CN109033460A - Sort method, device and equipment/terminal/server in a kind of information flow - Google Patents
Sort method, device and equipment/terminal/server in a kind of information flow Download PDFInfo
- Publication number
- CN109033460A CN109033460A CN201811004222.5A CN201811004222A CN109033460A CN 109033460 A CN109033460 A CN 109033460A CN 201811004222 A CN201811004222 A CN 201811004222A CN 109033460 A CN109033460 A CN 109033460A
- Authority
- CN
- China
- Prior art keywords
- user
- recommended
- information flow
- model
- dimension
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Analysis (AREA)
- Databases & Information Systems (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computing Systems (AREA)
- Computational Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Algebra (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- User Interface Of Digital Computer (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
Abstract
The embodiment of the present application provides sort method, device and the equipment/terminal/server in a kind of information flow, which comprises using user behavior data training Factorization machine model, obtains the hidden vector of user and the multiple dimensions of recommended object；According to the hidden vector of the user and the multiple dimensions of recommended object, the matching characteristic of user and each dimension of recommended object are extracted；Sample will be constituted after the processing of the matching characteristic of the user and each dimension of recommended object, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application uses Factorization machine and gradient boosted tree models coupling carries out the sequence that object is recommended in information flow, optimizes information flow order models, enables the object of recommendation that can more meet the point of interest of user, improves user experience.
Description
Technical field
This application involves sort method, device and the equipment/ends in Internet technical field more particularly to a kind of information flow
End/server.
Background technique
With the development of internet technology, information flow becomes the major way of user's daily life amusement, is calculated by sequence
Method, which provides the interested information of user (article/video etc.), becomes the important means that each information streaming application attracts user.Such as
What recommends the object for meeting its interest to user, is usually realized by the calculating of algorithm model.
But usually algorithm model can not precise expression information flow object and user in the sequencer procedure that information flow is recommended
Relationship between interest causes no normal direction user to recommend its practical interested information, to influence the use feeling of user.
Therefore, how more preferably to realize the sequence in information flow as technical problem urgently to be resolved in the prior art.
Summary of the invention
The embodiment of the present application provides sort method, device and the equipment/terminal/server in a kind of information flow, all
Or part solves problems of the prior art.
According to the one aspect of the embodiment of the present application, the sort method in a kind of information flow is provided, which comprises
Using user behavior data training Factorization machine model, the hidden vector of user and the multiple dimensions of recommended object are obtained；According to
The hidden vector of the user and the multiple dimensions of recommended object extract the matching characteristic of user and each dimension of recommended object；
Sample being constituted after the processing of the matching characteristic of the user and each dimension of recommended object, training gradient promotes tree-model, and
Tree-model is promoted by the gradient, Ordering and marking is provided.
According to the other side of the embodiment of the present application, the collator in a kind of information flow, described device are additionally provided
It include: that vector obtains module, setting is used to be obtained user using user behavior data training Factorization machine model and be recommended
The hidden vector of the multiple dimensions of object；Characteristic extracting module is arranged for according to the user and the multiple dimensions of recommended object
Hidden vector extracts the matching characteristic of user and each dimension of recommended object；Model training module, setting is for by the user
Sample is constituted with after the matching characteristic processing of each dimension of recommended object, trains gradient to promote tree-model, and by the gradient
It promotes tree-model and Ordering and marking is provided
According to the another aspect of the embodiment of the present application, additionally provide a kind of equipment/terminal/server, comprising: one or
Multiple processors；Storage device, for storing one or more programs, when one or more of programs are by one or more
A processor executes, so that one or more of processors realize the corresponding behaviour of sort method in information flow as described above
Make.
According to the another aspect of the embodiment of the present application, a kind of computer readable storage medium is additionally provided, is stored thereon
There is computer program, which realizes the sort method corresponding operation in information flow as described above when being executed by processor.
According to technical solution provided by the embodiments of the present application, the embodiment of the present application uses user behavior data training Factor minute
Solution machine model obtains the hidden vector of user and the multiple dimensions of recommended object.And it is multiple according to the user and recommended object
The hidden vector of dimension extracts the matching characteristic of user and each dimension of recommended object, and the user and recommended object is each
Sample is constituted after the matching characteristic processing of a dimension, training gradient promotes tree-model, and promotes tree-model by the gradient and provide
Ordering and marking.The embodiment of the present application uses Factorization machine model and gradient boosted tree models coupling carries out in information flow recommendation
The sequence of recommended object, optimizes information flow order models, enables the object of recommendation that can more meet the point of interest of user, change
Kind user experience.
Detailed description of the invention
Fig. 1 is the step flow chart of the sort method in a kind of information flow according to the embodiment of the present application one；
Fig. 2 is the structural block diagram of the collator in a kind of information flow according to the embodiment of the present application three；
Fig. 3 is the structural block diagram according to a kind of equipment/terminal/server of the embodiment of the present application five.
Specific embodiment
(identical label indicates identical element in several attached drawings) and embodiment with reference to the accompanying drawing, implement the application
The specific embodiment of example is described in further detail.Following embodiment is not limited to the application for illustrating the application
Range.
It will be understood by those skilled in the art that the terms such as " first ", " second " in the embodiment of the present application are only used for distinguishing
Different step, equipment or module etc., neither represent any particular technology meaning, also do not indicate that the inevitable logic between them is suitable
Sequence.
Embodiment one
Referring to Fig.1, the step flow chart of the sort method in a kind of information flow according to the embodiment of the present application one is shown.
It is worth noting that herein described step S101 to S103 does not represent the sequencing of its execution.
Sort method in the information flow of the present embodiment the following steps are included:
Step S101: it using user behavior data training Factorization machine model, obtains user and recommended object is multiple
The hidden vector of dimension.
Factorization machine model (FM, Factorization Machines) is being located to traditional logic regression model (LR)
Manage the optimization and improvement in high-order interaction feature problem.Logic Regression Models (LR) are by way of characteristic crossover, after combination
Feature be added in model as new feature, model complexity is O (N^2) (quantity that N is interaction feature), and Memorability is stronger
And generalization is on the weak side.FM indicates feature association by the similarity (inner product) of hidden vector by being hidden vector by characteristic present
This mode carrys out the generalization ability of dexterously lift scheme.The complexity of FM model is O (N*k) (k is the super ginseng of hidden vector dimension).
FM is substantially a linear model, influences the output of model between different item in a manner of linear combination.It introduces
Nonlinear model optimizes FM model, and gradient promotes tree algorithm (GBM) as tree-model can optimize FM nonlinear model.
The embodiment of the present application obtains user behavior data, and FM model is trained using the user behavior data, thus
The hidden vector of user Yu the multiple dimensions of recommended object are obtained according to the user behavior data.
In terms of model structure, FM model may be considered the neural network knot that even closer can portray this matching degree
Structure.The embodiment of the present application is using FM model decomposition user behavior data (that is, user is to the click of recommended object and shows number
According to), obtain the hidden vector that each dimension of object is recommended in user and information flow.And then it can be calculated and be used by these hidden vectors
The matching degree at family and information flow object.
Step S102: according to the hidden vector of the user and the multiple dimensions of recommended object, user and recommended couple are extracted
As the matching characteristic of each dimension.
In the application one in the specific implementation, the matching of user described in the embodiment of the present application and each dimension of recommended object is special
Sign includes:
It is recommended in subject side feature and information flow in information flow and is recommended object and End-user relevance feature.
It is recommended subject side feature in the information flow, i.e., is recommended subject side signal in information flow, this category feature is from each
A dimension features and is recommended Properties of Objects in information flow: temperature, consumption duration, quality etc..This category feature helps to screen
Fine work content promotes the quality criteria of recommendation.
Object and End-user relevance feature are recommended in the information flow for portraying in user and information flow recommended pair
The degree of association of elephant, the degree of association can be portrayed by the movement of the recommended object of click of user, can also pass through use
The duration that family plays recommended object is portrayed, to promote the personalization of recommendation.
Object and End-user relevance feature are recommended in the information flow using the hidden vector sum of object recommended in information flow
The hidden vector of user carries out matching degree and calculates acquisition.
Specifically, by the structured message for being recommended each dimension of object in user and information flow, to be embedded into a low-dimensional hidden
Vector calculates the matching degree that object is recommended in user and information flow.
The embodiment of the present application is that all users and recommended characteristics of objects are divided into the same space by FM model
Interior hidden vector, thus all vectors are all comparable.Not only user and that each dimension of object is recommended in information flow
With being recommended between object between degree or even user, in information flow, matching degree can be obtained by simple vector operation.
It is another in the application in the specific implementation, in the information flow be recommended the hidden hidden vector of vector sum user of object use version
This alignment mechanism enables described its that can be aligned.
Since FM needs training routine, the hidden vector of user is filled to be recommended in library and information flow between the hidden vector load of object and be deposited
It in the time difference, and is incomparable between the hidden vector of different Model of Version.
Specifically, since the time window of training routine is usually 4~6 hours, the data for retaining two versions are foots
To guarantee that the hidden vector of the overwhelming majority can be aligned.All hidden vectors can all be retained nearest two versions by the version alignment mechanism
Data.In the online computing module of FM, the logic of version alignment is realized, calculate matching with the hidden vector of latest edition after alignment
Degree.In addition, the quantity of version can be increased in the model training of more high frequency to ensure that model is aligned.
Step S103: sample will be constituted after the processing of the matching characteristic of the user and each dimension of recommended object, training
Gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.
Gradient promotes tree-model (GBM), and it is Boosting algorithm that full name, which is Gradient Boosting Machine,
It is a kind of.GBM main thought be established based on the gradient descent direction of the loss function for the base learner established before it is next
New base learner, purpose are just desirable to by integrating these base learners the loss function of model totality constantly be declined,
Model is continuously improved.
FM models coupling GBM model is carried out clicking rate by GBM and estimated by the application.FM models coupling GBM model is tree
Model, is capable of handling the complex relationship of nonlinearity between signal and target, also has better interpretation.
It follows that the embodiment of the present application obtains user and quilt using user behavior data training Factorization machine model
The hidden vector of the multiple dimensions of recommended.And according to the hidden vector of the user and the multiple dimensions of recommended object, user is extracted
With the matching characteristic of each dimension of recommended object, after the matching characteristic processing of the user and each dimension of recommended object
Sample is constituted, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application is adopted
The sequence for being recommended object in information flow recommendation is carried out with Factorization machine model and gradient boosted tree models coupling, is optimized
Information flow order models enable the object of recommendation that can more meet the point of interest of user, improve user experience.
Sort method in the information flow of the present embodiment can be by any suitable sequencing ability in information flow
Equipment executes, including but not limited to: various device ends or server-side, including but not limited to PC machine, tablet computer, movement are eventually
End etc..
Embodiment two
The present embodiment includes above-mentioned steps S101 to step S103.The step S102 further include:
The feature of extraction is normalized.
Since the feature of the embodiment of the present application extraction is such as without normalized, then the feature between different user is beaten
Branch makes a difference significantly, and the characteristic use GBM model training process for enabling FM model obtain is difficult to restrain.
Therefore the embodiment of the present application will return to the distribution server after the special decent normalization of extraction, be serviced by distribution
Device is back to log server rule.Click logs are also simultaneously via log server rule.By the version alignment mechanism
It is aligned, after cleaned, filtering and anti-cheating processing, extracts reflux features and be used for model training.
It follows that the embodiment of the present application obtains user and quilt using user behavior data training Factorization machine model
The hidden vector of the multiple dimensions of recommended.And according to the hidden vector of the user and the multiple dimensions of recommended object, user is extracted
With the matching characteristic of each dimension of recommended object, after the matching characteristic processing of the user and each dimension of recommended object
Sample is constituted, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application is adopted
The sequence for being recommended object in information flow recommendation is carried out with Factorization machine model and gradient boosted tree models coupling, is optimized
Information flow order models enable the object of recommendation that can more meet the point of interest of user, improve user experience.
Sort method in the information flow of the present embodiment can be by any suitable sequencing ability in information flow
Equipment executes, including but not limited to: various device ends or server-side, including but not limited to PC machine, tablet computer, movement are eventually
End etc..
Embodiment three
Referring to Fig. 2, the structural block diagram of the collator in a kind of information flow according to the embodiment of the present application three is shown.
Collator in the information flow of the present embodiment includes:
Vector obtain module 201, setting for using user behavior data training Factorization machine model, obtain user with
The hidden vector of the recommended multiple dimensions of object.
Characteristic extracting module 202, setting are extracted for the hidden vector according to the user and the multiple dimensions of recommended object
The matching characteristic of user and each dimension of recommended object.
Model training module 203 is arranged for handling the matching characteristic of the user and each dimension of recommended object
After constitute sample, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.
Factorization machine model (FM, Factorization Machines) is being located to traditional logic regression model (LR)
Manage the optimization and improvement in high-order interaction feature problem.Logic Regression Models (LR) are by way of characteristic crossover, after combination
Feature be added in model as new feature, model complexity is O (N^2) (quantity that N is interaction feature), and Memorability is stronger
And generalization is on the weak side.FM indicates feature association by the similarity (inner product) of hidden vector by being hidden vector by characteristic present
This mode carrys out the generalization ability of dexterously lift scheme.The complexity of FM model is O (N*k) (k is the super ginseng of hidden vector dimension).
FM is substantially a linear model, influences the output of model between different item in a manner of linear combination.It introduces
Nonlinear model optimizes FM model, and gradient promotes tree algorithm (GBM) as tree-model can optimize FM nonlinear model.
The embodiment of the present application obtains user behavior data, and FM model is trained using the user behavior data, thus
The hidden vector of user Yu the multiple dimensions of recommended object are obtained according to the user behavior data.
In terms of model structure, FM model may be considered the neural network knot that even closer can portray this matching degree
Structure.The embodiment of the present application is using FM model decomposition user behavior data (that is, user is to the click of recommended object and shows number
According to), obtain the hidden vector that each dimension of object is recommended in user and information flow.And then it can be calculated and be used by these hidden vectors
The matching degree at family and information flow object.
In the application one in the specific implementation, the matching of user described in the embodiment of the present application and each dimension of recommended object is special
Sign includes:
It is recommended in subject side feature and information flow in information flow and is recommended object and End-user relevance feature.
It is recommended subject side feature in the information flow, i.e., is recommended subject side signal in information flow, this category feature is from each
A dimension features and is recommended Properties of Objects in information flow: temperature, consumption duration, quality etc..This category feature helps to screen
Fine work content promotes the quality criteria of recommendation.
Object and End-user relevance feature are recommended in the information flow for portraying in user and information flow recommended pair
The degree of association of elephant, the degree of association can be portrayed by the movement of the recommended object of click of user, can also pass through use
The duration that family plays recommended object is portrayed, to promote the personalization of recommendation.
Object and End-user relevance feature are recommended in the information flow using the hidden vector sum of object recommended in information flow
The hidden vector of user carries out matching degree and calculates acquisition.
Specifically, by the structured message for being recommended each dimension of object in user and information flow, to be embedded into a low-dimensional hidden
Vector calculates the matching degree that object is recommended in user and information flow.
The embodiment of the present application is that all users and recommended characteristics of objects are divided into the same space by FM model
Interior hidden vector, thus all vectors are all comparable.Not only user and that each dimension of object is recommended in information flow
With being recommended between object between degree or even user, in information flow, matching degree can be obtained by simple vector operation.
It is another in the application in the specific implementation, in the information flow be recommended the hidden hidden vector of vector sum user of object use version
This alignment mechanism enables described its that can be aligned.
Since FM needs training routine, the hidden vector of user is filled to be recommended in library and information flow between the hidden vector load of object and be deposited
It in the time difference, and is incomparable between the hidden vector of different Model of Version.
Specifically, since the time window of training routine is usually 4~6 hours, the data for retaining two versions are foots
To guarantee that the hidden vector of the overwhelming majority can be aligned.All hidden vectors can all be retained nearest two versions by the version alignment mechanism
Data.In the online computing module of FM, the logic of version alignment is realized, calculate matching with the hidden vector of latest edition after alignment
Degree.In addition, the quantity of version can be increased in the model training of more high frequency to ensure that model is aligned.
Gradient promotes tree-model (GBM), and it is Boosting algorithm that full name, which is Gradient Boosting Machine,
It is a kind of.GBM main thought be established based on the gradient descent direction of the loss function for the base learner established before it is next
New base learner, purpose are just desirable to by integrating these base learners the loss function of model totality constantly be declined,
Model is continuously improved.
FM models coupling GBM model is carried out clicking rate by GBM and estimated by the application.FM models coupling GBM model is tree
Model, is capable of handling the complex relationship of nonlinearity between signal and target, also has better interpretation.
It follows that the embodiment of the present application obtains user and quilt using user behavior data training Factorization machine model
The hidden vector of the multiple dimensions of recommended.And according to the hidden vector of the user and the multiple dimensions of recommended object, user is extracted
With the matching characteristic of each dimension of recommended object, after the matching characteristic processing of the user and each dimension of recommended object
Sample is constituted, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application is adopted
The sequence for being recommended object in information flow recommendation is carried out with Factorization machine model and gradient boosted tree models coupling, is optimized
Information flow order models enable the object of recommendation that can more meet the point of interest of user, improve user experience.
Sort method in the information flow of the present embodiment can be by any suitable sequencing ability in information flow
Equipment executes, including but not limited to: various device ends or server-side, including but not limited to PC machine, tablet computer, movement are eventually
End etc..
Example IV
The present embodiment includes that above-mentioned vector obtains module, characteristic extracting module, model training module.The feature extraction mould
Block further include:
The feature of extraction is normalized.
Since the feature of the embodiment of the present application extraction is such as without normalized, then the feature between different user is beaten
Branch makes a difference significantly, and the characteristic use GBM model training process for enabling FM model obtain is difficult to restrain.
Therefore the embodiment of the present application will return to the distribution server after the special decent normalization of extraction, be serviced by distribution
Device is back to log server rule.Click logs are also simultaneously via log server rule.By the version alignment mechanism
It is aligned, after cleaned, filtering and anti-cheating processing, extracts reflux features and be used for model training.
It follows that the embodiment of the present application obtains user and quilt using user behavior data training Factorization machine model
The hidden vector of the multiple dimensions of recommended.And according to the hidden vector of the user and the multiple dimensions of recommended object, user is extracted
With the matching characteristic of each dimension of recommended object, after the matching characteristic processing of the user and each dimension of recommended object
Sample is constituted, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application is adopted
The sequence for being recommended object in information flow recommendation is carried out with Factorization machine model and gradient boosted tree models coupling, is optimized
Information flow order models enable the object of recommendation that can more meet the point of interest of user, improve user experience.
Sort method in the information flow of the present embodiment can be by any suitable sequencing ability in information flow
Equipment executes, including but not limited to: various device ends or server-side, including but not limited to PC machine, tablet computer, movement are eventually
End etc..
Embodiment five
Referring to Fig. 3, a kind of structural block diagram of equipment/terminal/server according to the embodiment of the present application five, this Shen are shown
Please specific embodiment the specific implementation of equipment/terminal/server is not limited.
As shown in figure 3, equipment/the terminal/server may include: one or more processor (processor)
302, storage device (memory) 304.
Wherein:
Processor 302 can specifically execute in the sort method embodiment in above- mentioned information stream for executing program 306
Correlation step.
Specifically, program 306 may include program code, which includes computer operation instruction.
Processor 302 may be central processor CPU or specific integrated circuit ASIC (Application
Specific Integrated Circuit), or be arranged to implement the integrated electricity of one or more of the embodiment of the present application
Road.The one or more processors that equipment/terminal/server includes can be same type of processor, such as one or more
CPU；It is also possible to different types of processor, such as one or more CPU and one or more ASIC.
Storage device 304, for storing one or more programs 306.Storage device 304 may be stored comprising high-speed RAM
Device, it is also possible to further include nonvolatile memory (non-volatile memory), for example, at least a magnetic disk storage.
Program 306 specifically can be used for so that processor 302 executes following operation: use the user behavior data training factor
Disassembler model obtains the hidden vector of user and the multiple dimensions of recommended object；It is multiple according to the user and recommended object
The hidden vector of dimension extracts the matching characteristic of user and each dimension of recommended object；The user and recommended object is each
Sample is constituted after the matching characteristic processing of a dimension, training gradient promotes tree-model, and promotes tree-model by the gradient and provide
Ordering and marking.
In a kind of optional embodiment, program 306 is also used to the user and each dimension of recommended object
Matching characteristic is normalized.
In a kind of optional embodiment, the matching characteristic of the user and each dimension of recommended object includes: letter
It ceases and is recommended object and End-user relevance feature in flow object side feature and information flow.
It is recommended object in a kind of optional embodiment, in the information flow and End-user relevance feature uses information
It is recommended the hidden vector of the hidden vector sum user of object in stream and carries out matching degree calculating acquisition.
The hidden hidden vector of vector sum user of object, which is recommended, in a kind of optional embodiment, in the information flow uses version
This alignment mechanism enables described its that can be aligned.
It follows that the embodiment of the present application obtains user and quilt using user behavior data training Factorization machine model
The hidden vector of the multiple dimensions of recommended.And according to the hidden vector of the user and the multiple dimensions of recommended object, user is extracted
With the matching characteristic of each dimension of recommended object, after the matching characteristic processing of the user and each dimension of recommended object
Sample is constituted, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.The embodiment of the present application is adopted
The sequence for being recommended object in information flow recommendation is carried out with Factorization machine model and gradient boosted tree models coupling, is optimized
Information flow order models enable the object of recommendation that can more meet the point of interest of user, improve user experience.
It may be noted that all parts/step described in the embodiment of the present application can be split as more according to the needs of implementation
The part operation of two or more components/steps or components/steps can also be combined into new component/step by multi-part/step
Suddenly, to realize the purpose of the embodiment of the present application.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communications portion, and/or be pacified from detachable media
Dress.When the computer program is executed by central processing unit (CPU), the above-mentioned function of limiting in the present processes is executed.
It should be noted that computer-readable medium described herein can be computer-readable signal media or computer-readable
Storage medium either the two any combination.Computer readable storage medium for example may be-but not limited to-
Electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.It is computer-readable
The more specific example of storage medium can include but is not limited to: have electrical connection, the portable computing of one or more conducting wires
Machine disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM
Or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device or above-mentioned
Any appropriate combination.In this application, computer readable storage medium can be it is any include or storage program it is tangible
Medium, the program can be commanded execution system, device or device use or in connection.And in this application,
Computer-readable signal media may include in a base band or as the data-signal that carrier wave a part is propagated, wherein carrying
Computer-readable program code.The data-signal of this propagation can take various forms, and including but not limited to electromagnetism is believed
Number, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be computer-readable storage medium
Any computer-readable medium other than matter, the computer-readable medium can be sent, propagated or transmitted for being held by instruction
Row system, device or device use or program in connection.The program code for including on computer-readable medium
It can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. or above-mentioned any conjunction
Suitable combination.
The calculating of the operation for executing the application can be write with one or more programming languages or combinations thereof
Machine program code, described program design language include object oriented program language-such as Java, Smalltalk, C+
+, further include conventional procedural programming language-such as " C " language or similar programming language.Program code can
Fully to execute, partly execute on the user computer on the user computer, be executed as an independent software package,
Part executes on the remote computer or executes on a remote computer or server completely on the user computer for part.
In situations involving remote computers, remote computer can pass through the network of any kind --- including local area network (LAN)
Or wide area network (WAN)-is connected to subscriber computer, or, it may be connected to outer computer (such as utilize Internet service
Provider is connected by internet).
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the application, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in the embodiment of the present application can be realized by way of software, can also be by hard
The mode of part is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor packet
Include receiving unit, resolution unit, information extracting unit and generation unit.Wherein, the title of these units is under certain conditions simultaneously
The restriction to the unit itself is not constituted, for example, receiving unit is also described as " receiving the web page browsing request of user
Unit ".
As on the other hand, present invention also provides a kind of computer-readable mediums, are stored thereon with computer program, should
The method as described in above-mentioned any embodiment is realized when program is executed by processor.
As on the other hand, present invention also provides a kind of computer-readable medium, which be can be
Included in device described in above-described embodiment；It is also possible to individualism, and without in the supplying device.Above-mentioned calculating
Machine readable medium carries one or more program, when said one or multiple programs are executed by the device, so that should
Device: using user behavior data training Factorization machine model, the hidden vector of user and the multiple dimensions of recommended object are obtained；
According to the hidden vector of the user and the multiple dimensions of recommended object, the matching for extracting user and each dimension of recommended object is special
Sign；Sample being constituted after the processing of the matching characteristic of the user and each dimension of recommended object, training gradient promotes tree-model,
And tree-model is promoted by the gradient, Ordering and marking is provided.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (12)
1. the sort method in a kind of information flow, which is characterized in that the described method includes:
Using user behavior data training Factorization machine model, the hidden vector of user and the multiple dimensions of recommended object are obtained；
According to the hidden vector of the user and the multiple dimensions of recommended object, of user and each dimension of recommended object are extracted
With feature；
Sample, training gradient boosted tree mould will be constituted after the processing of the matching characteristic of the user and each dimension of recommended object
Type, and tree-model is promoted by the gradient, Ordering and marking is provided.
2. the method according to claim 1, wherein described by the user and each dimension of recommended object
Sample is constituted after matching characteristic processing, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking
Further include:
The user and the matching characteristic of each dimension of recommended object are normalized.
3. the method according to claim 1, wherein the matching of the user and each dimension of recommended object is special
Sign includes:
It is recommended in subject side feature and information flow in information flow and is recommended object and End-user relevance feature.
4. according to the method described in claim 3, it is characterized in that, being recommended object and End-user relevance spy in the information flow
Sign carries out matching degree using the hidden vector of the hidden vector sum user of information flow object and calculates acquisition.
5. according to the method described in claim 4, it is characterized in that, the hidden vector of the hidden vector sum user of the information flow object uses
Version alignment mechanism enables described its that can be aligned.
6. the collator in a kind of information flow, which is characterized in that described device includes:
Vector obtains module, and setting is used to be obtained user using user behavior data training Factorization machine model and be recommended
The hidden vector of the multiple dimensions of object；
Characteristic extracting module, setting for the hidden vector according to the user and the multiple dimensions of recommended object, extract user and
The matching characteristic of the recommended each dimension of object；
Model training module is arranged for constituting sample after the matching characteristic processing by the user and each dimension of recommended object
This, training gradient promotes tree-model, and promotes tree-model by the gradient and provide Ordering and marking.
7. device according to claim 6, which is characterized in that the model training module, which is also set up, to be used for:
The user and the matching characteristic of each dimension of recommended object are normalized.
8. device according to claim 6, which is characterized in that the matching of the user and each dimension of recommended object is special
Sign includes:
It is recommended in subject side feature and information flow in information flow and is recommended object and End-user relevance feature.
9. device according to claim 8, which is characterized in that be recommended object in the information flow and End-user relevance is special
Sign carries out matching degree using the hidden vector of the hidden vector sum user of information flow object and calculates acquisition.
10. device according to claim 9, which is characterized in that the hidden hidden vector of vector sum user of information flow object is adopted
Enable described its that can be aligned with version alignment mechanism.
11. a kind of equipment/terminal/server, comprising:
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
Now such as method as claimed in any one of claims 1 to 5.
12. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
Such as method as claimed in any one of claims 1 to 5 is realized when execution.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811004222.5A CN109033460A (en) | 2018-08-30 | 2018-08-30 | Sort method, device and equipment/terminal/server in a kind of information flow |
| PCT/IB2018/057166 WO2020044098A2 (en) | 2018-08-30 | 2018-09-18 | Method and apparatus for sorting in information stream, and device/terminal/server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811004222.5A CN109033460A (en) | 2018-08-30 | 2018-08-30 | Sort method, device and equipment/terminal/server in a kind of information flow |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109033460A true CN109033460A (en) | 2018-12-18 |
Family
ID=64625867
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811004222.5A Pending CN109033460A (en) | 2018-08-30 | 2018-08-30 | Sort method, device and equipment/terminal/server in a kind of information flow |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109033460A (en) |
| WO (1) | WO2020044098A2 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111695960A (en) * | 2019-03-12 | 2020-09-22 | 阿里巴巴集团控股有限公司 | Object recommendation system, method, electronic device and storage medium |
| CN111950733A (en) * | 2019-05-15 | 2020-11-17 | 阿里巴巴集团控股有限公司 | Information flow sorting method and device and computer storage medium |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801760A (en) * | 2021-03-30 | 2021-05-14 | 南京蓝鲸人网络科技有限公司 | Sequencing optimization method and system of content personalized recommendation system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250461A (en) * | 2016-07-28 | 2016-12-21 | 北京北信源软件股份有限公司 | A kind of algorithm utilizing gradient lifting decision tree to carry out data mining based on Spark framework |
| CN107515909A (en) * | 2017-08-11 | 2017-12-26 | 深圳市耐飞科技有限公司 | A kind of video recommendation method and system |
| CN107562758A (en) * | 2016-06-30 | 2018-01-09 | 北京金山安全软件有限公司 | Information pushing method and device and electronic equipment |
| CN107729488A (en) * | 2017-10-17 | 2018-02-23 | 北京搜狐新媒体信息技术有限公司 | A kind of information recommendation method and device |
| CN108427708A (en) * | 2018-01-25 | 2018-08-21 | 腾讯科技（深圳）有限公司 | Data processing method, device, storage medium and electronic device |
-
2018
- 2018-08-30 CN CN201811004222.5A patent/CN109033460A/en active Pending
- 2018-09-18 WO PCT/IB2018/057166 patent/WO2020044098A2/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107562758A (en) * | 2016-06-30 | 2018-01-09 | 北京金山安全软件有限公司 | Information pushing method and device and electronic equipment |
| CN106250461A (en) * | 2016-07-28 | 2016-12-21 | 北京北信源软件股份有限公司 | A kind of algorithm utilizing gradient lifting decision tree to carry out data mining based on Spark framework |
| CN107515909A (en) * | 2017-08-11 | 2017-12-26 | 深圳市耐飞科技有限公司 | A kind of video recommendation method and system |
| CN107729488A (en) * | 2017-10-17 | 2018-02-23 | 北京搜狐新媒体信息技术有限公司 | A kind of information recommendation method and device |
| CN108427708A (en) * | 2018-01-25 | 2018-08-21 | 腾讯科技（深圳）有限公司 | Data processing method, device, storage medium and electronic device |
Non-Patent Citations (5)
| Title |
|---|
| CHEN CHENG等: ""Gradient Boosting Factorization Machines"", 《PROCEEDINGS OF THE 8TH ACM CONFERENCE ON RECOMMENDER SYSTEM》 * |
| FENG ZHOU等: ""A Novel Ensemble Strategy Combining Gradient Boosted Decision Trees and Factorization Machine Based Neural Network for Clicks Prediction"", 《2018 INTERNATIONAL CONFERENCE ON BIG DATA AND ARTIFICIAL INTELLIGENCE》 * |
| 匡俊等: ""基于特征工程的视频点击率预测算法"", 《华东师范大学学报（自然科学版）》 * |
| 潘洋等: ""融合因子分解机和用户行为预测的音乐推荐"", 《计算机工程与应用》 * |
| 胡平伍: ""移动广告点击率预测方法的研究与实现"", 《中国优秀硕士学位论文全文数据库 (经济与管理科学辑)》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111695960A (en) * | 2019-03-12 | 2020-09-22 | 阿里巴巴集团控股有限公司 | Object recommendation system, method, electronic device and storage medium |
| CN111950733A (en) * | 2019-05-15 | 2020-11-17 | 阿里巴巴集团控股有限公司 | Information flow sorting method and device and computer storage medium |
| CN111950733B (en) * | 2019-05-15 | 2024-06-11 | 阿里巴巴集团控股有限公司 | Method and device for ordering information streams and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020044098A3 (en) | 2020-08-13 |
| WO2020044098A2 (en) | 2020-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bonzanini | Mastering social media mining with Python | |
| CN108171276B (en) | Method and apparatus for generating information | |
| CN109543111A (en) | Recommendation information screening technique, device, storage medium and server | |
| US10503829B2 (en) | Book analysis and recommendation | |
| CN108510373A (en) | Paintings recommend method, paintings recommendation apparatus, equipment and storage medium | |
| CN108228906B (en) | Method and apparatus for generating information | |
| CN109522486A (en) | Method and apparatus for match information | |
| CN108805594A (en) | Information-pushing method and device | |
| CN109033460A (en) | Sort method, device and equipment/terminal/server in a kind of information flow | |
| CN109165344A (en) | Method and apparatus for pushed information | |
| CN108475283A (en) | Text Text summarization for searching for multiple video flowings | |
| CN107211106B (en) | Animation generation server, animation generation method, animation generation system, and computer-readable recording medium | |
| CN108256537A (en) | A kind of user gender prediction method and system | |
| CN110020162A (en) | User identification method and device | |
| CN108763532A (en) | For pushed information, show the method and apparatus of information | |
| CN107943895A (en) | Information-pushing method and device | |
| CN108900612A (en) | Method and apparatus for pushed information | |
| CN109101309A (en) | For updating user interface method and device | |
| CN107925698A (en) | The relevant inquiry system of file type | |
| CN109284367A (en) | Method and apparatus for handling text | |
| CN109389660A (en) | Image generating method and device | |
| Wang et al. | Research on the influencing factors of the communication effect of Tik Tok short videos about intangible cultural heritage | |
| Fuad et al. | Analysis and classification of mobile apps using topic modeling: A case study on Google Play Arabic apps | |
| CN110019018A (en) | File recommended method and file recommendation apparatus | |
| US20230351473A1 (en) | Apparatus and method for providing user's interior style analysis model on basis of sns text |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181218 |
|
| RJ01 | Rejection of invention patent application after publication |