Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing to embodiment party of the present inventionFormula is described in further detail.
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Following description is related toDuring accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodimentDescribed in embodiment do not represent and the consistent all embodiments of the present invention.On the contrary, they be only with it is such as appendedThe example of the consistent apparatus and method of some aspects being described in detail in claims, of the invention.
Fig. 1 is the application environment schematic diagram that method is recommended in a kind of application provided in an embodiment of the present invention, as described in Figure 1, itsThe structural representation of the application environment involved by the embodiment of the present invention is shown, this includes using the application environment of recommendation method：Server 101 and at least one terminal 102.
Terminal 102 is connected by wireless or cable network and server 101, and terminal 102 can be computer, intelligent handThe electronic equipments such as machine, tablet personal computer.
Server 101 can be the Internet application server, the Internet application server, can be carried for the Internet, applicationsFor background service.The Internet, applications provide the information exchange service such as voice, video, picture, word as one for intelligent terminalApplication program, have the advantages that voice, video, picture and word can be sent across common carrier, spanning operation system platform.
The Internet application server can be configured as a server that service is provided by internet, and the internet shouldCan be social networking application server with server, for example, being taken corresponding to the social network sites such as instant communication server, forum or microbloggingBusiness device, can also be the server that the business such as payment can be realized by internet, and the embodiment of the present invention takes to the Internet, applicationsThe type of business device limits without specific.
Certainly, the server 101 can also be other servers, such as multimedia resource shared server, and the present invention is realExample is applied to be not especially limited the type of the server.
Fig. 2A is that method flow diagram is recommended in a kind of application provided in an embodiment of the present invention, and the executive agent of this method is serviceDevice, referring to Fig. 2A, this method includes：
201st, each application and the correlation of the reference feature of multiple users in multiple applications are obtained.
The reference feature is used for the creditworthiness for representing user, and the reference feature can use reference fraction representation, user'sReference fraction can determine according to behavioral data on the line of user, can also associate behavioral data under the line according to user and determine,Or determined according to behavioral data on the line of user with behavioral data is associated under line, it can also be determined according to other data of userThe credit score, the embodiment of the present invention are not especially limited to this.
It should be noted that behavioral data can include social interaction behavioral data, virtual increment clothes on the line of the userAt least one of the data, economic behaviour data and amusement and leisure behavioral data of being engaged in data, association behavioral data can wrap under lineInclude in wearable device data, tourism trip data, O2O (Online to Offline, i.e., under line on line) service for life dataAt least one data, certainly, on the line behavioral data with behavioral data is associated under line can also include respectively or can be itHis data, the embodiment of the present invention are not especially limited to this.
The correlation is used to represent application and the degree of correlation of the reference feature of user, and the numerical value of the correlation is bigger, tableShow that application is bigger with the degree of correlation of the reference feature of user, numerical value is smaller, represents that application is related to the reference feature of userDegree is smaller.For example, for financial class application, higher is required to the creditworthiness of user, correspondingly, the financial class apply withThe degree of correlation of the reference feature at family is higher, and accessed correlation is bigger.
In embodiments of the present invention, each application and the correlation of the reference feature of multiple users in multiple applications are obtainedMethod includes step 2011 to 2013：
2011st, for any application in the plurality of application, according to the reference feature of the plurality of user, to the plurality of userIt is grouped.
According to the reference feature of the plurality of user or the distribution situation of the reference feature of the plurality of user, to the plurality of useFamily is grouped, and the number of users of every group of user can be made identical, can also make the number of users of every two groups of users in multigroup userWithin a preset range, the preset range can be defined as any fixed value to difference between amount, and the embodiment of the present invention is not made to thisIt is specific to limit.
So that the number of users of every group of user is identical as an example, specific group technology can be：If should with reference fraction representationReference feature, the plurality of user is arranged in order by the order of reference fraction from low to high or from high to low, using predetermined number asBenchmark, the multiple users arranged in order are grouped.
The predetermined number can be defined as any fixed value, can also be according to the determination of the number of users of the plurality of user and groupNumber determines, i.e., the ratio of the number of users of the plurality of user and group number is defined as into the predetermined number, it is, of course, also possible to according to itHis method determines the predetermined number, and the embodiment of the present invention is not especially limited to this.
2012nd, according to every group of user to the download situation of the application and the plurality of user to the download situation of the application, obtainTake the application and the correlation of the reference feature of every group of user.
Download situation to the application can be the number of users for downloading the application and the number of users for not downloading the applicationRatio；Can also be the download total degree to the application, or to download the number of users of the application and not downloading the applicationNumber of users etc. other download situations, the embodiment of the present invention is not construed as limiting to the specific meaning of the download situation.Correspondingly, rootIt is different according to the implication indicated by the download situation, obtain the application and the method for the correlation of the reference feature of every group of userDifference, the embodiment of the present invention are also not construed as limiting to specific acquisition methods.
According to the download situation of application, the correlation of application and the reference feature of user is obtained, can embody download should answerThe creditworthiness of user, and then can embody using the requirement height to user credit degree.For example, should for financial classWith when the number of users that the higher grouping user of creditworthiness downloads the application is more, representing the application and the credit of userThe correlation of degree is higher, i.e., higher with the correlation of the reference feature of user；And for amusement and leisure class application, if every groupThe number of users that user downloads the application is more or less the same, then it represents that the application is relatively low with the correlation of the reference degree of user, i.e.,It is relatively low with the correlation of the reference feature of user.
When the download situation of the application is downloads the number of users of the application and do not download the number of users of the application, obtainTaking the application and the method for the correlation of the reference feature of every group of user can be：
The application and the correlation of the reference feature of every group of user are obtained according to following formula,
Wherein, i=1,2 ..., n, n represent the user's group number obtained after being grouped to the plurality of user, IViRepresenting shouldUsing the correlation of the reference feature with i-th group of user, GiRepresent to download the number of users of the application, B in i-th group of useriTableShow the number of users for not downloading the application in i-th group of user, GTRepresent to download the number of users of the application in the plurality of user,BTRepresent not download the number of users of the application in the plurality of user, " ln " represents logarithm operation symbol, and " * " represents multiplyingSymbol, "/" represent division arithmetic symbol.
2013rd, according to the application and the correlation of the reference feature of every group of user, the application and the plurality of user's are obtainedThe correlation of reference feature.
Wherein, obtaining the application and the method for the correlation of the reference feature of the plurality of user can be：According to the applicationWith the correlation of the reference feature of every group of user, being averaged for the application and the correlation of the reference feature of all grouping users is obtainedValue or maximum, the application and the correlation of the reference feature of the plurality of user are retrieved as by the average value or maximum.Certainly,The application and the plurality of use can also be obtained using other method according to the application and the correlation of the reference feature of every group of userThe correlation of the reference feature at family, the embodiment of the present invention are not especially limited to this.
In an alternative embodiment of the invention, can also be by the application and the correlation of the reference feature of all grouping usersSummation, it is retrieved as the application and the correlation of the reference feature of the plurality of user；That is, according to following formula obtain the application with it is the plurality ofThe correlation of the reference feature of user.
Wherein, IV represents the application and the correlation of the reference feature of the plurality of user.By by this application and all pointsThe summation of the correlation of the reference feature of group user, is retrieved as the application and the correlation of the reference feature of the plurality of user, energyEnough accuracys for improving the accuracy of accessed correlation, and then subsequent applications recommendation being improved.
202nd, the plurality of application is grouped according to the correlation, obtains first and apply group, the first application group includesMultiple correlations are more than the application of predetermined threshold value.
Wherein, within a preset range, the lower limit of the preset range corresponds to correlation to the predetermined threshold value for the plurality of applicationMinimum value, the upper limit are the maximum that the plurality of application corresponds to correlation, and the predetermined threshold value could be arranged in the preset rangeAny value；For example, the predetermined threshold value can be 0.2.
In embodiments of the present invention, any correlation is more than the application of the predetermined threshold value, reflects the application and is levied with userBelieve that the degree of correlation of feature is higher, dividing first into by application of the correlation more than predetermined threshold value applies group, can realize and be directed toThe first application group is individually established recommends the forecast model of probability for obtaining application, and then can realize special for user's referenceThe high application of sign degree of correlation is recommended, and reaches and improves the purpose that accuracy rate is recommended in application.
In an alternative embodiment of the invention, after being grouped according to the correlation to the multiple application, is also obtainedTwo apply group, and described second includes application of multiple correlations less than or equal to the predetermined threshold value using group.
Because any correlation is less than or equal to the application of the predetermined threshold value, the application and user's reference feature are reflectedDegree of correlation is relatively low, therefore by the way that the plurality of application is divided into one group high with reference feature degree of correlation according to the predetermined threshold valueOne group low with reference feature degree of correlation, it can realize that basis is different from reference feature degree of correlation, be every group of applicationForecast model corresponding to foundation, it is different from reference feature degree of correlation so as to reach basis, obtained using different forecast modelsThe recommendation probability of corresponding application, is recommended further according to the recommendation probability, can further be improved using the accuracy rate recommended.
203rd, group is applied for first, the first forecast model corresponding to the reference feature foundation based on the plurality of user shouldFirst forecast model is used for the recommendation probability that application is determined based on reference feature.
In embodiments of the present invention, the method for the first forecast model corresponding to the reference feature foundation based on the plurality of userCan be：Using the user characteristics of the plurality of user as training sample, forecast model corresponding to every group of application is obtained by training.Wherein, the user characteristics is in addition to including reference feature, in addition to the user such as age, sex Figure Characteristics, online interaction spyThe features such as sign, history Download History, method can be SVM (Support Vector used by establishing forecast modelMachine, SVMs), or the machine learning method such as maximum entropy or random forest, other calculate can also be usedMethod establishes forecast model, and the embodiment of the present invention is to the particular user feature as training sample and used by establishing forecast modelMethod is not construed as limiting.
It should be noted that group is applied for higher with reference feature correlation first, i.e., in the first application groupRequire higher using the creditworthiness to user, when establishing forecast model corresponding to the first application group, it is special reference will to be includedThe user characteristics of multiple users of sign establishes first forecast model as training sample so that is recommending to be somebody's turn to do to any userDuring application in the first application group, it can recommend to be adapted to the user, the i.e. user to the user according to the reference feature of the userDownload the big application of possibility.
By every group of application to being grouped according to correlation establish corresponding to forecast model, can realize according to different pre-The recommendation probability that model obtains different application is surveyed, can when carrying out further according to the recommendation probability of the different application using recommendingImprove the accuracy rate that application is recommended.
In an alternative embodiment of the invention, for being less than or equal to the application of the predetermined threshold value including multiple correlationsSecond applies group, and the method for establishing the second forecast model for obtaining the recommendation probability each applied in the second application group can be withFor：User characteristics based on the plurality of user in addition to the reference feature establishes second forecast model.
Specifically, group, i.e. application pair in the second application group are applied for relatively low with reference feature correlation secondThe creditworthiness of user requires relatively low, when establishing forecast model corresponding to the first application group, by addition to reference featureThe user characteristics of the plurality of user establishes second forecast model as training sample so as to any user recommend thisDuring application in two application groups, it can be recommended to the user according to the other users feature in addition to the reference feature of the userIt is adapted to the application of the user, i.e. the user downloads the big application of possibility.It should be noted that second forecast model can beTraditional CT R (Click-Through Rate, clicking rate) prediction model.
By establishing the first application group and with reference feature correlation relatively low higher with reference feature correlation respectivelyThe forecast model of second application group, and when being carried out to user using recommending, recommended not to user by corresponding forecast modelThe same application applied in group, can further be improved using the accuracy rate recommended, and then can be improved user and be downloaded and recommend to answerProbability, with improve institute recommendation apply conversion ratio.
In yet another embodiment of the invention, in order to further improve using the accuracy rate recommended, first group is applied for this,Further packet transaction can also be made, the group result being grouped again with basis, establish sub- forecast model corresponding to every group of application,To obtain the recommendation probability each applied in every group of application.First method being further grouped using group can be included followingTwo kinds：
The first, the correlation applied in the first application group is divided into multiple sections by default size；By this firstThe application for belonging to same section using correlation in group divides same application group into；According to group result, based on the plurality of user'sReference is characterized as sub- forecast model corresponding to every group of application foundation, the sub- forecast model of the corresponding application group in the bigger section of correlationReference feature weight it is bigger.
Wherein, the default size can be defined as any fixed value, can also be according to point of correlation in the first application groupCloth is determined, is such as determined according to minimum relatedness and maximum correlation, and the 1/3 of maximum correlation and minimum relatedness difference is determinedSize is preset for this, the first application group according to this is preset into size is further divided into three and apply group, it is of course also possible to use otherMethod determines the default size, and the embodiment of the present invention is not especially limited to this.
For example, it is 0.9 to work as maximum correlation in the first application group, when minimum relatedness is 0.6, the default size is 0.1,Size is preset according to this, the correlation applied in the first application group is divided into three groups：[0.6,0.7),[0.7,0.8),[0.8,0.9], in the first application group, correlation size [0.6,0.7) application in section is classified as one group, correlation size exists[0.7,0.8) application in section is classified as one group, and application of the correlation size in [0.8,0.9] section is classified as one group.
, can be according to application and user's reference by making further packet transaction to the first application group according to the correlationThe degree of correlation of feature is different, establishes the different forecast model of reference feature weight, obtains every group of application after being grouped againRecommend probability, the purpose for further improving the accuracy that application is recommended can be reached.
Secondth, can also be according to using class for the application in the first application group higher with reference feature correlationType does further packet to the application in first application, further according to forecast model, specific method corresponding to group result foundationCan be：Application type in the first application group is classified as one group for the application of specified type, the 3rd is obtained and applies group；By thisApplication type is classified as one group for the other kinds of application in addition to the specified type in one application group, obtains the 4th and applies group；Group is applied based on the 3rd, the user characteristics based on the plurality of user establishes the first sub- forecast model；Group is applied based on the 4th,User characteristics based on the plurality of user establishes the second sub- forecast model；Wherein, the reference feature in the first sub- forecast modelWeight is more than the reference feature weight in the second sub- forecast model.
Wherein, the specified type can be defined as any kind in the types such as net purchase class, financial class by developer, shouldSpecified type can also require to determine according to the different of user credit degree, for example, by the specified type be defined as toFamily creditworthiness requires the type of higher application.It is, of course, also possible to determine the specified type using other method, the present invention is in factExample is applied to the concrete application type of specified type and determines that method is not construed as limiting.
For example, when the specified type is financial class, application type in the first application group is returned for the application of financial classGroup is applied for the 3rd, application type in the first application group is classified as into the 4th for the application of non-financial class applies group, is establishedFirst sub- forecast model is used for the recommendation probability for obtaining financial class application, and the second sub- forecast model is used to obtain non-financial class applicationRecommendation probability.
It should be noted that it is to require higher application to user credit degree to be applied due to the financial class, for example, financingUsing, equity investment application etc., financial class application is obtained by using the first bigger sub- forecast model of reference feature weightRecommend probability, improve influence of the user credit degree to recommendation probability, recommend accuracy rate so as to reach further raising applicationPurpose.And relatively low application is required user credit degree for other kinds of, for example, the application of social class, net purchase classUsing etc., the recommendation that non-financial class application is obtained by using the second relatively small sub- forecast model of reference feature weight is generalRate, influence of the user credit degree to recommendation probability is reduced, can also reach and improve the purpose that accuracy rate is recommended in application.
By using the different forecast model of reference feature weight, the recommendation probability to different type application is obtained, especiallyApplication for specified type, by improving the weight of reference feature, when obtaining the recommendation probability of the type application, Neng GoujinOne step improves the accuracy rate for recommending application, and then causes recommended application more to meet the download demand of user, so as toReach the purpose further improved using conversion ratio.
Make the process of further packet transaction to the first application group, any of above two method side can be usedMethod is realized, other method can also be used to make further packet transaction to the first application group, the embodiment of the present invention is not made to thisIt is specific to limit.
Above-mentioned steps 201 to step 203 is to establish to correspond to for the application in different grouping and/or different types of applicationForecast model process, when being grouped according to correlation and application type simultaneously, the application after packet is established correspondingThe process of forecast model can be represented with Fig. 2 B.Specifically, each application and the reference of multiple users in multiple applications are obtainedThe correlation of feature, the plurality of application is grouped according to correlation and application type, correlation is less than or equal to default thresholdThe application of value is classified as one group, obtains second and applies group, and correlation is the application of specified type with predetermined threshold value and application type greatlyOne group is classified as, the 3rd is obtained and applies group, correlation is more than the predetermined threshold value and application type is classified as the application of non-designated typeOne group, obtain the 4th and apply group.
, therefore, can be by traditional CTR prediction models because the degree of correlation of the second application group and reference feature is minimumAs second forecast model；The first vertical sub- forecast model is set up for the 3rd application to can be used for obtaining the 3rd applicationIn any application recommendation probability, each application that can also be directed in the 3rd application group individually established for obtaining corresponding pushing awayThe forecast model of probability is recommended, the reference feature weight of each forecast model is different, and the embodiment of the present invention is not construed as limiting to this；For4th applies group, and the second sub- forecast model is established according to the user characteristics including reference feature, the second son prediction mouldType can be logistic regression forecast model, and the reference feature weight of the second sub- forecast model is less than the feature of the first sub- forecast modelWeight, or less than the minimal characteristic weight in all forecast models corresponding to the 3rd application group.
204th, the reference feature based on user to be recommended and first forecast model, recommending to the user to be recommended shouldWith.
When receiving the recommendation application acquisition request that the user to be recommended sends, server obtains the user's to be recommendedUser characteristics, and in the first forecast model that the user characteristics input step 203 of the user to be recommended is established, with obtain thisThe recommendation probability of different application, is pushed away according to the recommendation probability selection of the different application to the user to be recommended in one application groupRecommend the application in the first application group.For example, application of the probability more than 0.5 will be recommended to recommend the user to be recommended, with thisThe terminal of user to be recommended shows the recommendation application, the display mode of recommendation application can according to Apply Names initial it is suitableSequence is shown, or according to recommending the order of probability from big to small to show, can also be shown using other method, present example pairThe method of being particularly shown is not construed as limiting.
For example, when terminal detects that user to be recommended opens application and recommends the page or on application service platform to recommendingDuring the trigger action of application option, terminal to server, which is sent, recommends application to obtain request, and recommendation application obtains request can be withThe user characteristics of the user to be recommended is carried, or carries user's mark of the user to be recommended, to enable the server to rootThe user characteristics of the user to be recommended is inquired according to user's mark of the user to be recommended, and then according to the use of the user to be recommendedFamily feature, recommend application to the user to be recommended.
The recommendation probability each applied in the first application group is obtained by the first forecast model, and according to the recommendation probabilityRecommend the application in the first application group to user, it is possible to increase the accuracy rate recommended the application in the first application group.
In an alternative embodiment of the invention, after for being grouped according to correlation to the plurality of application, obtain firstUsing group, the specific method recommended the application in the first application group can be：The user of the user to be recommended is specialSign inputs first forecast model, obtains the recommendation probability each applied in the first application group, according in the first application groupThe recommendation probability each applied, recommend to recommend probability to be more than answering for predetermined probabilities in the first application group to the user to be recommendedWith.
The predetermined probabilities could be arranged to it is any be more than 0 numerical value for being less than 1, can also be obtained according to the user to be recommended selectionThe number of the recommendation application taken is determined, for example, when user selects acquisition 50 to recommend application, the predetermined probabilities are defined as0.7, when user selects to obtain 100 recommendation probability, the predetermined probabilities are defined as 0.5；It is of course also possible to use its other partyMethod determines the predetermined probabilities, and the embodiment of the present invention is not especially limited to this.
The application in the first application group is recommended by the above method, can according to the difference of the predetermined probabilities, toThe user to be recommended recommends the application of varying number, so as to better meet the recommended requirements of user.
In yet another embodiment of the invention, after being grouped according to correlation to the plurality of application, obtain secondUsing group, the specific method recommended the application in the second application group can be：By except the reference of the user to be recommendedThe user characteristics of the user to be recommended beyond feature inputs the second class forecast model, obtains each should in the second application groupRecommendation probability, according to the recommendation probability each applied in the second application group, recommend this second should to the user to be recommendedWith the application for recommending probability to be more than the predetermined probabilities in group.
In embodiments of the present invention, the application in the first application group only can be recommended to user, it is less to recommend to userQuantity, with the higher application of reference feature degree of correlation so that user more clearly can check and download recommended answerWith；Further, when receiving user and obtaining the request for more recommending application, then answering into user's recommendation the second application groupWith to meet that user obtains more the needs of recommending application；It is of course also possible to when receiving recommendation application acquisition request, togetherWhen to user recommend first application group and second application group in application, will corresponding to different application group recommend application use pagingThe method of display is shown, to distinguish the application recommended using different forecast models so that while the page is cleaner and tidier, Xiang YongThe page is recommended in displaying information content bigger application in family.
In an alternative embodiment of the invention, the page is recommended to show specified function choosing-item in the application, this specifies work(Energy option is used to obtain the recommendation application in the first application group, and this specifies the viewing area annex of function choosing-item can be with display remindingInformation, it is special with user's reference that the prompt message, which is used to prompt user to trigger this to specify the recommendation accessed by function choosing-item to apply,Levy the big application of correlation, such as financial class, the application of net purchase class.When terminal detects that user specifies the triggering of function choosing-item to thisDuring operation, send to obtain to server and specify the request for recommending application, to cause server to be pushed away according to the request using the firstMethod is recommended to recommend to apply to user.
Recommend to apply by the above method, it is possible to increase the specific aim that application is recommended, and then can recommend more to accord with to userThe application of user's request is closed, so as to reach the purpose improved using conversion ratio.
By according to the predetermined threshold value by the plurality of application be divided into reference feature degree of correlation it is high first application group andSecond low with reference feature degree of correlation applies group, is obtained according to forecast model corresponding with every group of application every in every group of applicationThe recommendation probability of individual application, and carried out according to the recommendation probability using recommendation, it is possible to increase the success rate that application is recommended.
It should be noted that for first making the result after further packet transaction using group to this, acquisition is grouped againIt is every in the method for the recommendation probability each applied in every group of application afterwards, with the above-mentioned acquisition first application group and the second application groupThe method of the recommendation probability of individual application similarly, does not repeat herein.
What the embodiment of the present invention was provided applies recommendation method, is used to represent application and the reference feature of user by obtainingDegree of correlation correlation, and according in multiple applications each application and the reference feature of multiple users correlation, pair withFirst that the reference feature degree of correlation of user is high applies group, the first prediction established using the reference feature based on multiple usersModel, the recommendation probability each applied in the first application group is obtained, is carried out according to the recommendation probability using recommendation, it is possible to increaseTo the accuracy rate in the first application group using recommendation so that the application recommended more meets the download demand of user, so as toEnough conversion ratios for improving application；By using the different forecast model of reference feature weight, obtain and different type application is pushed awayProbability is recommended, can further improve a pair recommendation accuracy rate for the application different from reference feature degree of correlation.
Fig. 3 is that one kind provided in an embodiment of the present invention applies recommendation apparatus block diagram.Reference picture 3, the device include software phaseClosing property acquisition module 301, grouping module 302, model building module 303 and recommending module 304.
Correlation acquisition module 301, for obtaining each application and the phase of the reference feature of multiple users in multiple applicationsGuan Xing；
Grouping module 302, for the correlation that is got according to the correlation acquisition module 301 to the multipleUsing being grouped, obtain first and apply group, described first includes application of multiple correlations more than predetermined threshold value using group；
Model building module 303, first for being obtained for the grouping module 302 applies group, based on the multipleFirst forecast model corresponding to the reference feature foundation of user, first forecast model are used to determine application based on reference featureRecommendation probability；
Recommending module 304, the institute established for the reference feature based on user to be recommended and the model building module 303The first forecast model is stated, recommends application to the user to be recommended.
In the first possible implementation provided by the invention, the recommending module 304 is used for：
The reference feature of the user to be recommended is inputted into first forecast model, obtained every in the first application groupThe recommendation probability of individual application, according to the recommendation probability each applied in the described first application group, recommend to the user to be recommendedProbability is recommended to be more than the application of predetermined probabilities in the first application group.
In second provided by the invention possible implementation, the model building module 303 is additionally operable to：
For applying group by being grouped obtain second, the use based on the multiple user in addition to the reference featureFamily feature establishes the second forecast model, and second forecast model is used for true based on the user characteristics in addition to the reference featureSurely the recommendation probability applied, described second includes application of multiple correlations less than or equal to the predetermined threshold value using group.
In the third possible implementation provided by the invention, the recommending module 304 is used for：
By the user characteristics of the user to be recommended in addition to the reference feature of the user to be recommended input described theTwo forecast models, the recommendation probability each applied in the second application group is obtained, each should according in the described second application groupRecommendation probability, recommend to recommend probability to be more than answering for the predetermined probabilities in the second application group to the user to be recommendedWith.
In the 4th kind of possible implementation provided by the invention, the model building module 303 is used for：
Application type in described first application group is classified as one group for the application of specified type, the 3rd is obtained and applies group；
Application type in described first application group is classified as one for the other kinds of application in addition to the specified typeGroup, obtain the 4th and apply group；
Group is applied based on the described 3rd, the user characteristics based on the multiple user establishes the first sub- forecast model；
Group is applied based on the described 4th, the user characteristics based on the multiple user establishes the second sub- forecast model；
Wherein, the reference feature weight in the described first sub- forecast model is more than the reference in the described second sub- forecast modelFeature weight.
In the 5th kind of possible implementation provided by the invention, the model building module 303 is additionally operable to：
The correlation applied in described first application group is divided into multiple sections by default size；
The application that correlation in described first application group is belonged to same section divides same application group into；
According to group result, the reference based on the multiple user is characterized as son prediction mould corresponding to every group of application foundationType, the reference feature weight of the sub- forecast model of the corresponding application group in the bigger section of correlation are bigger.
In the 6th kind of possible implementation provided by the invention, the correlation acquisition module 301 is used for：
For any application in the multiple application, according to the reference feature of the multiple user, to the multiple useFamily is grouped；
The download situation of the application and the multiple user are obtained to the download situation of the application according to every group of userTake the application and the correlation of the reference feature of every group of user；
According to the application and the correlation of the reference feature of every group of user, the application is obtained with the multiple user'sThe correlation of reference feature.
In the 7th kind of possible implementation provided by the invention, the correlation acquisition module 301 is used for：
The application and the correlation of the reference feature of every group of user are obtained according to following formula：
Wherein, i=1,2 ..., n, n represent the user's group number obtained after being grouped to the multiple user, IViRepresentThe application and the correlation of the reference feature of i-th group of user, GiRepresent to download the user of the application in i-th group of userQuantity, BiRepresent not download the number of users of the application, G in i-th group of userTRepresent to download institute in the multiple userState the number of users of application, BTRepresent not download the number of users of the application in the multiple user.
In the 8th kind of possible implementation provided by the invention, the correlation acquisition module 301 is used for according to following formulaObtain the application and the correlation of the reference feature of the multiple user：
Wherein, IV represents the application and the correlation of the reference feature of the multiple user.
It should be noted that：The application recommendation apparatus that above-described embodiment provides is when recommending to apply, only with above-mentioned each functionThe division progress of module, can be as needed and by above-mentioned function distribution by different function moulds for example, in practical applicationBlock is completed, i.e., the internal structure of equipment is divided into different functional modules, to complete all or part of work(described aboveEnergy.In addition, the application recommendation apparatus that above-described embodiment provides recommends embodiment of the method to belong to same design with application, it is specific realExisting process refers to embodiment of the method, repeats no more here.
Fig. 4 is a kind of structural representation of device 400 provided in an embodiment of the present invention.For example, device 400 can be providedFor a server.Reference picture 4, device 400 include processing component 422, and it further comprises one or more processors, Yi JiyouMemory resource representated by memory 432, can be by the instruction of the execution of processing component 422, such as application program for storing.The application program stored in memory 432 can include it is one or more each correspond to the module of one group of instruction.In addition, processing component 422 is configured as execute instruction, recommend method to perform above-mentioned application.
Device 400 can also include the power management that a power supply module 426 is configured as performs device 400, and one hasLine or radio network interface 450 are configured as device 400 being connected to network, and input and output (I/O) interface 458.DressPutting 400 can operate based on the operating system for being stored in memory 432, such as Windows ServerTM, Mac OS XTM,UnixTM,LinuxTM, FreeBSDTMIt is or similar.
One of ordinary skill in the art will appreciate that hardware can be passed through by realizing all or part of step of above-described embodimentTo complete, by program the hardware of correlation can also be instructed to complete, described program can be stored in a kind of computer-readableIn storage medium, storage medium mentioned above can be read-only storage, disk or CD etc..
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit andWithin principle, any modification, equivalent substitution and improvements made etc., it should be included in the scope of the protection.