CN108038541A - Method, apparatus, equipment and the computer-readable medium that CTR is estimated - Google Patents
Method, apparatus, equipment and the computer-readable medium that CTR is estimated Download PDFInfo
- Publication number
- CN108038541A CN108038541A CN201711288753.7A CN201711288753A CN108038541A CN 108038541 A CN108038541 A CN 108038541A CN 201711288753 A CN201711288753 A CN 201711288753A CN 108038541 A CN108038541 A CN 108038541A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- real
- estimated
- deep neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- User Interface Of Digital Computer (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention proposes a kind of method that CTR is estimated, and comprises the following steps：Primitive character training step：Input is trained to the input layer of deep neural network by multiple hidden layers of the deep neural network after the primitive character of existing object is handled；Real-time characteristic training step：The real-time characteristic of existing object is inputted to last hidden layer of the deep neural network, joint training is carried out with primitive character；Estimate clicking rate output step：Clicking rate is estimated by what the output layer of the deep neural network exported existing object.By first individually being trained to primitive character, joint training then is carried out in conjunction with real-time characteristic, can effectively can be learnt with the information of real-time characteristic by model, so that the accuracy for improving the timeliness of model and estimating.
Description
Technical field
The present invention relates to big data technical field, more particularly to method and device, equipment and the computer that a kind of CTR is estimated
Computer-readable recording medium.
Background technology
Information flow recommend clicking rate (Click-Through-Rate, CTR) estimate refer to information flow recommend platform according to
User interest custom etc., estimates clicking rate of the user to institute's recommendation.And real-time deep model refers to estimate mould in design CTR
In order to give full play to the importance of real-time characteristic during type, a kind of technology that depth model and real-time characteristic are combined.Wherein
The real-time statistics feature includes the history CTR of Current Content, past one hour, the CTR of one day, and history shows number, clicks on
The features such as number.
And traditional design method is including following several：
(1) feature extraction and design higher order combination feature are manually carried out, is among these joining real-time statistics feature.
The step of Feature Engineering, can be roughly divided into：Characteristic Design, acquisition, processing.
However, since the extraction and combination of feature directly affect the final effect of model, therefore this rely on Feature Engineering
Model is for the required cost of labor of characteristic aspect work and to calculate cost higher, and be generally unattainable preferably
Effect.
(2) feature extracting method based on model, it would be desirable to which the primitive character being combined is input to a kind of solution such as GBDT
Combinations of features is carried out in the high model of the property released, obtains the feature of high-order, such a method a degree of can save feature work
Cost of labor, but the problem of also need to face a large amount of calculating.
(3) deep-neural-network (Deep Neural Networks, DNN), by primitive character or the spy of simple combination
Sign is input to DNN model learnings, solves the problems, such as that linear model can not carry out characteristic crossover combination, it also avoid feature work
Journey is brought a large amount of manually with calculating cost, effectively improves the generalization ability of model.
However, the input of real-time statistics feature and other feature vectors at the same time as DNN networks is subjected to model training.Through
The study of depth model is crossed, these real-time characteristics there can be a Multiple Combination with primitive character, and combined situation is unreadable and from setting
Meter person controls, its effect of these real-time characteristics can be weakened.
The content of the invention
The embodiment of the present invention provides method, apparatus, equipment and the computer-readable medium that a kind of CTR is estimated, at least to solve
Above technical problem certainly of the prior art.
In a first aspect, an embodiment of the present invention provides a kind of method that CTR is estimated, comprise the following steps：
Primitive character training step：Inputted after the primitive character of existing object is handled to the input of deep neural network
Layer, is trained by multiple hidden layers of the deep neural network；
Real-time characteristic training step：The real-time characteristic of existing object is inputted to last of the deep neural network
Hidden layer, joint training is carried out with primitive character；
Estimate clicking rate output step：Click is estimated by what the output layer of the deep neural network exported existing object
Rate.
With reference to first aspect, the present invention is in the first implementation of first aspect, the primitive character training step
In specifically include：
Sliding-model control is carried out to continuous primitive character；
Feature vector is formed after feature after sliding-model control is carried out embedded processing；
Feature vector is inputted to the input layer of the deep neural network model.
With reference to first aspect, the present invention is in second of implementation of first aspect, the real-time characteristic training step
Specially：Real-time characteristic is directly inputted into last hidden layer of the deep neural network.
With reference to first aspect, the present invention is in the third implementation of first aspect, the real-time characteristic training step
Specifically include：
Structure includes the shallow-layer neutral net of at least one layer of hidden layer；
Real-time characteristic is inputted to the input layer of the shallow-layer neutral net；
By the output layer of the shallow-layer neutral net by result export to the deep neural network last is hidden
Containing layer.
Second aspect, an embodiment of the present invention provides the device that a kind of CTR is estimated, including：
Primitive character training module, for the defeated of input after the primitive character of existing object is handled to deep neural network
Enter layer, be trained by multiple hidden layers of the deep neural network；
Real-time characteristic training module, for input after the real-time characteristic of existing object is handled to the deep neural network
Last hidden layer, with primitive character carry out joint training；
Clicking rate output module is estimated, for exporting estimating for existing object by the output layer of the deep neural network
Clicking rate.
With reference to second aspect, the present invention is in the first implementation of second aspect, the primitive character training module
Specifically include：
Sliding-model control submodule, for carrying out sliding-model control to continuous primitive character；
Vectorization handle submodule, for by after sliding-model control feature carry out embedded processing after formed feature to
Amount；
Feature input submodule, the deep neural network model is inputted by feature vector.
With reference to second aspect, the present invention is in second of implementation of second aspect, the real-time characteristic training module
Last hidden layer specifically for real-time characteristic to be directly inputted into the deep neural network.
With reference to second aspect, the present invention is in the third implementation of second aspect, the real-time characteristic training module
Specifically include：
Submodule is built, the shallow-layer neutral net of at least one layer of hidden layer is included for building；
Input submodule, for inputting real-time characteristic to the input layer of the shallow-layer neutral net；
Output sub-module, exports result to the depth nerve net for the output layer by the shallow-layer neutral net
Last hidden layer of network.
The function of described device can also be performed corresponding software and be realized by hardware realization by hardware.It is described
Hardware or software include the one or more and corresponding module of above-mentioned function.
In a possible design, the structure for the device that CTR is estimated includes processor and memory, the memory
For storing the program for supporting the devices estimated of CTR to perform the method that CTR is estimated in above-mentioned first aspect, the processor by with
It is set to for performing the program stored in the memory.The device that the CTR is estimated can also include communication interface, be used for
The device that CTR is estimated and other equipment or communication.
The third aspect, an embodiment of the present invention provides a kind of computer-readable medium, the device institute estimated for storing CTR
Computer software instructions, the program involved by method that its CTR for including being used to perform above-mentioned first aspect is estimated.
A technical solution in above-mentioned technical proposal has the following advantages that or beneficial effect：By first to primitive character list
Solely training, is then carried out joint training in conjunction with real-time characteristic, can effectively can be carried out with the information of real-time characteristic by model
Study, so that the accuracy for improving the timeliness of model and estimating.
Above-mentioned general introduction is merely to illustrate that the purpose of book, it is not intended to is limited in any way.Except foregoing description
Schematical aspect, outside embodiment and feature, it is further by reference to attached drawing and the following detailed description, the present invention
Aspect, embodiment and feature would is that what is be readily apparent that.
Brief description of the drawings
In the accompanying drawings, unless specified otherwise herein, otherwise represent the same or similar through the identical reference numeral of multiple attached drawings
Component or element.What these attached drawings were not necessarily to scale.It should be understood that these attached drawings depict only according to the present invention
Some disclosed embodiments, and should not serve to limit the scope of the present invention.
Fig. 1 is the model schematic that the CTR of embodiment one is estimated；
Fig. 2 is the step flow chart of the CTR predictor methods of embodiment one；
Fig. 3 is the specific steps flow chart of the step S110 of embodiment one；
Fig. 4 is the model schematic that the CTR of embodiment two is estimated；
Fig. 5 is the method flow diagram that the CTR of embodiment two is estimated；
Fig. 6 is the specific steps flow chart of the step S220 of embodiment two；
Fig. 7 is the connection block diagram for the device that the CTR of embodiment three is estimated；
Fig. 8 is the connection block diagram for the device that the CTR of example IV is estimated；
Fig. 9 is that the equipment that the CTR of embodiment five is estimated connects block diagram.
Embodiment
Hereinafter, some exemplary embodiments are simply just described.As one skilled in the art will recognize that
Like that, without departing from the spirit or scope of the present invention, described embodiment can be changed by various different modes.
Therefore, attached drawing and description are considered essentially illustrative rather than restrictive.
The embodiment of the present invention is aimed to solve the problem that in the prior art when there are during multiple terminal devices, meeting is at the same time to the voice of user
The technical problem that information is responded.The embodiment of the present invention is broadly divided into two parts, first passes through deep neural network to original spy
Sign is individually trained, and then in last hidden layer of neural depth model, is added real-time characteristic and is carried out joint training,
Last output estimation clicking rate CTR again.The expansion for carrying out technical solution by following embodiments separately below describes.
Embodiment one
The present embodiment one provides a kind of predictor method of CTR, it is based primarily upon on the basis of deep neural network model,
Real-time characteristic is inputted into the neural network model last hidden layer and carries out joint training.It is specific as shown in Figure 1, its
The model schematic estimated for the CTR of embodiment one.Model includes depth used by the method that the CTR of the present embodiment one is estimated
Neutral net, the deep neural network include input layer, multiple hidden layers and output layer.Wherein, the input layer receives former
Beginning feature, is then trained primitive character in hidden layer, and real-time characteristic then is input to the deep neural network
Last hidden layer in, with primitive character carry out joint training.
Specific steps flow chart as shown in Fig. 2, its for the CTR predictor methods of the embodiment of the present invention one step flow chart,
The CTR predictor methods of the present embodiment one comprise the following steps：
S110：Primitive character training step, inputs to deep neural network after the primitive character of existing object is handled
Input layer, is trained by multiple hidden layers of the deep neural network.
Wherein, the existing object can be：Document, Domestic News, video, music etc..If existing object is news
Information, then its primitive character include：The content keyword of information, the field, label etc..
Specifically, when the primitive character of existing object is inputted deep neural network, it is also necessary to which primitive character is carried out
Processing, specific processing step as shown in figure 3, including：
S111：Sliding-model control is carried out to continuous primitive character.
In this step, continuous feature is subjected to sliding-model control, obtains discrete characteristic value.
S112：Feature after sliding-model control is subjected to embedded processing and forms feature vector.
In this step, discrete characteristic value is subjected to embedded processing, obtains multiple feature vectors.
S113：Feature vector is inputted to the input layer of the deep neural network model.
In this step, the feature vector value of acquisition is inputted to the input layer of deep neural network, with to feature vector
It is trained.It is special using carrying out learning the combined crosswise between each feature entirely by the way of connecting in the deep neural network
Property.
S120：Real-time characteristic training step, the real-time characteristic of existing object is inputted to the deep neural network most
The latter hidden layer, joint training is carried out with primitive character.
The characteristic information of the real-time characteristic combination current slot of the existing object.For example if existing object is news
Information, then the real-time characteristic can include：The work of the touching quantity of user, current slot historical user in current slot
Jerk etc..
Specifically, in the present embodiment one, the real-time characteristic of existing object is directly inputted into deep neural network most
In the latter hidden layer, joint training is carried out.
S130：Clicking rate output step is estimated, the pre- of existing object is exported by the output layer of the deep neural network
Estimate clicking rate.
Finally, after the deep neural network completes the training to existing object, export last prediction result, i.e., it is pre-
Estimate clicking rate CTR.
Embodiment two
With embodiment one difference lies in：The present embodiment two first passes through shallow-layer god in the real-time characteristic training step
Real-time characteristic is trained through network, then is inputted into last hidden layer of deep neural network, specific scheme is such as
Under：
As shown in figure 4, it is the model schematic that the CTR of embodiment two is estimated.The method that the CTR of the present embodiment two is estimated
Used model includes deep neural network and shallow-layer neutral net.The deep neural network includes input layer, Duo Geyin
Containing layer and output layer.Wherein, the input layer receives primitive character, and then primitive character is trained in hidden layer.Institute
Stating shallow-layer neutral net includes one or two hidden layer, after receiving real-time characteristic and being trained, is output to the depth god
In last hidden layer through network, joint training is carried out with primitive character.
Specific steps flow chart is as shown in figure 5, it is the method flow diagram that the CTR of the present embodiment two is estimated.It is of the invention real
Apply example two and provide a kind of method that CTR is estimated, comprise the following steps：
S210：Primitive character training step, inputs to deep neural network after the primitive character of existing object is handled
Input layer, is trained by multiple hidden layers of the deep neural network.
S220：Real-time characteristic training step, after the real-time characteristic input shallow-layer neutral net of existing object is trained
Export to last hidden layer of the deep neural network, joint training is carried out with primitive character.
In this step S220, first the real-time characteristic of existing object is handled, specific processing step as described in Figure 6,
Including：
S221：Structure includes the shallow-layer neutral net of at least one layer of hidden layer.
S222：Real-time characteristic is inputted to the input layer of the shallow-layer neutral net.
S223：By the output layer of the shallow-layer neutral net by result export to the deep neural network last
A hidden layer.
It is the individually designed shallow-layer neutral net of the real-time characteristic in this step S220, the shallow-layer nerve net
Network can include one or two hidden layer.
S230：Clicking rate output step is estimated, the pre- of existing object is exported by the output layer of the deep neural network
Estimate clicking rate.
After the shallow-layer neutral net of addition, the input of real-time statistics feature can be made by the way of connecting entirely, most
Last hidden layer with deep neural network splices afterwards, so that effective combined crosswise information between real-time characteristic obtains
To be arrived by model learning.
At the same time it can also will be considered to need the primitive character for carrying out combined crosswise with real-time characteristic also may be used in characteristic Design
To be added among shallow-layer neutral net.It is such to design the validity that both ensure that real-time characteristic, also can be by network by fact
Combination and real-time characteristic and other combined crosswises that will not weaken between the feature of real-time characteristic effect between Shi Tezheng is special
Inquiry learning comes out.
Embodiment three
The present embodiment three corresponds to embodiment one, there is provided the device that a kind of CTR is estimated.Referring to Fig. 7, it is this implementation
The connection block diagram for the device that the CTR of example three is estimated.The embodiment of the present invention three provides the device that a kind of CTR is estimated, and specifically includes：
Primitive character training module 110, for input after the primitive character of existing object is handled to deep neural network
Input layer, be trained by multiple hidden layers of the deep neural network.
In the present embodiment three, the primitive character training module 110 specifically includes：
Sliding-model control submodule 111, for carrying out sliding-model control to continuous primitive character；
Vectorization handles submodule 112, for forming feature after the feature after sliding-model control is carried out embedded processing
Vector；
Feature input submodule 113, the deep neural network model is inputted by feature vector.
Real-time characteristic training module 120, for input after the real-time characteristic of existing object is handled to depth nerve
Last hidden layer of network, joint training is carried out with primitive character.In the present embodiment three, the real-time characteristic trains mould
Real-time characteristic is directly inputted into last hidden layer of the deep neural network by block.
Clicking rate output module 130 is estimated, for exporting existing object by the output layer of the deep neural network
Estimate clicking rate.
The present embodiment three is identical with the principle of embodiment one, and so it will not be repeated.
Example IV
The present embodiment four is corresponding with embodiment two, there is provided the device that a kind of CTR is estimated, it is specific as follows：
As shown in figure 8, the connection block diagrams of device estimated of CTR for the present embodiment four.The embodiment of the present invention four provides
The device that a kind of CTR is estimated, including：
Primitive character training module 210, for input after the primitive character of existing object is handled to deep neural network
Input layer, be trained by multiple hidden layers of the deep neural network.
Real-time characteristic training module 220, for the real-time characteristic input shallow-layer neutral net of existing object to be trained
Export afterwards to last hidden layer of the deep neural network, joint training is carried out with primitive character.The real-time characteristic
Training module specifically includes：
Submodule 221 is built, the shallow-layer neutral net of at least one layer of hidden layer is included for building.
Input submodule 222, for inputting real-time characteristic to the input layer of the shallow-layer neutral net.
Output sub-module 223, exports result to depth god for the output layer by the shallow-layer neutral net
Last hidden layer through network.
Clicking rate output module 230 is estimated, for exporting existing object by the output layer of the deep neural network
Estimate clicking rate.
The application mode of the present embodiment four is identical with embodiment two with principle, and so it will not be repeated.
Embodiment five
The embodiment of the present invention five provides the equipment that a kind of CTR is estimated, as shown in figure 9, the equipment includes：310 He of memory
Processor 320,310 memory of memory contain the computer program that can be run on the processor 320.The processor 320 performs
The method that the CTR in above-described embodiment is estimated is realized during the computer program.The number of the memory 310 and processor 320
Amount can be one or more.
The equipment further includes：
Communication interface 330, for communicating with external device, carries out data interaction.
Memory 310 may include high-speed RAM memory, it is also possible to further include nonvolatile memory (non-
Volatile memory), a for example, at least magnetic disk storage.
If memory 310, processor 320 and the independent realization of communication interface 330, memory 310,320 and of processor
Communication interface 330 can be connected with each other by bus and complete mutual communication.The bus can be Industry Standard Architecture
Structure (ISA, Industry Standard Architecture) bus, external equipment interconnection (PCI, Peripheral
Component) bus or extended industry-standard architecture (EISA, Extended Industry Standard
Component) bus etc..The bus can be divided into address bus, data/address bus, controlling bus etc..For ease of representing, Fig. 9
In only represented with a thick line, it is not intended that an only bus or a type of bus.
Optionally, in specific implementation, if memory 310, processor 320 and communication interface 330 are integrated in one piece of core
On piece, then memory 310, processor 320 and communication interface 330 can complete mutual communication by internal interface.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or the spy for combining the embodiment or example description
Point is contained at least one embodiment of the present invention or example.Moreover, particular features, structures, materials, or characteristics described
It may be combined in any suitable manner in any one or more of the embodiments or examples.In addition, without conflicting with each other, this
The technical staff in field can be by the different embodiments or example described in this specification and different embodiments or exemplary spy
Sign is combined and combines.
In addition, term " first ", " second " are only used for description purpose, and it is not intended that instruction or hint relative importance
Or the implicit quantity for indicating indicated technical characteristic.Thus, " first " is defined, the feature of " second " can be expressed or hidden
Include at least one this feature containing ground.In the description of the present invention, " multiple " are meant that two or more, unless otherwise
It is clearly specific to limit.
Any process or method described otherwise above description in flow chart or herein is construed as, and represents to include
Module, fragment or the portion of the code of the executable instruction of one or more the step of being used for realization specific logical function or process
Point, and the scope of the preferred embodiment of the present invention includes other realization, wherein can not press shown or discuss suitable
Sequence, including according to involved function by it is basic at the same time in the way of or in the opposite order, carry out perform function, this should be of the invention
Embodiment person of ordinary skill in the field understood.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (such as computer based system including the system of processor or other can be held from instruction
The system of row system, device or equipment instruction fetch and execute instruction) use, or combine these instruction execution systems, device or set
It is standby and use.For the purpose of this specification, " computer-readable medium " can any can be included, store, communicate, propagate or pass
Defeated program is for instruction execution system, device or equipment or the dress used with reference to these instruction execution systems, device or equipment
Put.
Computer-readable medium described in the embodiment of the present invention can be that computer-readable signal media or computer can
Read storage medium either the two any combination.The more specifically example of computer-readable recording medium is at least (non-poor
Property list to the greatest extent) including following：Electrical connection section (electronic device) with one or more wiring, portable computer diskette box (magnetic
Device), random access memory (RAM), read-only storage (ROM), erasable edit read-only storage (EPROM or flash
Memory), fiber device, and portable read-only storage (CDROM).In addition, computer-readable recording medium even can be with
It is the paper or other suitable media that can print described program on it, because can be for example by being carried out to paper or other media
Optical scanner, is then handled described electronically to obtain into edlin, interpretation or if necessary with other suitable methods
Program, is then stored in computer storage.
In embodiments of the present invention, computer-readable signal media can be included in a base band or as a carrier wave part
The data-signal of propagation, wherein carrying computer-readable program code.The data-signal of this propagation can use a variety of
Form, includes but not limited to electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media is also
Can be any computer-readable medium beyond computer-readable recording medium, which can send, pass
Either transmission is broadcast for instruction execution system, input method or device use or program in connection.Computer can
The program code for reading to include on medium can be transmitted with any appropriate medium, be included but not limited to：Wirelessly, electric wire, optical cable, penetrate
Frequently (Radio Frequency, RF) etc., or above-mentioned any appropriate combination.
It should be appreciated that each several part of the present invention can be realized with hardware, software, firmware or combinations thereof.Above-mentioned
In embodiment, software that multiple steps or method can be performed in memory and by suitable instruction execution system with storage
Or firmware is realized.If, and in another embodiment, can be with well known in the art for example, realized with hardware
Any one of row technology or their combination are realized：With the logic gates for realizing logic function to data-signal
Discrete logic, have suitable combinational logic gate circuit application-specific integrated circuit, programmable gate array (PGA), scene
Programmable gate array (FPGA) etc..
Those skilled in the art are appreciated that to realize all or part of step that above-described embodiment method carries
Suddenly it is that relevant hardware can be instructed to complete by program, the program can be stored in a kind of computer-readable storage medium
In matter, the program upon execution, including one or a combination set of the step of embodiment of the method.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing module, can also
That unit is individually physically present, can also two or more units be integrated in a module.Above-mentioned integrated mould
Block can both be realized in the form of hardware, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized in the form of software function module and as independent production marketing or in use, can also be stored in a computer
In readable storage medium storing program for executing.The storage medium can be read-only storage, disk or CD etc..
In conclusion the embodiment of the present invention by first individually being trained to primitive character, is then carried out in conjunction with real-time characteristic
Joint training, can effectively can be learnt with the information of real-time characteristic by model, thus improve model timeliness and
The accuracy estimated.
The above description is merely a specific embodiment, but protection scope of the present invention is not limited thereto, any
Those familiar with the art the invention discloses technical scope in, its various change or replacement can be readily occurred in,
These should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the guarantor of the claim
Protect subject to scope.
Claims (10)
1. a kind of method that CTR is estimated, it is characterised in that including
Primitive character training step：Inputted after the primitive character of existing object is handled to the input layer of deep neural network, by
Multiple hidden layers of the deep neural network are trained；
Real-time characteristic training step：The real-time characteristic of existing object is inputted to last of the deep neural network and is implied
Layer, joint training is carried out with primitive character；
Estimate clicking rate output step：Clicking rate is estimated by what the output layer of the deep neural network exported existing object.
2. the method that CTR according to claim 1 is estimated, it is characterised in that specific in the primitive character training step
Including：
Sliding-model control is carried out to continuous primitive character；
Feature after sliding-model control is subjected to embedded processing and forms feature vector；
Feature vector is inputted to the input layer of the deep neural network model.
3. the method that CTR according to claim 1 is estimated, it is characterised in that the real-time characteristic training step is specially：
Real-time characteristic is directly inputted into last hidden layer of the deep neural network.
4. the method that CTR according to claim 1 is estimated, it is characterised in that the real-time characteristic training step specifically wraps
Include：
Structure includes the shallow-layer neutral net of at least one layer of hidden layer；
Real-time characteristic is inputted to the input layer of the shallow-layer neutral net；
Result is exported to last hidden layer of the deep neural network by the output layer of the shallow-layer neutral net.
A kind of 5. device that CTR is estimated, it is characterised in that including：
Primitive character training module, the input for input after the primitive character of existing object is handled to deep neural network
Layer, is trained by multiple hidden layers of the deep neural network；
Real-time characteristic training module, for being inputted after the real-time characteristic of existing object is handled to the deep neural network most
The latter hidden layer, joint training is carried out with primitive character；
Clicking rate output module is estimated, click is estimated for the output layer output existing object by the deep neural network
Rate.
6. the method that CTR is estimated according to claim 5, it is characterised in that the primitive character training module specifically includes：
Sliding-model control submodule, for carrying out sliding-model control to continuous primitive character；
Vectorization handles submodule, for forming feature vector after the feature after sliding-model control is carried out embedded processing；
Feature input submodule, the deep neural network model is inputted by feature vector.
7. the method that CTR is estimated according to claim 5, it is characterised in that the real-time characteristic training module is specifically used for
Real-time characteristic is directly inputted into last hidden layer of the deep neural network.
8. the method that CTR is estimated according to claim 5, it is characterised in that the real-time characteristic training module specifically includes：
Submodule is built, the shallow-layer neutral net of at least one layer of hidden layer is included for building；
Input submodule, for inputting real-time characteristic to the input layer of the shallow-layer neutral net；
Output sub-module, exports result to the deep neural network for the output layer by the shallow-layer neutral net
Last hidden layer.
9. the equipment that a kind of CTR is estimated, it is characterised in that the equipment includes：
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are performed by one or more of processors so that one or more of processors
Realize the method that the CTR as described in any in claim 1-4 is estimated.
10. a kind of computer-readable medium, it is stored with computer program, it is characterised in that when the program is executed by processor
Realize the method that the CTR as described in any in claim 1-4 is estimated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711288753.7A CN108038541B (en) | 2017-12-07 | 2017-12-07 | CTR (China train redundancy) estimation method, device, equipment and computer readable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711288753.7A CN108038541B (en) | 2017-12-07 | 2017-12-07 | CTR (China train redundancy) estimation method, device, equipment and computer readable medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108038541A true CN108038541A (en) | 2018-05-15 |
| CN108038541B CN108038541B (en) | 2021-02-19 |
Family
ID=62095987
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711288753.7A Active CN108038541B (en) | 2017-12-07 | 2017-12-07 | CTR (China train redundancy) estimation method, device, equipment and computer readable medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108038541B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615060A (en) * | 2018-11-27 | 2019-04-12 | 深圳前海微众银行股份有限公司 | CTR predictor method, device and computer readable storage medium |
| CN110039537A (en) * | 2019-03-15 | 2019-07-23 | 北京精密机电控制设备研究所 | A kind of automatic measure on line multi joint motion planing method neural network based |
| CN110083688A (en) * | 2019-05-10 | 2019-08-02 | 北京百度网讯科技有限公司 | Search result recalls method, apparatus, server and storage medium |
| CN110399979A (en) * | 2019-06-17 | 2019-11-01 | 深圳大学 | Click rate pre-estimation system and method based on field programmable gate array |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105160548A (en) * | 2015-08-20 | 2015-12-16 | 北京奇虎科技有限公司 | Method and apparatus for predicting advertisement click-through rate |
| CN107423430A (en) * | 2017-08-03 | 2017-12-01 | 北京京东尚科信息技术有限公司 | Data processing method, device and computer-readable recording medium |
-
2017
- 2017-12-07 CN CN201711288753.7A patent/CN108038541B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105160548A (en) * | 2015-08-20 | 2015-12-16 | 北京奇虎科技有限公司 | Method and apparatus for predicting advertisement click-through rate |
| CN107423430A (en) * | 2017-08-03 | 2017-12-01 | 北京京东尚科信息技术有限公司 | Data processing method, device and computer-readable recording medium |
Non-Patent Citations (2)
| Title |
|---|
| KINTOCAI: "闲聊DNN CTR预估模型", 《CSDN HTTPS://BLOG.CSDN.NET/BIGHEADYUSHAN/ARTICLE/DETAILS/77588397》 * |
| WEINAN ZHANG等: "Deep Learning over Multi-field Categorical Data-A Case Study on User Response Prediction", 《ECIR2016》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615060A (en) * | 2018-11-27 | 2019-04-12 | 深圳前海微众银行股份有限公司 | CTR predictor method, device and computer readable storage medium |
| CN109615060B (en) * | 2018-11-27 | 2023-06-30 | 深圳前海微众银行股份有限公司 | CTR estimation method, CTR estimation device and computer-readable storage medium |
| CN110039537A (en) * | 2019-03-15 | 2019-07-23 | 北京精密机电控制设备研究所 | A kind of automatic measure on line multi joint motion planing method neural network based |
| CN110083688A (en) * | 2019-05-10 | 2019-08-02 | 北京百度网讯科技有限公司 | Search result recalls method, apparatus, server and storage medium |
| CN110083688B (en) * | 2019-05-10 | 2022-03-25 | 北京百度网讯科技有限公司 | Search result recall method, device, server and storage medium |
| CN110399979A (en) * | 2019-06-17 | 2019-11-01 | 深圳大学 | Click rate pre-estimation system and method based on field programmable gate array |
| CN110399979B (en) * | 2019-06-17 | 2022-05-13 | 深圳大学 | Click rate pre-estimation system and method based on field programmable gate array |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108038541B (en) | 2021-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108038541A (en) | Method, apparatus, equipment and the computer-readable medium that CTR is estimated | |
| CN107885886A (en) | To the method, apparatus and server of information recommendation sort result | |
| CN109522531A (en) | Official documents and correspondence generation method and device, storage medium and electronic device | |
| CN103425678B (en) | The system of selection of the recommendation information of a kind of business object and device | |
| CN108090218A (en) | Conversational system generation method and device based on deeply study | |
| CN110990600B (en) | Multimedia file recommendation method, multimedia file recommendation device, multimedia file parameter adjustment device, multimedia file recommendation medium and electronic equipment | |
| CN108197251A (en) | A kind of big data operation and maintenance analysis method, device and server | |
| CN111582938A (en) | Advertisement putting method and device and electronic equipment | |
| CN110443494A (en) | Data distributing method and device | |
| CN108052597A (en) | Method, apparatus, equipment and the computer-readable medium of news push | |
| CN107864405A (en) | A kind of Forecasting Methodology, device and the computer-readable medium of viewing behavior type | |
| CN114519435A (en) | Model parameter updating method, model parameter updating device and electronic equipment | |
| CN111986318A (en) | Method and device for selecting steel to be processed | |
| CN105955743A (en) | Resource data transfer request generating method, device and system | |
| CN107871088A (en) | A kind of information processing method, device, terminal and computer-readable recording medium | |
| CN117172887B (en) | Commodity recommendation model training method and commodity recommendation method | |
| CN104133679B (en) | The method and apparatus for showing function of application | |
| CN110516050A (en) | A kind of construction method of the multipath Training scene of knowledge based map | |
| CN113255079A (en) | Artificial intelligence-based package design method and device | |
| CN108509582A (en) | A kind of answering method of information, terminal device and computer readable storage medium | |
| CN107679889A (en) | The recognition methods of potential customers a kind of and terminal device | |
| CN116028823A (en) | Loss calculation method and device based on multi-mode semantic interaction | |
| CN110827078A (en) | Information recommendation method, device, equipment and storage medium | |
| CN113808192B (en) | House pattern generation method, device, equipment and storage medium | |
| CN109753708A (en) | A kind of payment amount prediction technique, device and readable storage medium storing program for executing |
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 |