CN109460512A - Recommendation information processing method, device, equipment and storage medium - Google Patents

Recommendation information processing method, device, equipment and storage medium Download PDF

Info

Publication number
CN109460512A
CN109460512A CN201811251040.8A CN201811251040A CN109460512A CN 109460512 A CN109460512 A CN 109460512A CN 201811251040 A CN201811251040 A CN 201811251040A CN 109460512 A CN109460512 A CN 109460512A
Authority
CN
China
Prior art keywords
classification
score
target object
preset
initial
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201811251040.8A
Other languages
Chinese (zh)
Other versions
CN109460512B (en
Inventor
梁超
崔瑞
李天浩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Beijing Co Ltd
Original Assignee
Tencent Technology Beijing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Beijing Co Ltd filed Critical Tencent Technology Beijing Co Ltd
Priority to CN201811251040.8A priority Critical patent/CN109460512B/en
Publication of CN109460512A publication Critical patent/CN109460512A/en
Application granted granted Critical
Publication of CN109460512B publication Critical patent/CN109460512B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The application is to belong to information technology field about a kind of recommendation information processing method, device, equipment and storage medium.The described method includes: obtaining the attribute information and corresponding initial classification score collection of target object；Determine target object in the score of preset each candidate classification using preset prediction model based on attribute information and initial classification score collection；According to target object in the score of preset each candidate classification, recommendation information is pushed to target object.Thus, realize the attribute information and initial classification score collection according to target object, recommend its interested information for target object, it is determined since initial classification score collection can obtain diversity according to preset classification, alternatively, can determine according to the historical behavior data of target object, accurate information recommendation is carried out so as to be embodied as any object, the clicking rate for improving recommendation information, improves user experience.

Description

Recommendation information processing method, device, equipment and storage medium
Technical field
This application involves information technology field, in particular to a kind of recommendation information processing method, device, equipment and storage are situated between Matter.
Background technique
Personalized information recommending technology can recommend its interested information to user, and therefore, the technology is gradually in network It is more and more applied in access.
In the related technology, usually by collaborative filtering mode, for user recommend with the interested information of its similar users, or Person is that user recommends information similar with its interested information, still, by means of which for user's recommendation information when, if with Family is new user, due to that can not obtain information related with user and its interested information, to can not accurately push away for user Information is recommended, i.e., the application range of this information recommendation mode is small, poor user experience.
Summary of the invention
The application proposes a kind of recommendation information processing method, device, equipment and storage medium, to solve in the related technology, It can not be the accurate recommendation information of user, i.e., if user is new user when being user's recommendation information by collaborative filtering mode The technical issues of application range of this information recommendation mode is small, poor user experience.
The application one side embodiment provides a kind of recommendation information processing method, this method comprises: obtaining target object Attribute information and corresponding initial classification score collection；Based on the attribute information and the initial classification score collection, using default Prediction model, determine the target object in the score of preset each candidate classification, the preset prediction model, for benefit It is got with attribute information and classification the score training of sample object, wherein the classification score of the sample object concentrates packet The quantity of the classification score contained meets preset condition；According to the target object preset each candidate classification score, to The target object pushes recommendation information.
The application another aspect embodiment provides a kind of recommendation information processing unit, which includes: the first acquisition module, For obtaining the attribute information and corresponding initial classification score collection of target object；First determining module, for being based on the category Property information and the initial classification score collection determine the target object in preset each time using preset prediction model Select the score of classification, the preset prediction model, to get using the attribute information of sample object and the training of classification score , wherein the classification score of the sample object concentrates the quantity for the classification score for including to meet preset condition；Pushing module, For the score according to the target object in preset each candidate classification, Xiang Suoshu target object push recommendation information.
The another aspect embodiment of the application provides a kind of computer equipment, including memory, processor and is stored in storage On device and the computer program that can run on a processor, when the processor executes described program, previous embodiment institute is realized The recommendation information processing method stated.
The application another further aspect embodiment provides a kind of computer readable storage medium, is stored thereon with computer program, When the program is executed by processor, recommendation information processing method described in previous embodiment is realized.
Recommendation information processing method, device, equipment and storage medium provided by the embodiments of the present application are obtaining target object Attribute information and corresponding initial classification score collection after, attribute information and initial classification score collection can be based on, using default Prediction model, determine target object in the score of preset each candidate classification, thus according to target object preset every The score of a candidate's classification, pushes recommendation information to target object.Hereby it is achieved that according to the attribute information of target object and just Beginning classification obtains diversity, recommends its interested information for target object, since initial classification score collection can be according to preset class Mesh obtain diversity determine, alternatively, can according to the historical behavior data of target object determine, so as to be embodied as any object into The accurate information recommendation of row, improves the clicking rate of recommendation information, improves user experience.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not The application can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the application Example, and together with specification it is used to explain the principle of the application.
Fig. 1 is the flow diagram according to the recommendation information processing method shown in one exemplary embodiment of the application；
Fig. 2 is the exemplary diagram according to the recommendation information processing method shown in one exemplary embodiment of the application；
Fig. 3-4 is the application scenario diagram according to the recommendation information processing method shown in one exemplary embodiment of the application；
Fig. 5 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application；
Fig. 6 is the application scenario diagram according to the recommendation information processing method shown in another exemplary embodiment of the application；
Fig. 7 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application；
Fig. 8-9 is the application scenarios according to the recommendation information processing method shown in another exemplary embodiment of the application Figure；
Figure 10 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application；
Figure 11 is the exemplary diagram according to the preset prediction model training process shown in one exemplary embodiment of the application；
Figure 12 is the structural schematic diagram according to the recommendation information processing unit shown in one exemplary embodiment of the application；
Figure 13 is the structural schematic diagram according to the computer equipment shown in one exemplary embodiment of the application.
Through the above attached drawings, it has been shown that the specific embodiment of the application will be hereinafter described in more detail.These attached drawings It is not intended to limit the range of the application design in any manner with verbal description, but is by referring to specific embodiments Those skilled in the art illustrate the concept of the application.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment Described in embodiment do not represent all embodiments consistent with the application.On the contrary, they be only with it is such as appended The example of the consistent device and method of some aspects be described in detail in claims, the application.
Each embodiment of the application is directed in the related technology, by collaborative filtering mode, when being user's recommendation information, if user It then can not be the accurate recommendation information of user, i.e., the application range of this information recommendation mode is small, poor user experience for new user Problem proposes a kind of recommendation information processing method.
Recommendation information processing method provided by the embodiments of the present application is obtaining the attribute information of target object and corresponding After initial classification score collection, target can be determined using preset prediction model based on attribute information and initial classification score collection Object preset each candidate classification score, thus according to target object preset each candidate classification score, to Target object pushes recommendation information, wherein preset prediction model, to obtain diversity using the attribute information and classification of sample object What training obtained, wherein the classification score of sample object concentrates the quantity for the classification score for including to meet preset condition.As a result, The attribute information and initial classification score collection according to target object are realized, recommends its interested information for target object, by Diversity can be obtained according to preset classification in initial classification score collection to determine, alternatively, can be according to the historical behavior of target object Data determine, carry out accurate information recommendation so as to be embodied as any object, improve the clicking rate of recommendation information, improve User experience.
With reference to the accompanying drawing, recommendation information processing method provided by the present application, device, equipment and storage medium are carried out detailed It describes in detail bright.
Fig. 1 is combined first, and recommendation information processing method provided by the embodiments of the present application is described in detail.
Fig. 1 is the flow diagram according to the recommendation information processing method shown in one exemplary embodiment of the application.
As shown in Figure 1, the recommendation information processing method, may comprise steps of:
Step 101, the attribute information and corresponding initial classification score collection of target object are obtained.
Specifically, recommendation information processing method provided by the embodiments of the present application, can be pushed away by provided by the embodiments of the present application Recommend information processing unit execution.Wherein, the recommendation information processing unit, can be configured in arbitrarily can provide video for user Or in the application program of the information such as text, to carry out accurate information recommendation for any object, the click of recommendation information is improved Rate.Wherein, the computer equipment in the present embodiment can be any hardware device having data processing function, such as intelligently Mobile phone, personal digital assistant, tablet computer, desktop computer etc..
Wherein, target object is the object that recommendation information processing unit will push recommendation information to it.The category of target object Property information, may include the attributes such as gender, the age of target object.It should be noted that target object, can be history registry Object is also possible to register object for the first time, herein with no restriction.
Initial classification score is concentrated, target object the obtaining in each classification that can be obtained including recommendation information processing unit Point.Wherein, classification can be amusement class, class of making laughs, news category etc..It should be noted that target object is in each classification Score can be the numerical value between -1 to 1.
When specific implementation, when due to user's registration, the information such as gender, age may have been filled in, then, for The target object of the information such as gender, age is filled in, the content that can directly fill according to it obtains the attribute letter of target object Breath, and the target object for not filling in the information such as gender, age, can by pre-set default property information, as The attribute information of target object.
It is understood that during it browses information, usually feeling to it for history registry object The various information of interest has carried out multiple click.In the embodiment of the present application, row is clicked by the history to history registry object For etc. data analyzed, can determine that history registry object in the score of each classification, and then determines history registry object pair The initial classification score collection answered.That is, if target object is history registry object, it can historical behavior data to target object Dissection process is carried out, determines the corresponding initial classification score collection of target object.
Wherein, historical behavior data, may include target object within a preset period of time, to the corresponding information of each classification The data such as number of clicks.Wherein, preset time period can according to need setting.For example, can be history registry object from for the first time The registration moment started to the period between current time, alternatively, whens waiting in being also possible to nearest one week or being one month nearest Between section.
In the exemplary embodiment, it is believed that target object within a preset period of time, to the corresponding information of certain classification Number of clicks is higher, then the corresponding score of the classification is higher.
In addition, registering for object for for the first time, click behavior may not generated to nothing due to registering object for the first time Method determines the corresponding initial classification score collection of registration object for the first time using the historical behavior data for registering object for the first time.So, exist In the embodiment of the present application, a classification can be preset and obtain diversity, as the initial classification score collection for registering object for the first time. That is, can determine that preset classification obtains diversity is the corresponding initial classes of target object if target object is to register object for the first time Mesh obtains diversity.
Wherein, preset classification obtains diversity, can be according to the determination of the historical behavior data of history registry object, can also Being determined according to other way, herein with no restriction.
Step 102, target object is determined using preset prediction model based on attribute information and initial classification score collection In the score of preset each candidate classification.
In the embodiment of the present application, preset prediction model, to obtain diversity using the attribute information and classification of sample object What training obtained, the classification score of sample object concentrates the quantity for the classification score for including to meet preset condition.
Wherein, preset condition can according to need setting.
For example, in the embodiment of the present application, preset condition can be set are as follows: classification score concentrates the classification score for including Quantity is greater than preset threshold, concentrates the classification score for including so as to from history registry object, choose classification score in advance Quantity be greater than preset threshold object then obtain diversity using the attribute information of sample object and classification as sample object, Initial model is trained, preset prediction model is obtained.Wherein, initial model can be neural network model or other Model.It should be noted that the training generating process of preset prediction model will be illustrated in the following embodiments, this Place is not described.
Further, after obtaining preset prediction model, can be based on attribute information and initial classification score collection, by with Lower step 102a-102c determines target object in the score of preset each candidate classification.
Step 102a determines that the corresponding attribute vector of attribute information, and initial classification score concentrate each classification corresponding Classification vector.
Specifically, it may be predetermined that the corresponding classification vector of each classification, thus obtaining initial classification score collection Afterwards, it can obtain initial classification score from the corresponding classification vector of predetermined each classification and concentrate each classification pair The classification vector answered.It should be noted that the dimension of classification vector, can according to need any setting, herein with no restriction.
In the exemplary embodiment, each classification can be preset and respectively corresponds unique label, for example, class of making laughs is corresponding Marked as 1, amusement class is corresponding marked as 2, and it is right respectively to obtain each label for and the random initializtion by meeting Gaussian Profile The initial vector answered, is then again trained initial model, during generating preset prediction model, to each initial vector It is updated, obtains the corresponding classification vector of each label.After obtaining initial classification score collection, if initial classification score collection In include classification be make laughs class and amusement class, then can according to label 1 and label 2, respectively from predetermined each classification to In amount, classification vector corresponding with label 1 and label 2 is searched, so that obtaining initial classification score concentrates make laughs class and joy The corresponding classification vector of happy class.
It should be noted that initial classification score concentrates the corresponding classification vector of each classification, any way can be passed through It determines, only needs uniquely characterize corresponding classification.The above-mentioned initial classification score of determination concentrates the corresponding class of each classification The example of mesh vector, only schematically illustrates, and should not be understood as the limitation to technical scheme, and those skilled in the art exist It on the basis of this, can according to need, any setting determines that initial classification score concentrates the side of the corresponding classification vector of each classification Formula, the application to this with no restriction.
In addition, determining the mode of the corresponding attribute vector of attribute information, each classification is concentrated with the initial classification score of determination The mode of corresponding classification vector is similar, and details are not described herein again.
Step 102b determines target object according to the corresponding classification vector of each classification and the corresponding score of each classification Corresponding total classification vector.
Specifically, after determining each corresponding classification vector of classification and the corresponding score of each classification, it can be by each class The corresponding classification vector of mesh is so corresponding with the classification that be allocated as accumulating, then the result summation after work product is averaging again, to obtain The corresponding total classification vector of target object.
As an example it is assumed that it includes three classifications that the corresponding initial classification score of target object, which is concentrated, each classification is corresponding Score be respectively w1, w2, w3, corresponding classification vector is respectively a1, a2, a3, then the corresponding total classification vector of target object For (a1*w1+a2*w2+a3*w3)/3.
Step 102c is based on the corresponding total classification vector of target object and attribute vector, using preset prediction model, really Set the goal object preset each candidate classification score.
Wherein, candidate classification, all classifications that can recommend for recommendation information processing unit.It should be noted that target Object is scored at the numerical value between -1 to 1 in preset each candidate classification.
Specifically, obtaining preset prediction model to be based on the corresponding total classification vector of target object and attribute vector An input vector, and then utilize preset prediction model, determine target object in the score of preset each candidate classification, Can total classification vector corresponding to target object and attribute vector carry out certain processing, such as by determining target object pair The total classification vector and attribute vector answered are spliced, so that be spliced into vector is inputted preset prediction model, i.e., Using preset prediction model, determine target object in the score of preset each candidate classification.
Such as, it is assumed that it is preset candidate classification include varieties of society class, obvious information class, publish in instalments animated type, cross-talk play class, 32 classifications such as TV play film clips class, traffic class.Assuming that target object A is to register object for the first time, the age of target object is 30 In year, gender is male, obtains diversity according to preset classification, determines that the corresponding initial classification score of target object A is concentrated, society hundred State class is corresponding to be scored at 0.235323, and star's information class is corresponding to be scored at 0.323445, and legal system class is corresponding to be scored at 0.353222, and age corresponding attribute vector is B1=[0 01 0], the corresponding attribute vector of gender is B2=[0 10 1], the corresponding classification vector of varieties of society class is C1=[0 0000101 1], the corresponding classification vector of star's information class For C2=[0 1000001 1], the corresponding classification vector of legal system class is C3=[1 0000101 0], then may be used With determine the corresponding total classification vector of target object A for C=(C1*0.235323+C2*0.323445+C3*0.353222)/3, The corresponding total classification vector C of target object A and attribute vector B1 and B2 are spliced into a vector, input preset prediction mould Type, then score of the available target object A as shown in Figure 2 in preset each candidate classification.
It is understood that the initial classification score of target object concentrates the classification score for including in practice Quantity may be very much.In order to reduce data volume to be treated in recommendation information treatment process, treatment effeciency, Jin Erti are improved Height can be concentrated to the speed of target object push recommendation information merely with initial classification score in the embodiment of the present application The attribute information of the corresponding score of part classification and target object determines target object default using preset prediction model Each of candidate classification score.It is obtained for example, the initial classification score concentration of target object includes that classification 1-40 is corresponding Point, i.e., the quantity of classification score is 40, in the embodiment of the present application, 5 classifications that can be concentrated merely with initial classification score Score (such as the corresponding score of classification 1-5) and the attribute information of target object determine mesh using preset prediction model Object is marked in the score of preset each candidate classification.
That is, in the embodiment of the present application, a first threshold can be preset, if the initial classes of target object Mesh score concentrates the quantity for the classification score for including to be greater than first threshold, for example, first threshold is 5, the initial classes of target object It includes 40 classification scores that mesh score, which is concentrated, is greater than first threshold 5, then can concentrate extraction section classification from initial classification score Corresponding score, to based on attribute information and refer to classification score collection, utilizes preset prediction as reference classification score collection Model determines target object in the score of preset each candidate classification.That is, if the initial classification score of target object concentrates packet The quantity of the classification score contained is greater than first threshold
According to default rule, concentrates and extracted with reference to classification score collection from initial classification score；
Correspondingly, step 102 can be accomplished by the following way:
Determine target object preset using preset prediction model based on attribute information and with reference to classification score collection The score of each candidate's classification.
Wherein, first threshold can according to need setting.Default rule can be by way of randomly selecting, from Initial classification score concentrates the corresponding score of classification for extracting preset quantity, as reference classification score collection.Wherein, present count Amount, i.e., the quantity for the classification score concentrated with reference to classification score, can be less than or equal to first threshold.
It extracts specifically, being concentrated from initial classification score with reference to after classification score collection, can first determining attribute information correspondence Attribute vector, and the corresponding classification vector of each classification is concentrated with reference to classification score, then according to the corresponding class of each classification Mesh vector and the corresponding score of each classification determine the corresponding total classification vector of target object, then corresponding based on target object Total classification vector and attribute vector determine target object obtaining in preset each candidate classification using preset prediction model Point.Concrete implementation process and principle, similar with process described in step 102a-102c, details are not described herein again.
Step 103, recommendation is pushed to target object in the score of preset each candidate classification according to target object Breath.
Specifically, determining that target object, can be by the former of highest scoring after the score of preset each candidate classification Candidate classification corresponding several information of the corresponding information of a candidate's classification as recommendation information or highest scoring As recommendation information, it is pushed to target object.That is, can be difference to a plurality of recommendation information that target object pushes The corresponding a plurality of information of candidate classification is also possible to the corresponding a plurality of information of same candidate classification, herein with no restriction.
It should be noted that the information bar number of recommendation information, can be held according to the data volume and target object of each information The factors such as the data volume that can show of screen of mobile phone, apparatus such as computer determine.
It is understood that by recommendation information processing method provided by the embodiments of the present application, if target object is for the first time Object is registered, then browses first screen information in target object, and shield and do not click on the corresponding information of any classification in head, but directly It clicks after refreshing, diversity can be obtained according to preset classification, determine that initial classification score collection, such as preset classification score are concentrated Class of making laughs is corresponding to be scored at 0.353222, and news category is corresponding to be scored at 0.323222, then can determine initial classification score Concentrating includes make laughs class and its corresponding score 0.353222, news category and its corresponding score 0.323222, is then based on mesh The attribute information and initial classification score collection for marking object determine target object preset each using preset prediction model The score of candidate classification, to be target object in next screen according to target object in the score of preset each candidate classification Push recommendation information.
If target object shields information in browsing head, and after head screen clicks the corresponding information of certain classifications, then clicks brush Newly, then initial classification score collection can be determined according to the click behavior of target object, is then based on the attribute information of target object And initial classification score collection, using preset prediction model, determine target object in the score of preset each candidate classification, from And according to target object in the score of preset each candidate classification, it is that target object pushes recommendation information in next screen.Wherein, Target object is after head screen clicks the corresponding information of certain classifications, then initial classification score is concentrated, such purpose score can be with For the value greater than 0, the score of the corresponding classification of the information that other target objects do not click on can be the value less than 0.
Further, by according to target object in the click behavior of first screen, the second screen, can be target pair in third screen As pushing recommendation information.When the quantity for the classification score for including in the initial classification score concentration of target object as a result, is smaller, base Target pair is determined that is, using preset prediction model in the attribute information of target object and corresponding initial classification score collection As the score in preset each candidate classification, and then recommendation information is pushed to target object, until the initial classes of target object When mesh score concentrates the quantity for the classification score for including to be greater than first threshold, then can attribute information based on target object and from Initial classification score concentrates the reference classification score collection extracted to determine target object preset using preset prediction model The score of each candidate's classification, and then recommendation information is pushed to target object.
Below with reference to Fig. 3 and application scenario diagram shown in Fig. 4, to recommendation information processing method provided by the embodiments of the present application It is illustrated.Wherein, client can provide the information such as video or text for target object.
As shown in figure 3, (the step 1) after target object clicks refreshing in the client, the place of recommendation information processing unit Reason device can obtain the attribute information and corresponding initial classification score collection of target object first according to the data stored in memory (step 2).It wherein, can history row to the target object stored in memory if target object is history registry object Dissection process is carried out for data, to determine the corresponding initial classification score collection of target object, if target object is registration pair for the first time As preset classification can then be obtained to the initial classification score collection that diversity is determined as target object.Then recommendation information processing dress The processor set can attribute information and corresponding initial classification score collection based on the target object of acquisition, using preset pre- Model is surveyed, determines target object in the score (step 3) of each candidate classification, further according to target object in preset each candidate The score of classification obtains recommendation information (step 4) from memory, and is pushed to the client (step 5) where target object.
Refresh (step 1) as shown in figure 4, clicking in the client in target object, and the place of recommendation information processing unit Device is managed according to the data stored in memory, obtains (step after the attribute information and corresponding initial classification score collection of target object It is rapid 2), however, it is determined that the initial classification score of target object concentrates the quantity for the classification score for including to be greater than first threshold, then can be with According to default rule, concentrates and extracted with reference to classification score collection (step 3) from initial classification score.For example, first threshold is 5, Default rule is to randomly select from initial classification score concentration with reference to classification score collection, wherein concentrates packet with reference to classification score The classification score of the quantity containing first threshold, it is determined that the initial classification score of target object concentrates the quantity for the classification score for including It is 40, when being greater than first threshold 5, then can randomly selects 5 classifications from the corresponding score of 40 classifications and respectively correspond Score, as reference classification score collection.Then the processor of recommendation information processing unit can be based on the attribute of target object Information and reference classification score collection determine target object in the score (step of each candidate classification using preset prediction model 4) recommendation information (step 5), is obtained from memory in the score of preset each candidate classification further according to target object, and Client (the step 6) being pushed to where target object.
Recommendation information processing method provided by the embodiments of the present application is obtaining the attribute information of target object and corresponding After initial classification score collection, target can be determined using preset prediction model based on attribute information and initial classification score collection Object preset each candidate classification score, thus according to target object preset each candidate classification score, to Target object pushes recommendation information.Hereby it is achieved that being target according to the attribute information of target object and initial classification score collection Its interested information of object recommendation is determined since initial classification score collection can obtain diversity according to preset classification, alternatively, can To be determined according to the historical behavior data of target object, accurate information recommendation is carried out so as to be embodied as any object, is mentioned The high clicking rate of recommendation information, improves user experience.
It, can through above-mentioned analysis it is found that after the attribute information and corresponding initial classification score collection for obtaining target object To determine target object in preset each time using preset prediction model based on attribute information and initial classification score collection The score of classification is selected, to push recommendation to target object in the score of preset each candidate classification according to target object Breath.In a kind of possible way of realization, can also further it be classified to each classification, for example, classification is when making laughs class, also The class that can will make laughs is further divided into second levels classification, the application such as cross-talk play, hilarious animal, daily joke and is referred to as label, To which in the embodiment of the present application, the attribute information, initial classification score collection, each label for being also based on target object are corresponding Score determines target object in the score of preset each candidate classification, and then to target object using preset prediction model Push recommendation information.The recommendation information processing method of the application is carried out furtherly for above situation below with reference to Fig. 5 It is bright.
Fig. 5 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application.
As shown in figure 5, the recommendation information processing method may comprise steps of:
Step 201, the attribute information, corresponding initial classification score collection and initial labels for obtaining target object obtain diversity.
Wherein, it includes the target object that can obtain of recommendation information processing unit in each label that initial labels score, which is concentrated, Score.Wherein, label is the classification for obtain after further classifying to classification.
Specifically, the acquisition modes of the attribute information of target object and corresponding initial classification score collection, are referred to The associated description of step 101 in embodiment is stated, details are not described herein again.
It is understood that during it browses information, usually feeling to it for history registry object The various information of interest has carried out multiple click.In the embodiment of the present application, row is clicked by the history to history registry object For etc. data analyzed, can determine that history registry object in the score of each label, and then determines history registry object pair The initial labels answered obtain diversity.That is, if target object is history registry object, it can historical behavior data to target object Dissection process is carried out, determines that the corresponding initial labels of target object obtain diversity.
Wherein, historical behavior data, may include target object within a preset period of time, to the corresponding information of each label The data such as number of clicks.Wherein, preset time period can according to need setting.For example, can be history registry object from for the first time The registration moment started to the period between current time, alternatively, whens waiting in being also possible to nearest one week or being one month nearest Between section.
It is registered for object for for the first time, may not generate click behavior due to registering object for the first time, thus can not benefit With the historical behavior data for registering object for the first time, determine that the corresponding initial labels of registration object obtain diversity for the first time.So, in this Shen It please can preset a label and obtain diversity in embodiment, obtain diversity as the initial labels for registering object for the first time.That is, if Target object is to register object for the first time, then can determine that preset label obtains diversity is the corresponding initial labels score of target object Collection.
Further, the attribute information, corresponding initial classification score collection and initial labels score of target object are obtained After collection, diversity can be obtained based on attribute information, initial classification score collection and initial labels, using preset prediction model, determined Score of the target object in preset each candidate classification.
Specifically, attribute information, initial classification score collection and initial labels can be based on by following steps 202-205 Diversity is obtained, using preset prediction model, determines target object in the score of preset each candidate classification.
Step 202, determine that the corresponding attribute vector of attribute information, initial classification score concentrate the corresponding classification of each classification Vector and initial labels score concentrate the corresponding label vector of each label.
Step 203, according to the corresponding classification vector of each classification and the corresponding score of each classification, target object pair is determined The total classification vector answered.
Step 204, according to the corresponding label vector of each label and the corresponding score of each label, target object pair is determined The total label vector answered.
Specifically, determining that the corresponding attribute vector of attribute information, initial classification score concentrate the corresponding classification of each classification The process of vector, and according to the corresponding classification vector of each classification and the corresponding score of each classification, determine that target object is corresponding Total classification vector process, be referred to the associated description of step 102 in above-described embodiment, details are not described herein again.
In addition, determining that initial labels score concentrates the corresponding label vector of each label, and corresponding according to each label Label vector and the corresponding score of each label, determine the process of the corresponding total label vector of target object, with determining initial classes Mesh score concentrates the corresponding classification vector of each classification, and corresponding according to the corresponding classification vector of each classification and each classification Score determines that the process of the corresponding total classification vector of target object is similar, and details are not described herein again.
In addition, above-mentioned steps 203 and step 204 can first carry out step 203, then hold when being executed without sequencing Row step 204 can also first carry out step 204, then execute step 203, can also be performed simultaneously step 203 and step 204, Step 203 and step 204 need after step 202, execute before step 205, the application to this with no restriction.
Step 205, mesh is determined using preset prediction model based on attribute vector, total classification vector and total label vector Object is marked in the score of preset each candidate classification.
Specifically, the determining corresponding total classification vector of target object, total label vector and attribute vector are spliced into one A vector, that is, using preset prediction model, determines target object preset every as the input of preset prediction model The score of a candidate's classification.
Step 206, recommendation is pushed to target object in the score of preset each candidate classification according to target object Breath.
Wherein, the specific implementation process and principle of above-mentioned steps 206, is referred to the detailed description of above-described embodiment, this Place repeats no more.
Below with reference to application scenario diagram shown in fig. 6, recommendation information processing method provided by the embodiments of the present application is carried out Explanation.Wherein, client can provide the information such as video or text for target object.
As shown in fig. 6, (the step 1) after target object clicks refreshing in the client, the place of recommendation information processing unit Manage device first according to the data stored in memory, can obtain target object attribute information and corresponding initial classification score Collection, initial labels obtain diversity (step 2).Then the processor of recommendation information processing unit can be based on attribute information and corresponding Initial classification score collection, initial labels obtain diversity, using preset prediction model, determine target object in preset each candidate Score (the step 3) of classification obtains from memory and recommends further according to target object in the score of preset each candidate classification Information (step 4), and it is pushed to the client (step 5) where target object.
Diversity is obtained by attribute information, initial classification score collection and initial labels based on target object, utilization is preset Prediction model determines target object in the score of preset each candidate classification, and then according to target object preset each The score of candidate classification pushes recommendation information to target object, further improves the accurate of the recommendation information pushed to user The clicking rate of property and recommendation information, improves user experience.
It should be noted that in the embodiment of the present application, label can also be further subdivided into multiple three-level classifications, it will Multiple three-level classifications are further subdivided into multiple level Four classifications, etc., thus the attribute information based on target object, it is corresponding just Beginning classification obtains diversity, initial labels obtain diversity, initial three-level classification obtains diversity, initial level Four classification obtains diversity etc., using default Prediction model, determine target object in the score of preset each candidate classification, and then according to target object preset every The score of a candidate's classification, pushes recommendation information to target object, to further increase the standard of the recommendation information pushed to user The clicking rate of true property and recommendation information, improves user experience.
It, can through above-mentioned analysis it is found that after the attribute information and corresponding initial classification score collection for obtaining target object To determine target object in preset each time using preset prediction model based on attribute information and initial classification score collection The score of classification is selected, to push recommendation to target object in the score of preset each candidate classification according to target object Breath.In a kind of possible way of realization, in order to improve the accuracy of the recommendation information pushed to target object, it is also based on The attribute information of target object, initial classification score collection and initial classification score concentrate the quantity for the classification score for including, and utilize Preset prediction model determines that target object in the score of preset each candidate classification, and then pushes to target object and recommends Information.The recommendation information processing method of the application is further described for above situation below with reference to Fig. 7.
Fig. 7 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application.
As shown in fig. 7, the recommendation information processing method may comprise steps of:
Step 301, the attribute information and corresponding initial classification score collection of target object are obtained.
Wherein, the specific implementation process and principle of above-mentioned steps 301 is referred to the detailed of step 101 in above-described embodiment Thin description, details are not described herein again.
Step 302, the quantity that the classification score for including is concentrated according to initial classification score, determines classification score vector.
Wherein, classification score vector can uniquely characterize the quantity that initial classes mesh score concentrates the classification score for including.
The specific mode for determining classification score vector, classification corresponding with the initial classification score each classification of concentration of determination The mode of vector is similar, and details are not described herein again.
It should be noted that the dimension of classification score vector, can according to need any setting, herein with no restriction.
Further, the attribute information, corresponding initial classification score collection and classification score vector of target object are obtained Afterwards, target can be determined using preset prediction model based on classification score vector, attribute information and initial classification score collection Score of the object in preset each candidate classification.
Specifically, can be obtained by following steps 303-305 based on classification score vector, attribute information and initial classification Diversity determines target object in the score of preset each candidate classification using preset prediction model.
Step 303, determine that the corresponding attribute vector of attribute information, and initial classification score concentrate the corresponding class of each classification Mesh vector.
Step 304, according to the corresponding classification vector of each classification and the corresponding score of each classification, target object pair is determined The total classification vector answered.
It should be noted that above-mentioned steps 302 can execute before step 303, can also hold after the step 304 Row, can also be performed simultaneously with step 303, only need to execute after step 301, before 305.
Step 305, it is based on classification score vector, attribute vector and total classification vector, using preset prediction model, is determined Score of the target object in preset each candidate classification.
Step 306, recommendation is pushed to target object in the score of preset each candidate classification according to target object Breath.
Specifically, the specific implementation process and principle of above-mentioned steps 303-306, the correlation for being referred to above-described embodiment are retouched It states, details are not described herein again.
Below with reference to application scenario diagram shown in Fig. 8, recommendation information processing method provided by the embodiments of the present application is carried out Explanation.Wherein, client can provide the information such as video or text for target object.
As shown in figure 8, (the step 1) after target object clicks refreshing in the client, the place of recommendation information processing unit Manage device first according to the data stored in memory, can obtain target object attribute information and initial classification score collection (step 2) quantity that the classification score for including, is then concentrated according to initial classification score, determines classification score vector (step 3).Then, The processor of recommendation information processing unit can be based on attribute information, initial classification score collection and classification score vector, using pre- If prediction model, determine target object in the score (step 4) of preset each candidate classification, to exist according to target object The score of preset each candidate classification, obtains recommendation information (step 5), and be pushed to where target object from memory Client (step 6).
By attribute information, initial classification score collection and classification score vector based on target object, using preset pre- Model is surveyed, determines target object in the score of preset each candidate classification, and then according to target object in preset each time The score for selecting classification pushes recommendation information to target object, further improves the accuracy of the recommendation information pushed to user And the clicking rate of recommendation information, improve user experience.
It should be noted that in the application one kind preferably way of realization, as shown in figure 9, in target object in client (step 1) after refreshing is clicked in end, the processor of recommendation information processing unit can obtain first according to the data stored in memory The attribute information, initial classification score collection, initial labels of target object is taken to obtain diversity (step 2), if initial classification score is concentrated Classification score quantity be greater than first threshold, initial labels score concentrate label score quantity be greater than second threshold, then It can concentrate and be extracted with reference to classification score collection from initial classification score, concentrate and take out from initial labels score according to default rule Take reference label to obtain diversity, then according to reference classification score concentrate include classification score quantity, determine classification score to Amount concentrates the quantity for the label score for including according to reference label score, determines label score vector (step 3).Then, recommend The processor of information processing unit can attribute information based on target object, obtain with reference to classification score collection, reference label diversity, Classification score vector and label score vector determine target object in preset each candidate class using preset prediction model Purpose score (step 4), and then obtain and recommend from memory in the score of preset each candidate classification according to target object Information (step 5), and it is pushed to the client (step 6) where target object, to further increase the recommendation pushed to user The accuracy of information and the clicking rate of recommendation information improve user experience.
By above-mentioned analysis it is found that can be based on after the attribute information and initial classification score collection for obtaining target object Attribute information, initial classification score collection determine target object in preset each candidate classification using preset prediction model Score, to push recommendation information to target object in the score of preset each candidate classification according to target object.It ties below Figure 10 is closed, in the recommendation information processing method of the application, the training generating process of preset prediction model is illustrated.
Figure 10 is the flow diagram according to the recommendation information processing method shown in another exemplary embodiment of the application.
As shown in Figure 10, the recommendation information processing method can with the following steps are included:
Step 401, according to the historical behavior data of each history registry object, determine that each history registry object is corresponding Classification obtains diversity.
Wherein, historical behavior data, may include each history registry object within a preset period of time, it is corresponding to each classification The data such as the number of clicks of information.
It is understood that during it browses information, usually feeling to it for history registry object The various information of interest has carried out multiple click.In the embodiment of the present application, by each history registry object to each classification The historical behaviors data such as number of clicks of corresponding information carry out dissection process, can determine that each history registry object is corresponding Classification obtains diversity.
Step 402, the quantity that the classification score for including is concentrated according to the corresponding classification score of each history registry object, from Sample object is obtained in history registry object.
Wherein, the classification score of sample object concentrates the quantity for the classification score for including to meet preset condition.Preset condition, It can according to need setting.
It is understood that the corresponding classification score of each history registry object concentrates the quantity for the classification score for including, It is usually different, and the corresponding classification score of the history registry object having concentrates the quantity for the classification score for including that may compare More, the corresponding classification score of some history registry objects concentrates the quantity for the classification score for including may be fewer.In the application In embodiment, in order to improve the accuracy and reliability of prediction model prediction result, preset condition be can be set are as follows: classification score The quantity for the classification score that concentration includes is greater than preset third threshold value, so that when the training preset prediction model of generation In training sample, sample object is a fairly large number of history registry pair that corresponding classification score concentrates the classification score for including As.
Wherein, third threshold value can according to need any setting, for example be set as 100,200 etc., so as to from going through History is registered in object, and the history registry that corresponding classification score concentrates the quantity for the classification score for including to be greater than third threshold value is chosen Object, as sample object.
Step 403, the attribute information of sample object is obtained.
Specifically, the information such as gender, age may have been filled in when due to user's registration, then, implement in the application In example, the content that can be directly filled according to it obtains the attribute information of sample object.
Step 404, diversity is obtained using the attribute information of sample object and corresponding classification, initial model is trained, To generate preset prediction model.
Specifically, can be obtained by following steps 403a-403c using the attribute information and corresponding classification of sample object Diversity is trained initial model, to generate preset prediction model.
Step 403a obtains diversity using the attribute information and corresponding classification of each sample object, carries out to initial model Training determines each sample object in the prediction score of preset each candidate classification.
Step 403b, according to each sample object preset each candidate classification prediction score, with each sample pair As the difference of corresponding classification score concentration classification score, prediction error value is determined.
Step 403c carries out backpropagation amendment to initial model, to generate preset prediction mould according to prediction error value Type.
Wherein, initial model can be the neural network model connected entirely, or, or other models, herein not It is restricted.The neural network that the embodiment of the present application is three layers with initial model, hidden node quantity is respectively 256,128,128, Activation primitive is to be illustrated for correcting linear unit (Rectified linear unit, abbreviation ReLU).
Specifically, can determine the attribute of some sample object after pre-setting the structure and each parameter of initial model The corresponding attribute vector of information and corresponding classification score concentrate the corresponding classification vector of each classification, and according to each classification Corresponding classification vector and the corresponding score of each classification determine the corresponding total classification vector of the sample object, then by total class Mesh vector sum attribute vector is spliced into input of the vector as initial model, to obtain the sample using initial model Prediction score of the object in preset each candidate classification.Then, due to the class of the corresponding classification score concentration of the sample object Mesh may not include all candidate classifications, therefore in the embodiment of the present application, can be from the sample object preset every In the prediction score of a candidate's classification, selects classification score corresponding with the sample object and the pre- of identical classification is concentrated to measure Point, then the corresponding classification score of the sample object is concentrated into all kinds of purpose scores, classification identical as what is selected is predicted to divide it Between make it is poor, and will make difference after each result square value sum after be averaging again, as prediction error value.And then it is missed according to prediction Difference carries out backpropagation amendment to initial model.
It is then possible to continue through aforesaid way, diversity is obtained using the corresponding classification of another sample object and attribute is believed Breath is trained to by primary modified initial model, determines the sample object in the prediction of preset each candidate classification Score, and according to the sample object in the prediction score of preset each candidate classification, classification corresponding with the sample object obtains The difference of classification score in diversity determines corresponding prediction error value, and according to the prediction error value, to by primary modified Initial model continues backpropagation amendment, until obtaining diversity and attribute letter using all corresponding classifications of sample object Breath produces preset prediction model after continuing multiple backpropagation amendment to revised initial model.
When specific implementation, can using generate confrontation network thought, by discriminator come to each sample object pre- If candidate classification prediction score, corresponding with each sample object classification score concentrates the difference of classification score to be divided Analysis, so that prediction error value is obtained, to carry out backpropagation amendment to initial model.
It should be noted that obtain diversity and attribute information using the corresponding classification of each sample object, to initial model into When row training, the quantity for the classification score that the corresponding classification score of each sample object is concentrated can be identical, be also possible to Different.Correspondingly, in the training generating process of preset prediction model, the corresponding classification of each sample object obtains diversity, with Using preset prediction model, target object is determined in the score of preset each candidate classification, target object is corresponding just Beginning classification score concentrate, the quantity of classification score be also possible to it is identical, be also possible to it is different, herein with no restriction.
In addition, the corresponding attribute vector of the attribute information of sample object and corresponding classification score concentrate each classification pair The classification vector answered can be obtained by meeting the random initializtion of Gaussian Profile, in the process being trained to initial model In, each attribute vector and classification vector can be updated.
By the above process, it can train and generate preset prediction model, thus pushing recommendation information to target object When, can attribute information based on target object and initial classification score collection using preset prediction model determine target object In the score of preset each candidate classification, so according to target object preset each candidate classification score, to target Object Push recommendation information.
It should be noted that the corresponding classification score of sample object concentrates the classification score for including in practice Quantity may be very much.It, in the embodiment of the present application, can be from each sample in order to enhance the generalization ability of preset prediction model The corresponding classification score of this object is concentrated, the corresponding score of extraction section classification, thus corresponding using each sample object Classification obtains diversity and attribute information, when being trained to initial model, can obtain merely with the corresponding classification of each sample object The corresponding score of part classification and attribute information in diversity, are trained initial model.Specific extraction process, Ke Yican According to the description in above-described embodiment to the process with reference to classification score collection is extracted, details are not described herein again.
Specifically, can first concentrate from the corresponding classification score of each sample object, it is corresponding to randomly select part classification Score obtains diversity as part classification, then described using initial classification score collection and attribute information through the foregoing embodiment, To the mode that initial model is trained, diversity and attribute information are obtained using the part classification that each sample object extracts, to first Beginning model is trained, and determines each sample object in the prediction score of preset each candidate classification, further according to each sample Object predicts to obtain diversity preset each candidate classification, and classification score corresponding with each sample object concentrates classification score Difference, determine prediction error value, thus according to each prediction error value, multiple backpropagation amendment carried out to initial model.By This, can complete the wheel training to initial model.
It is then possible to concentrate again from the corresponding classification score of each sample object, it is corresponding to randomly select part classification It is allocated as obtaining diversity for part classification, next proceeds through aforesaid way, the part classification score extracted using each sample object Collection and attribute information, are trained initial model, determine that each sample object is measured in preset each candidate the pre- of classification Point, diversity is predicted to obtain in preset each candidate classification further according to each sample object, class corresponding with each sample object Mesh score concentrates the difference of classification score, determines prediction error value, to be carried out to initial model more according to each prediction error value Secondary backpropagation amendment.Thus, it is possible to complete the another wheel training to initial model.By taking turns training, that is, produce default more Prediction model.
It should be noted that by the multiple extraction for obtaining diversity to the corresponding classification of each sample object, and then to initial Model carries out more wheel training, and when generating preset prediction model, the how many wheels of specific training can according to need setting.In addition, When obtaining diversity to the corresponding classification of same sample object and extracting, the quantity that the classification extracted every time obtains diversity can be identical, It can also be different, herein with no restriction.
It is understood that in order to improve the accuracy and reliability of prediction model prediction result, in the embodiment of the present application In, diversity and attribute information can be obtained using the corresponding classification of each sample incessantly, the mark of each sample object can also be utilized One or more of diversity, classification score vector, label score vector are signed to obtain, initial model is trained, it is pre- to generate If prediction model.
In a kind of preferably way of realization, it can use from the classification score of each sample object and concentrate the reference extracted Classification obtains diversity, concentrated from the corresponding label score of each sample object the reference label extracted obtain diversity, classification score vector, Label score vector, attribute information, are trained initial model, to generate preset prediction model.
Specifically, diversity can obtained to the corresponding classification of certain sample object respectively and label obtains after diversity extracts, Determine that the reference classification score after extracting concentrates the corresponding classification vector of each classification, and the reference label score after extraction is concentrated The corresponding label vector of each label.Then according to the corresponding classification vector of each classification and the corresponding score of each classification, really The fixed corresponding total classification vector of the sample object, according to the corresponding label vector of each label and the corresponding score of each label, Determine the corresponding total label vector of the sample object.The number for the classification score for including is concentrated further according to the classification score after extraction Amount, determines classification score vector, and the quantity for the label score for including is concentrated according to the label score after extraction, determines label score Vector determines corresponding attribute vector according to the attribute information of the sample object.And then by total classification vector, classification score to Amount, total label vector, label score vector, attribute vector are spliced, as the input of initial model, to determine the sample pair As the prediction score in preset each candidate classification, and by discriminator, to the sample object in preset candidate classification It predicts that score, classification score corresponding with the sample object concentrate the difference of classification score to be analyzed, is missed to obtain prediction Difference, to carry out backpropagation amendment to initial model.Diversity is obtained to the classification of all sample objects by utilization and label obtains Diversity carries out obtained reference classification score collection after once extracting, reference label obtains diversity and attribute information, label score to The training in rotation to initial model can be completed after carrying out multiple backpropagation amendment to initial model in amount, classification score vector Practice.Diversity is obtained by the classification to all sample objects and label obtains diversity and repeatedly extracted, it is more to be carried out to initial model Wheel training, that is, produce preset prediction model.
For the training process shown in Figure 11, diversity is being obtained to the corresponding classification of certain sample object A1 respectively and label obtains After diversity is extracted, determining that the reference classification score after extracting concentrates the corresponding classification vector of each classification is c1 and c2, It is tag1, tag2 and tag3 that reference label score after extraction, which concentrates the corresponding label vector of each label,.Then it is based on c1, c2 And the corresponding score W of c1c1, the corresponding score W of c2c2, can be according to C=(c1*Wc1+c2*Wc2)/2 determine sample object A1 Corresponding total classification vector C is based on the corresponding score W of tag1, tag2, tag3 and tag1tag1, the corresponding score W of tag2tag2、 The corresponding score W of tag3tag3, according to Tag=(tag1*Wtag1+tag2*Wtag2+tag3*Wtag3)/3 determine sample object A1 Corresponding total label vector Tag.The quantity N for the classification score for including is concentrated according to the reference classification score after extractionc=2, it can To determine classification score vector, the quantity N for the label score for including is concentrated according to the reference label score after extractiontag=3, it can To determine label score vector, according to the attribute information gender of sample object A1 and age, the corresponding category of gender can be determined Property vector and age corresponding attribute vector.In turn by total classification vector C, classification score vector, total label vector Tag, label Score vector, gender and age corresponding attribute vector are spliced, and as the input of initial model, can determine the sample This object A1 preset each candidate classification prediction score, by discriminator, to sample object A1 in preset candidate The prediction score of classification, classification score corresponding with sample object A1 are concentrated the difference of classification score to be analyzed, can be obtained To prediction error value, to carry out backpropagation amendment to initial model.By being utilized respectively the institutes such as sample object A2, A3, A4 again There is sample object is corresponding to obtain diversity, attribute information, label score vector and classification score with reference to classification score collection, reference label Vector, by the above-mentioned means, the training in rotation to initial model can be completed after carrying out multiple backpropagation amendment to initial model Practice.Diversity is obtained by the classification to all sample objects such as A1, A2, A3, A4 and label obtains diversity and repeatedly extracted, to first Beginning model carries out more wheel training, that is, produces preset prediction model.
It should be noted that in the embodiment of the present application, if according to the corresponding attribute of attribute information of each sample object Vector, total classification vector, total label vector, label score vector and classification score vector, are trained initial model, generate Preset prediction model then utilizes preset prediction model, determine target object in the score of each candidate classification, it is corresponding Need according to the corresponding attribute vector of attribute information of target object, total classification vector, total label vector, label score vector and Classification score vector.That is, in the training process for carrying out preset prediction model, the input of model is generated with training is utilized Preset prediction model determines that mode input of the target object in the score of each candidate classification is corresponding, and sample object The attribute for including in attribute information and the attribute information of target object is identical, for example all includes gender, age.
In the exemplary embodiment, a kind of recommendation information processing unit is additionally provided.
Figure 12 is the structural schematic diagram according to the recommendation information processing unit shown in one exemplary embodiment of the application.
Referring to Fig.1 shown in 2, the recommendation information processing unit of the application includes: that the first acquisition module 110, first determines mould Block 120 and pushing module 130.
Wherein, first obtain module 110, for obtain target object attribute information and corresponding initial classification score Collection；
First determining module 120, for being based on attribute information and initial classification score collection, using preset prediction model, Determine target object in the score of preset each candidate classification, the preset prediction model, to utilize the category of sample object Property information and the training of classification score get, wherein the classification score of the sample object concentrates the classification score for including Quantity meets preset condition；
Pushing module 130, for, in the score of preset each candidate classification, being pushed to target object according to target object Recommendation information.
Specifically, recommendation information processing unit provided by the embodiments of the present application, can execute provided by the embodiments of the present application Recommendation information processing method.The device can be configured in can provide the application of the information such as video or text arbitrarily for user In program, to carry out accurate information recommendation for any object, improve the click of recommendation information by recommendation information processing unit Rate.
Specifically, above-mentioned first obtains module 110, it is specifically used for:
If target object is history registry object, dissection process is carried out to the historical behavior data of target object, is determined The corresponding initial classification score collection of target object；
Alternatively,
If target object is to register object for the first time, it is determined that it is the corresponding initial classes of target object that preset classification, which obtains diversity, Mesh obtains diversity.
In a kind of possible way of realization, above-mentioned apparatus can also include:
Second obtains module, obtains diversity for obtaining the corresponding initial labels of target object；
Correspondingly, above-mentioned first determining module 120, is specifically used for:
Diversity is obtained based on attribute information, initial classification score collection and initial labels, using preset prediction model, determines mesh Object is marked in the score of preset each candidate classification.
In alternatively possible way of realization, the initial classification score of target object concentrates the number for the classification score for including Amount is greater than first threshold；
Correspondingly, above-mentioned apparatus, can also include:
Abstraction module refers to classification score collection for concentrating to extract from initial classification score according to default rule；
Correspondingly, above-mentioned first determining module 120, is also used to:
Determine target object preset using preset prediction model based on attribute information and with reference to classification score collection The score of each candidate's classification.
In alternatively possible way of realization, above-mentioned apparatus can also include:
Second determining module determines that classification obtains for concentrating the quantity for the classification score for including according to initial classification score Divide vector；
Correspondingly, above-mentioned first determining module 120, is also used to:
Target is determined using preset prediction model based on classification score vector, attribute information and initial classification score collection Score of the object in preset each candidate classification.
In alternatively possible way of realization, above-mentioned first determining module 120 is also used to:
Determine that the corresponding attribute vector of attribute information, and initial classification score concentrate the corresponding classification vector of each classification；
According to the corresponding classification vector of each classification and the corresponding score of each classification, the corresponding total class of target object is determined Mesh vector；
Target object is determined using preset prediction model based on classification score vector, attribute vector and total classification vector In the score of preset each candidate classification.
In alternatively possible way of realization, above-mentioned apparatus can also include:
Third determining module determines each history registry for the historical behavior data according to each history registry object The corresponding classification of object obtains diversity；
Third obtains module, for concentrating the classification score for including according to the corresponding classification score of each history registry object Quantity, obtain sample object from history registry object；
4th obtains module, for obtaining the attribute information of sample object；
Training module, for using sample object attribute information and corresponding classification obtain diversity, to initial model carry out Training, to generate preset prediction model.
In alternatively possible way of realization, above-mentioned training module is specifically used for:
Diversity is obtained using the attribute information and corresponding classification of each sample object, initial model is trained, is determined Prediction score of each sample object in preset each candidate classification；
Prediction score according to each sample object in preset each candidate classification, class corresponding with each sample object Mesh score concentrates the difference of classification score, determines prediction error value；
According to prediction error value, backpropagation amendment is carried out to initial model, to generate preset prediction model.
It should be noted that the aforementioned explanation to recommendation information processing method embodiment is also applied for the embodiment Recommendation information processing unit, realization principle is similar, and details are not described herein again.
Recommendation information processing unit provided by the embodiments of the present application is obtaining the attribute information of target object and corresponding After initial classification score collection, target can be determined using preset prediction model based on attribute information and initial classification score collection Object preset each candidate classification score, thus according to target object preset each candidate classification score, to Target object pushes recommendation information.Hereby it is achieved that being target according to the attribute information of target object and initial classification score collection Its interested information of object recommendation is determined since initial classification score collection can obtain diversity according to preset classification, alternatively, can To be determined according to the historical behavior data of target object, accurate information recommendation is carried out so as to be embodied as any object, is mentioned The high clicking rate of recommendation information, improves user experience.
In the exemplary embodiment, a kind of computer equipment is additionally provided.
Figure 13 is the structural schematic diagram according to the computer equipment shown in one exemplary embodiment of the application.Figure 13 is shown Computer equipment be only an example, should not function to the embodiment of the present application and use scope bring any restrictions.
Referring to Fig.1 3, which includes: memory 210 and processor 220, and the memory 210 is stored with Computer program, when the computer program is executed by processor 220, so that the processor 220 executes such as previous embodiment The recommendation information processing method.
In a kind of optional way of realization, which can also include: memory 210 and processor 220, the bus 230 of different components (including memory 210 and processor 220) is connected, memory 210 is stored with computer journey Sequence realizes recommendation information processing method described in the embodiment of the present application when processor 220 executes described program.
It should be noted that the aforementioned explanation to recommendation information processing method embodiment is also applied for the embodiment Computer equipment, realization principle is similar, and details are not described herein again.
Computer equipment provided by the embodiments of the present application, in the attribute information for obtaining target object and corresponding initial classes After mesh obtains diversity, it can determine that target object exists using preset prediction model based on attribute information and initial classification score collection The score of preset each candidate classification, thus according to target object preset each candidate classification score, to target pair As pushing recommendation information.Hereby it is achieved that being pushed away according to the attribute information of target object and initial classification score collection for target object Its interested information is recommended, is determined since initial classification score collection can obtain diversity according to preset classification, alternatively, can basis The historical behavior data of target object determine, carry out accurate information recommendation so as to be embodied as any object, improve and push away The clicking rate for recommending information, improves user experience.
In the exemplary embodiment, the application also proposed a kind of computer readable storage medium.
Above-mentioned computer readable storage medium, is stored thereon with computer program, when which is executed by processor, realizes The recommendation information processing method.
It should be noted that the aforementioned explanation to recommendation information processing method embodiment is also applied for the embodiment Computer readable storage medium, realization principle is similar, and details are not described herein again.
Computer readable storage medium provided by the embodiments of the present application in the attribute information for obtaining target object, and corresponds to Initial classification score collection after, mesh can be determined using preset prediction model based on attribute information and initial classification score collection Object is marked in the score of preset each candidate classification, thus according to target object preset each candidate classification score, Recommendation information is pushed to target object.Hereby it is achieved that being mesh according to the attribute information of target object and initial classification score collection Its interested information of object recommendation is marked, is determined since initial classification score collection can obtain diversity according to preset classification, alternatively, It can be determined according to the historical behavior data of target object, carry out accurate information recommendation so as to be embodied as any object, The clicking rate for improving recommendation information, improves user experience.
In the exemplary embodiment, the application also proposed a kind of computer program product, when in computer program product Instruction when being executed by processor, execute such as the recommendation information processing method in previous embodiment.
Computer program product provided by the embodiments of the present application, obtain target object attribute information and it is corresponding just After beginning classification obtains diversity, target pair can be determined using preset prediction model based on attribute information and initial classification score collection As the score in preset each candidate classification, thus according to target object preset each candidate classification score, to mesh Mark Object Push recommendation information.Hereby it is achieved that being target pair according to the attribute information of target object and initial classification score collection As recommending its interested information, determined since initial classification score collection can obtain diversity according to preset classification, alternatively, can be with It is determined according to the historical behavior data of target object, carries out accurate information recommendation so as to be embodied as any object, improve The clicking rate of recommendation information, improves user experience.
In the description of the present application, it is to be understood that term " first ", " second " are used for description purposes only, and cannot It is interpreted as indication or suggestion relative importance or implicitly indicates the quantity of indicated technical characteristic.Define as a result, " the One ", the feature of " second " can explicitly or implicitly include one or more of the features.In the description of the present application, The meaning of " plurality " is two or more, unless otherwise specifically defined.
It should be appreciated that each section of the application can be realized with hardware, software, firmware or their combination.Above-mentioned In embodiment, software that multiple steps or method can be executed in memory and by suitable instruction execution system with storage Or firmware is realized.It, and in another embodiment, can be under well known in the art for example, if realized with hardware Any one of column technology or their combination are realized: having a logic gates for realizing logic function to data-signal Discrete logic, with suitable combinational logic gate circuit specific integrated circuit, programmable gate array (PGA), scene Programmable gate array (FPGA) etc..
Those skilled in the art are understood that realize all or part of step that above-described embodiment method carries It suddenly is that relevant hardware can be instructed to complete by program, the program can store in a kind of computer-readable storage medium In matter, which when being executed, includes the steps that one or a combination set of embodiment of the method.
It, can also be in addition, can integrate in a processing module in each functional unit in each embodiment of the application It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould Block both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module is such as Fruit is realized and when sold or used as an independent product in the form of software function module, also can store in a computer In read/write memory medium.
Storage medium mentioned above can be read-only memory, disk or CD etc..Although having been shown and retouching above Embodiments herein is stated, it is to be understood that above-described embodiment is exemplary, and should not be understood as the limit to the application System, those skilled in the art can be changed above-described embodiment, modify, replace and become within the scope of application Type.

Claims (11)

1. a kind of recommendation information processing method characterized by comprising
Obtain the attribute information and corresponding initial classification score collection of target object；
The target object is determined using preset prediction model based on the attribute information and the initial classification score collection In the score of preset each candidate classification, the preset prediction model, to utilize the attribute information and classification of sample object Score training is got, wherein it is default that the classification score of the sample object concentrates the quantity for the classification score for including to meet Condition；
Score according to the target object in preset each candidate classification, Xiang Suoshu target object push recommendation information.
2. the method as described in claim 1, which is characterized in that the determination target object is in preset each candidate class Before purpose score, further includes:
It obtains the corresponding initial labels of the target object and obtains diversity；
Score of the determination target object in preset each candidate classification, comprising:
Diversity is obtained based on the attribute information, the initial classification score collection and the initial labels, utilizes preset prediction mould Type determines the target object in the score of preset each candidate classification.
3. the method as described in claim 1, which is characterized in that if the initial classification score concentration of the target object includes The quantity of classification score is greater than first threshold, then the determination target object must divide it in preset each candidate classification Before, further includes:
According to default rule, concentrates and extracted with reference to classification score collection from the initial classification score；
Score of the determination target object in preset each candidate classification, comprising:
Determine the target object pre- using preset prediction model based on the attribute information and with reference to classification score collection If each of candidate classification score.
4. the method as described in claim 1, which is characterized in that the determination target object is in preset each candidate class Before purpose score, further includes:
The quantity that the classification score for including is concentrated according to the initial classification score, determines classification score vector；
Score of the determination target object in preset each candidate classification, comprising:
Based on the classification score vector, the attribute information and the initial classification score collection, using preset prediction model, Determine the target object in the score of preset each candidate classification.
5. the method as claimed in claim 4, which is characterized in that the determination target object is in preset each time Select the score of classification, comprising:
Determine the corresponding attribute vector of the attribute information and the initial classification score concentrate the corresponding classification of each classification to Amount；
According to each corresponding classification vector of classification and the corresponding score of each classification, determine that the target object is corresponding Total classification vector；
It is determined based on the classification score vector, the attribute vector and total classification vector using preset prediction model Score of the target object in preset each candidate classification.
6. the method as described in claim 1, which is characterized in that the corresponding initial classification score of the acquisition target object Collection, comprising:
If the target object is history registry object, dissection process is carried out to the historical behavior data of the target object, Determine the corresponding initial classification score collection of the target object；
Alternatively,
If the target object is to register object for the first time, it is determined that it is that the target object is corresponding just that preset classification, which obtains diversity, Beginning classification obtains diversity.
7. the method as described in claim 1-6 is any, which is characterized in that the determination target object is preset each Before the score of candidate classification, further includes:
According to the historical behavior data of each history registry object, determine that the corresponding classification of each history registry object obtains diversity；
The quantity that the classification score for including is concentrated according to the corresponding classification score of each history registry object, from the history registry Sample object is obtained in object；
Obtain the attribute information of the sample object；
Diversity is obtained using the attribute information and corresponding classification of the sample object, initial model is trained, to generate State preset prediction model.
8. the method for claim 7, which is characterized in that the attribute information using the sample object and corresponding Classification obtains diversity, is trained to initial model, comprising:
Diversity is obtained using the attribute information and corresponding classification of each sample object, initial model is trained, is determined each Prediction score of the sample object in preset each candidate classification；
According to each sample object in the prediction score of preset each candidate classification, classification corresponding with each sample object obtained The difference of classification score, determines prediction error value in diversity；
According to the prediction error value, backpropagation amendment is carried out to the initial model, to generate the preset prediction mould Type.
9. a kind of recommendation information processing unit characterized by comprising
First obtains module, for obtaining the attribute information and corresponding initial classification score collection of target object；
First determining module, for being based on the attribute information and the initial classification score collection, using preset prediction model, Determine the target object in the score of preset each candidate classification, the preset prediction model, to utilize sample object Attribute information and the training of classification score get, wherein the classification score of the sample object concentrates the classification for including to obtain The quantity divided meets preset condition；
Pushing module, for, in the score of preset each candidate classification, Xiang Suoshu target object to push away according to the target object Send recommendation information.
10. a kind of computer equipment, which is characterized in that including memory, processor and store on a memory and can handle The computer program run on device, when the processor executes described program, to realize a method as claimed in any one of claims 1-8 push away Recommend information processing method.
11. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor When execution, recommendation information processing method a method as claimed in any one of claims 1-8 is realized.
CN201811251040.8A 2018-10-25 2018-10-25 Recommendation information processing method, device, equipment and storage medium Active CN109460512B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811251040.8A CN109460512B (en) 2018-10-25 2018-10-25 Recommendation information processing method, device, equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811251040.8A CN109460512B (en) 2018-10-25 2018-10-25 Recommendation information processing method, device, equipment and storage medium

Publications (2)

Publication Number Publication Date
CN109460512A true CN109460512A (en) 2019-03-12
CN109460512B CN109460512B (en) 2022-04-22

Family

ID=65608417

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811251040.8A Active CN109460512B (en) 2018-10-25 2018-10-25 Recommendation information processing method, device, equipment and storage medium

Country Status (1)

Country Link
CN (1) CN109460512B (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110727868A (en) * 2019-10-12 2020-01-24 腾讯音乐娱乐科技（深圳）有限公司 Object recommendation method, device and computer-readable storage medium
CN110827120A (en) * 2019-10-18 2020-02-21 郑州大学 GAN network-based fuzzy recommendation method and device, electronic equipment and storage medium
CN111651669A (en) * 2020-05-20 2020-09-11 拉扎斯网络科技（上海）有限公司 Information recommendation method and device, electronic equipment and computer-readable storage medium
CN112148973A (en) * 2020-09-04 2020-12-29 京东数字科技控股股份有限公司 Data processing method and device for information push
CN112395499A (en) * 2020-11-03 2021-02-23 腾讯科技（深圳）有限公司 Information recommendation method and device, electronic equipment and storage medium
CN112650931A (en) * 2021-01-04 2021-04-13 杭州情咖网络技术有限公司 Content recommendation method
CN112784083A (en) * 2019-11-04 2021-05-11 阿里巴巴集团控股有限公司 Method and device for acquiring category prediction model and feature extraction model
CN113010788A (en) * 2021-03-19 2021-06-22 成都欧珀通信科技有限公司 Information pushing method and device, electronic equipment and computer readable storage medium
CN113468426A (en) * 2021-07-01 2021-10-01 北京明略软件系统有限公司 Information recommendation method and device, electronic equipment and readable storage medium
CN113487379A (en) * 2021-06-24 2021-10-08 上海淇馥信息技术有限公司 Product recommendation method and device based on conversation and electronic equipment
CN113596528A (en) * 2020-04-30 2021-11-02 北京达佳互联信息技术有限公司 Training method and device of video push model, server and storage medium

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102207972A (en) * 2011-06-15 2011-10-05 中山大学 Television program recommending method and device for digital television
US20140143233A1 (en) * 2012-11-20 2014-05-22 Alibaba Group Holding Limited Method and system for recommending target object information
CN104933074A (en) * 2014-03-20 2015-09-23 华为技术有限公司 News ordering method and device and terminal equipment
CN105005582A (en) * 2015-06-17 2015-10-28 深圳市腾讯计算机系统有限公司 Recommendation method and device for multimedia information
CN106815217A (en) * 2015-11-30 2017-06-09 北京云莱坞文化传媒有限公司 Story recommends method and story recommendation apparatus
CN106959966A (en) * 2016-01-12 2017-07-18 腾讯科技（深圳）有限公司 A kind of information recommendation method and system
CN107016058A (en) * 2017-03-10 2017-08-04 浙江工业大学 A kind of recommendation Forecasting Methodology based on attribute information preference self study
CN107256494A (en) * 2017-05-18 2017-10-17 深圳大学 A kind of item recommendation method and device
CN107563500A (en) * 2017-09-14 2018-01-09 北京奇艺世纪科技有限公司 A kind of video recommendation method and system based on user's head portrait
CN107609060A (en) * 2017-08-28 2018-01-19 百度在线网络技术（北京）有限公司 Resource recommendation method and device
CN108076154A (en) * 2017-12-21 2018-05-25 广东欧珀移动通信有限公司 Application message recommends method, apparatus and storage medium and server
CN108093303A (en) * 2016-11-23 2018-05-29 深圳市优朋普乐传媒发展有限公司 A kind of video pushing method and device
CN108304429A (en) * 2017-05-16 2018-07-20 腾讯科技（深圳）有限公司 Information recommendation method, device and computer equipment
CN108648031A (en) * 2018-03-28 2018-10-12 苏州迪尔未来网络科技有限公司 Products Show method and device

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102207972A (en) * 2011-06-15 2011-10-05 中山大学 Television program recommending method and device for digital television
US20140143233A1 (en) * 2012-11-20 2014-05-22 Alibaba Group Holding Limited Method and system for recommending target object information
CN104933074A (en) * 2014-03-20 2015-09-23 华为技术有限公司 News ordering method and device and terminal equipment
CN105005582A (en) * 2015-06-17 2015-10-28 深圳市腾讯计算机系统有限公司 Recommendation method and device for multimedia information
CN106815217A (en) * 2015-11-30 2017-06-09 北京云莱坞文化传媒有限公司 Story recommends method and story recommendation apparatus
CN106959966A (en) * 2016-01-12 2017-07-18 腾讯科技（深圳）有限公司 A kind of information recommendation method and system
CN108093303A (en) * 2016-11-23 2018-05-29 深圳市优朋普乐传媒发展有限公司 A kind of video pushing method and device
CN107016058A (en) * 2017-03-10 2017-08-04 浙江工业大学 A kind of recommendation Forecasting Methodology based on attribute information preference self study
CN108304429A (en) * 2017-05-16 2018-07-20 腾讯科技（深圳）有限公司 Information recommendation method, device and computer equipment
CN107256494A (en) * 2017-05-18 2017-10-17 深圳大学 A kind of item recommendation method and device
CN107609060A (en) * 2017-08-28 2018-01-19 百度在线网络技术（北京）有限公司 Resource recommendation method and device
CN107563500A (en) * 2017-09-14 2018-01-09 北京奇艺世纪科技有限公司 A kind of video recommendation method and system based on user's head portrait
CN108076154A (en) * 2017-12-21 2018-05-25 广东欧珀移动通信有限公司 Application message recommends method, apparatus and storage medium and server
CN108648031A (en) * 2018-03-28 2018-10-12 苏州迪尔未来网络科技有限公司 Products Show method and device

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110727868A (en) * 2019-10-12 2020-01-24 腾讯音乐娱乐科技（深圳）有限公司 Object recommendation method, device and computer-readable storage medium
CN110727868B (en) * 2019-10-12 2022-07-15 腾讯音乐娱乐科技（深圳）有限公司 Object recommendation method, device and computer-readable storage medium
CN110827120A (en) * 2019-10-18 2020-02-21 郑州大学 GAN network-based fuzzy recommendation method and device, electronic equipment and storage medium
CN112784083A (en) * 2019-11-04 2021-05-11 阿里巴巴集团控股有限公司 Method and device for acquiring category prediction model and feature extraction model
CN113596528A (en) * 2020-04-30 2021-11-02 北京达佳互联信息技术有限公司 Training method and device of video push model, server and storage medium
CN111651669A (en) * 2020-05-20 2020-09-11 拉扎斯网络科技（上海）有限公司 Information recommendation method and device, electronic equipment and computer-readable storage medium
CN112148973A (en) * 2020-09-04 2020-12-29 京东数字科技控股股份有限公司 Data processing method and device for information push
CN112148973B (en) * 2020-09-04 2023-11-07 京东科技控股股份有限公司 Data processing method and device for information push
CN112395499A (en) * 2020-11-03 2021-02-23 腾讯科技（深圳）有限公司 Information recommendation method and device, electronic equipment and storage medium
CN112650931B (en) * 2021-01-04 2023-05-30 杭州情咖网络技术有限公司 Content recommendation method
CN112650931A (en) * 2021-01-04 2021-04-13 杭州情咖网络技术有限公司 Content recommendation method
CN113010788A (en) * 2021-03-19 2021-06-22 成都欧珀通信科技有限公司 Information pushing method and device, electronic equipment and computer readable storage medium
CN113010788B (en) * 2021-03-19 2023-05-23 成都欧珀通信科技有限公司 Information pushing method and device, electronic equipment and computer readable storage medium
CN113487379B (en) * 2021-06-24 2023-01-13 上海淇馥信息技术有限公司 Product recommendation method and device based on conversation mode and electronic equipment
CN113487379A (en) * 2021-06-24 2021-10-08 上海淇馥信息技术有限公司 Product recommendation method and device based on conversation and electronic equipment
CN113468426A (en) * 2021-07-01 2021-10-01 北京明略软件系统有限公司 Information recommendation method and device, electronic equipment and readable storage medium
CN113468426B (en) * 2021-07-01 2024-01-30 北京明略软件系统有限公司 Information recommendation method and device, electronic equipment and readable storage medium

Also Published As

Publication number Publication date
CN109460512B (en) 2022-04-22

Similar Documents

Publication Publication Date Title
CN109460512A (en) Recommendation information processing method, device, equipment and storage medium
CN110413877A (en) A kind of resource recommendation method, device and electronic equipment
CN104680453B (en) Course based on student's attribute recommends method and system
CN107783703B (en) E-book and e-book topic exchange method calculate equipment, storage medium
Hausmann et al. The atlas of economic complexity: Mapping paths to prosperity
CN110737783B (en) Method and device for recommending multimedia content and computing equipment
US11023545B2 (en) Method and device for displaying recommended contents
CN109582875A (en) A kind of personalized recommendation method and system of online medical education resource
CN110162690A (en) Determine user to the method and apparatus of the interest-degree of article, equipment and storage medium
CN110097412A (en) Item recommendation method, device, equipment and storage medium
CN112800097A (en) Special topic recommendation method and device based on deep interest network
US9401097B2 (en) Method and apparatus for providing emotion expression service using emotion expression identifier
CN111061946A (en) Scenario content recommendation method and device, electronic equipment and storage medium
CN107888950A (en) A kind of method and system for recommending video
JP6884116B2 (en) Information processing equipment, information processing methods, and programs
CN108829808A (en) A kind of page personalized ordering method, apparatus and electronic equipment
EP3739470A1 (en) Method and apparatus for performing categorised matching of videos, and selection engine
CN110348930A (en) Business object data processing method, the recommended method of business object information and device
CN107656787A (en) Newsy method, computing device, computer-readable storage medium are given birth to based on e-book
CN108074016A (en) Customer relationship intensity prediction method, device and equipment based on position social networks
CN108596695A (en) Entity method for pushing and system
CN105637501A (en) Utilizing social information for recommending an application
CN108648058A (en) Model sequencing method and device, electronic equipment, storage medium
CN111782947A (en) Search content display method and device, electronic equipment and storage medium
CN110162609A (en) For recommending the method and device asked questions to user

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
GR01 Patent grant
GR01 Patent grant