to the target user
It states to recommend to apply and be applied in pond.
For the present embodiment, the recommendation pond includes that one or more is intended to the application recommended to target user, when to be recommended
When being more than preset threshold value using the corresponding installation predicted value, just recommend the application to the intended application.Wherein, described pre-
It can be any number in installation value section to set threshold value, such as when 0 expression does not install, indicates installation, the preset threshold value
Can be the sections numerical value such as 0.5,0.76,0.9, the present embodiment does not limit this.
In the present embodiment, it according to the installation predicted value being calculated by Fusion Model, is applied to recommending to apply in pond
It is screened, to recommend it to install the big application software of probability, it can be achieved that further increasing application platform application to target user
Conversion ratio is installed in the exposure of proposed algorithm.
Embodiment six
The alternatively possible realization method of the embodiment of the present invention further includes implementing on the basis of shown in embodiment one
Step shown in example six, wherein
The step S105, including：Correspondence it is described recommend using the multiple installation predicted values applied in pond by from greatly to
Small sequence is ranked up, and recommends the installation predicted value for the preset quantity for coming foremost is corresponding to answer to the target user
With.
For the present embodiment, the recommendation pond is using including multiple applications for being intended to recommend to target user, according to
After the multiple installation predicted values for the multiple users to be recommended of correspondence that step S104 is obtained, to the multiple installation predicted value by from
Small sequence is arrived greatly to be ranked up, and then filters out the installation predicted value for the preset quantity for coming foremost, and to the target
User recommends the corresponding application of installation predicted value of the preset quantity.
In the present embodiment, it according to the installation predicted value being calculated by Fusion Model, is applied to recommending to apply in pond
It is screened, to recommend it to install the big application software of probability, it can be achieved that further increasing application platform application to target user
Conversion ratio is installed in the exposure of proposed algorithm.
In addition, recommendation apparatus is applied an embodiment of the present invention provides a kind of, as shown in figure 4, described device includes：
Gather generation module, for according to the application for always applying pond, generating comprising being combined into the of element by application two-by-two
One set；According to the set of applications of target user's installation and recommend to apply pond, generate comprising successively by an application of set of applications
With the second set for recommending to be combined into element using an application in pond；Wherein, the element representation is recommended after having installed an application
Another application；
Prediction model acquisition module, for obtaining at least two basic forecast models；
Feature vector acquisition module, for according to the first set, the second set and at least two basis
Prediction model obtains feature vector；
Predicted value acquisition module is installed, for described eigenvector to be inputted Fusion Model, obtains recommending to apply Chi Zhongying
Installation predicted value；Wherein, the Fusion Model is used to characterizing described eigenvector and recommends using applying in pond with described
The incidence relation of predicted value is installed；
Recommending module, for according to the installation predicted value, recommending to apply to the target user.
Application recommendation apparatus provided by the invention, it can be achieved that：It is obtained by using by least two basic forecast Model Fusions
To Fusion Model recommend to apply to realize to target user, compared with traditional only recommended models of single algorithm, pass through
The installation probability for the recommendation application that Fusion Model of the present invention obtains is more acurrate, and then can realize application platform application proposed algorithm
Exposure installation conversion ratio is obviously improved.In addition, the application recommendation apparatus can also be achieved：It is calculated according to by Fusion Model
Installation predicted value, to recommending using using screening, to expose it to target user, to install the big application of probability soft in pond
Part is, it can be achieved that further increase the exposure installation conversion ratio of application platform application proposed algorithm.
The embodiment of the method provided in an embodiment of the present invention that above-mentioned offer may be implemented using recommendation apparatus, concrete function are real
Referring now to the explanation in embodiment of the method, details are not described herein.
In addition, the embodiment of the present invention additionally provides a kind of computer equipment, as shown in Figure 5.Calculating described in the present embodiment
Machine equipment can be the equipment such as server, personal computer and the network equipment.The computer equipment include processor 502,
The devices such as memory 503, input unit 504 and display unit 505.It will be understood by those skilled in the art that setting shown in Fig. 5
Standby structure devices do not constitute the restriction to all devices, may include components more more or fewer than diagram, or combine certain
A little components.Memory 503 can be used for storing application program 501 and each function module, and the operation of processor 502 is stored in memory
503 application program 501, to execute various function application and the data processing of equipment.Memory can be built-in storage
Or external memory, or including both built-in storage and external memory.Built-in storage may include read-only memory (ROM), can
Programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory or with
Machine memory.External memory may include hard disk, floppy disk, ZIP disks, USB flash disk, tape etc..Memory disclosed in this invention includes
But it is not limited to the memory of these types.Memory disclosed in this invention is only used as example rather than as restriction.
Input unit 504 is used to receive the input of signal, and receives keyword input by user.Input unit 504 can
Including touch panel and other input equipments.Touch panel collects user on it or neighbouring touch operation (for example is used
Family uses the operations of any suitable object or attachment on touch panel or near touch panel such as finger, stylus), and root
According to the corresponding attachment device of preset driven by program；Other input equipments can include but is not limited to physical keyboard, function
It is one or more in key (such as broadcasting control button, switch key etc.), trace ball, mouse, operating lever etc..Display unit
505 can be used for showing information input by user or be supplied to the information of user and the various menus of computer equipment.Display is single
The forms such as liquid crystal display, Organic Light Emitting Diode can be used in member 505.Processor 502 is the control centre of computer equipment, profit
With the various pieces of various interfaces and the entire computer of connection, by running or executing the software being stored in memory 502
Program and/or module, and the data being stored in memory are called, perform various functions and handle data.
As one embodiment, the computer equipment includes：One or more processors 502, memory 503, one
Or multiple application programs 501, wherein one or more of application programs 501 are stored in memory 503 and are configured as
Executed by one or more of processors 502, one or more of programs 501 be configured to carry out above example one to
Recommendation method is applied described in six.
Computer equipment provided by the invention, it can be achieved that：It is obtained by using by least two basic forecast Model Fusions
Fusion Model recommend to apply to realize to target user, compared with traditional only recommended models of single algorithm, pass through this
The installation probability for the recommendation application that invention Fusion Model obtains is more acurrate, and then can realize the exposure of application platform application proposed algorithm
Light installation conversion ratio is obviously improved.In addition, the computer equipment can also be achieved：According to the peace being calculated by Fusion Model
Predicted value is filled, it, can to recommending using, using screening, the big application software of probability being installed to expose it to target user in pond
Realize the exposure installation conversion ratio for further increasing application platform application proposed algorithm.
The embodiment of the method for above-mentioned offer may be implemented in computer equipment provided in an embodiment of the present invention, and concrete function is realized
The explanation in embodiment of the method is referred to, details are not described herein.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing module, it can also
That each unit physically exists alone, can also two or more units be integrated in a module.Above-mentioned integrated mould
The form that hardware had both may be used in block is realized, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized in the form of software function module and when sold or used as an independent product, can also be stored in a computer
In read/write memory medium.One of ordinary skill in the art will appreciate that realizing that all or part of step of above-described embodiment can be with
It is completed by hardware, relevant hardware can also be instructed to complete by program, which can be stored in a computer can
It reads in storage medium, storage medium may include memory, disk or CD etc..
The above is only some embodiments of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (11)
1. method is recommended in a kind of application, which is characterized in that include the following steps：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；Pacified according to target user
Pond is applied in the set of applications of dress and recommendation, and it includes that the application by set of applications and recommendation successively apply group using the one of pond to generate
The second set of synthesized element；Wherein, the element representation recommends another application after having installed an application；
Obtain at least two basic forecast models；
According to the first set, the second set and at least two basic forecasts model, feature vector is obtained；
Described eigenvector is inputted Fusion Model, obtains recommending the installation predicted value using applying in pond；Wherein, the fusion
Model is used to characterize described eigenvector and the incidence relation recommended using the installation predicted value applied in pond；
According to the installation predicted value, recommend to apply to the target user.
2. method is recommended in application according to claim 1, which is characterized in that the Fusion Model is given birth to by following steps
At：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；
Obtain the set of applications of historical user's installation；The corresponding recommendation of historical user is obtained using the installation applied in pond
Value；
It generates the application comprising the set of applications installed successively by the historical user and recommends the application combination using pond
At the third set of element；
Obtain at least two basic forecast models；
According to the first set, the third set and at least two basic forecasts model, feature vector is obtained；
According to described eigenvector and the corresponding installation value, training sample is generated；
According to the training sample, the Fusion Model is generated by regression algorithm training.
3. method is recommended in application according to claim 2, which is characterized in that the regression algorithm is calculated for logistic regression
Method.
4. method is recommended in application according to claim 1, which is characterized in that at least two basic forecasts model passes through
Following steps generate：
According to total application for applying pond, generate comprising the first set for being combined into element by application two-by-two；
Obtain the set of applications of historical user's installation；The corresponding recommendation of historical user is obtained using the installation applied in pond
Value；
It generates the application comprising the set of applications installed successively by the historical user and recommends the application combination using pond
At the third set of element；
According to the first set, the third set, foundation characteristic vector is obtained；
According to described eigenvector and the corresponding installation value, basic model training sample is generated；
According to the basic model training sample, training generates at least two basic forecast models.
5. method is recommended in application according to claim 2 or 4, which is characterized in that the corresponding institute of the acquisition historical user
The installation value recommended using being applied in pond is stated, including：
Historical user is obtained to recommending using the installation behavior applied in pond；
Corresponding installation value is obtained according to the installation behavior；Wherein, if historical user installs described recommend using answering in pond
It is 1 with then installation value, installation value is 0 if historical user does not install the recommendation using the application in pond.
6. method is recommended in application according to claim 1, which is characterized in that it is described according to the first set, described the
Two set and at least two basic forecasts model, obtain feature vector, including：
According to the first set and the second set, foundation characteristic vector is obtained；
The foundation characteristic vector is inputted at least two basic forecasts model respectively, is obtained and at least two basis
Prediction model is corresponding to be recommended using at least two basic models installation predicted value applied in pond；
Predicted value is installed according at least two basic models described in the foundation characteristic vector sum, obtains feature vector.
7. application recommendation method according to claim 1 or 6, which is characterized in that at least two basic forecasts model,
Including：Bayes predictive model, decision tree prediction model, neural network prediction model and SVM prediction models.
8. method is recommended in application according to claim 1, which is characterized in that it is described according to the installation predicted value, to institute
It states target user and recommends application, including：
It determines that the installation predicted value is more than preset threshold value, recommends the recommendation to apply in pond to the target user and apply.
9. method is recommended in application according to claim 1, which is characterized in that it is described according to the installation predicted value, to institute
It states target user and recommends application, including：
Recommend to be ranked up by sequence from big to small using the multiple installation predicted values applied in pond correspondence is described, to described
Target user recommends to come the corresponding application of installation predicted value of the preset quantity of foremost.
10. a kind of applying recommendation apparatus, which is characterized in that including：
Gather generation module, for according to the application for always applying pond, generating comprising the first collection for being combined into element by application two-by-two
It closes；The set of applications and recommendation installed according to target user apply pond, generation to include the application by set of applications successively and push away
Recommend the second set that element is combined into using an application in pond；Wherein, the element representation installed recommend after an application it is another
Using；
Prediction model acquisition module, for obtaining at least two basic forecast models；
Feature vector acquisition module, for according to the first set, the second set and at least two basic forecast
Model obtains feature vector；
Predicted value acquisition module is installed, for described eigenvector to be inputted Fusion Model, is obtained recommending using being applied in pond
Predicted value is installed；Wherein, the Fusion Model is used to characterize described eigenvector and recommends using the installation applied in pond with described
The incidence relation of predicted value；
Recommending module, for according to the installation predicted value, recommending to apply to the target user.
11. a kind of computer equipment, which is characterized in that it includes：
One or more processors；
Memory；
One or more application program, wherein one or more of application programs are stored in the memory and are configured
To be executed by one or more of processors, one or more of programs are configured to：It executes according to claim 1 to 9
Any one of them application recommendation method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810272484.3A CN108520034B (en) | 2018-03-29 | 2018-03-29 | Application recommendation method and device and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810272484.3A CN108520034B (en) | 2018-03-29 | 2018-03-29 | Application recommendation method and device and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108520034A true CN108520034A (en) | 2018-09-11 |
| CN108520034B CN108520034B (en) | 2021-01-05 |
Family
ID=63431249
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810272484.3A Active CN108520034B (en) | 2018-03-29 | 2018-03-29 | Application recommendation method and device and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108520034B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543132A (en) * | 2018-11-22 | 2019-03-29 | 深圳墨世科技有限公司 | Content recommendation method, device, electronic equipment and storage medium |
| CN110046952A (en) * | 2019-01-30 | 2019-07-23 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN111311104A (en) * | 2020-02-27 | 2020-06-19 | 第四范式（北京）技术有限公司 | Configuration file recommendation method, device and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105117440A (en) * | 2015-08-11 | 2015-12-02 | 北京奇虎科技有限公司 | Method and apparatus for determining to-be-recommended application (APP) |
| WO2016099867A1 (en) * | 2014-12-19 | 2016-06-23 | Paypal, Inc. | App store update notification and warning system |
| CN107045673A (en) * | 2017-03-31 | 2017-08-15 | 杭州电子科技大学 | Public bicycles changes in flow rate amount Forecasting Methodology based on heap Model Fusion |
| CN107103036A (en) * | 2017-03-22 | 2017-08-29 | 广州优视网络科技有限公司 | Using acquisition methods, equipment and the programmable device for downloading probability |
-
2018
- 2018-03-29 CN CN201810272484.3A patent/CN108520034B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016099867A1 (en) * | 2014-12-19 | 2016-06-23 | Paypal, Inc. | App store update notification and warning system |
| CN105117440A (en) * | 2015-08-11 | 2015-12-02 | 北京奇虎科技有限公司 | Method and apparatus for determining to-be-recommended application (APP) |
| CN107103036A (en) * | 2017-03-22 | 2017-08-29 | 广州优视网络科技有限公司 | Using acquisition methods, equipment and the programmable device for downloading probability |
| CN107045673A (en) * | 2017-03-31 | 2017-08-15 | 杭州电子科技大学 | Public bicycles changes in flow rate amount Forecasting Methodology based on heap Model Fusion |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543132A (en) * | 2018-11-22 | 2019-03-29 | 深圳墨世科技有限公司 | Content recommendation method, device, electronic equipment and storage medium |
| CN110046952A (en) * | 2019-01-30 | 2019-07-23 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN110046952B (en) * | 2019-01-30 | 2021-12-10 | 创新先进技术有限公司 | Recommendation model training method and device, and recommendation method and device |
| CN111311104A (en) * | 2020-02-27 | 2020-06-19 | 第四范式（北京）技术有限公司 | Configuration file recommendation method, device and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108520034B (en) | 2021-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107103057B (en) | A kind of resource supplying method and device | |
| US11048530B1 (en) | Predictive action modeling to streamline user interface | |
| Wauters et al. | A nearest neighbour extension to project duration forecasting with artificial intelligence | |
| CN111046286B (en) | Object recommendation method and device and computer storage medium | |
| CN109062919B (en) | Content recommendation method and device based on deep reinforcement learning | |
| Vignes et al. | Gene regulatory network reconstruction using Bayesian networks, the Dantzig Selector, the Lasso and their meta-analysis | |
| CN107613022A (en) | Content delivery method, device and computer equipment | |
| CN103502899B (en) | Dynamic prediction Modeling Platform | |
| CN109359247B (en) | Content pushing method, storage medium and computer equipment | |
| US8442925B2 (en) | Music recommendation method and apparatus | |
| CN108769125B (en) | Application recommendation method and device, storage medium and computer equipment | |
| CN106294830A (en) | The recommendation method and device of multimedia resource | |
| CN113614756A (en) | System and method for forecasting alerts with programmable human-machine hybrid ensemble learning | |
| CN109729395B (en) | Video quality evaluation method and device, storage medium and computer equipment | |
| CN108520034A (en) | Using recommendation method, apparatus and computer equipment | |
| CN105975641A (en) | Video recommendation method ad device | |
| CN106250532A (en) | Application recommendation method, device and server | |
| JP6975692B2 (en) | Method of presenting information related to the basis of the predicted value output by the computer system and the predictor. | |
| CN110971659A (en) | Recommendation message pushing method and device and storage medium | |
| CN107862551B (en) | Method and device for predicting network application promotion effect and terminal equipment | |
| CN108604332A (en) | System and method for generating prediction | |
| KR101790788B1 (en) | Collaborative networking with optimized inter-domain information quality assessment | |
| JP2011203991A (en) | Information processing apparatus, information processing method, and program | |
| CN113689144A (en) | Quality assessment system and method for product description | |
| CN108564220A (en) | User gender prediction method, apparatus, storage medium and computer 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: 20200417 Address after: 310052 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: 510640 Guangdong city of Guangzhou province Whampoa Tianhe District Road No. 163 Xiping Yun Lu Yun Ping square B radio tower 15 layer self unit 02 Applicant before: GUANGZHOU UC NETWORK TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |